Sponsored by DataCamp. Glasp is paid for this email and earns a commission when you sign up through the links below.

If learning Python, SQL or AI has been on your list, this is a free week to start, so we asked DataCamp how to use it well.

Kei and Kazuki

Learn AI engineering, data analytics, SQL and Python for $0

Learn AI engineering for $0.

Learn data analytics for $0.

Learn Python for $0.

Learn SQL for $0.

This week only.

All 700+ DataCamp courses, career tracks, and certifications are free from November 2nd-8th. No credit card required.

Start Now

We asked DataCamp five questions about the week. Their answers, in their words:

1. What is Free Access Week, and what can someone use for free from November 2 to 8? Is anything left out, and do they need a credit card?

For seven days, DataCamp’s entire course library is available for $0. From November 2 to 8, you get access to 700+ courses, every career and skill track, hands-on projects, practice exercises, assessments and even industry-recognized certifications. Python, SQL, Power BI, data analytics, AI engineering: all of it. No credit card. No trial that quietly turns into a subscription. Just create a free account and start learning.

2. Our readers are founders, builders, students, and knowledge workers. Which course, track or project would you suggest for each of them, depending on where they are starting from?

Pick the one that matches where you are today:

• Founders: start with the AI Fundamentals track or AI Business Fundamentals track to see where AI can boost your business’s efficiency.

• Builders: go straight to the Associate AI Engineer for Developers track. You’ll be working with LLMs, embeddings and APIs, not watching slides about them.

• Students and knowledge workers: Python Fundamentals, SQL Fundamentals, or Power BI Fundamentals, all skills that can make your resume stand out on the job market.

All FREE courses HERE

3. If a reader has seven days and about 30 minutes a day, what should they do first, and what can they realistically finish by the end of the week?

Thirty minutes a day is about 3.5 hours across the week. That’s more than enough, if you don’t waste it. Most people sample ten courses and finish none. Don’t. On day one, pick one goal and one track. Then show up at the same time every day. Realistically, you can finish a full introductory course like Introduction to SQL, Python, or Introduction to AI for Work (about 2 hours each) and use the rest of the week on a guided project with a real dataset. One week won’t make you an expert. But it will take you from “I’ve been meaning to start” to “I’ve finished something.”

4. What helps people keep what they learned after the week ends, rather than starting and dropping it?

Three things make the difference. First, write code from the very first lesson. Every DataCamp course is interactive, so you build, break things and fix them instead of watching passive videos. That’s what makes it stick. Second, finish one thing. A completed course or project gives you momentum. Ten half-started ones give you guilt. Third, make it visible. Build a guided project with a real dataset and add it to your free portfolio, or share what you learned with your network. The habit matters more than the hours: if 30 minutes a day worked during Free Access Week, keep the same slot the week after.

5. What happens when the week ends: what stays available on a free account, and is there an offer for those who want to continue?

Your free account stays yours. On the free plan you keep access to the first chapter of every course, skill assessments, all cheat sheets and tutorials, live code-alongs, and competitions. If the week clicked and you want to keep going, premium unlocks the whole library, so you can carry on with the track you started and turn one good week into a skill that can change your career.

Start Now

Hi friends,

Muse is Meta’s personal agent. You can use it in the iPhone and Android apps, or in the browser at muse.ai. Now I can ask Muse about what I saved in Glasp: my web highlights, Kindle highlights, bookmarks, and the summaries I made with YouTube Summary. It answers from what I kept, with links and timestamps.

Setting it up takes one message. Here is the first step.

If you want to reread or highlight this newsletter, save it to Glasp.

Save to Glasp

Connect Glasp to Muse from your phone

1. Send one message. Open a chat in the Muse app and send:

“Please create a custom connector for Glasp using this MCP server: https://glasp.co/api/mcp“

Muse asks whether it may share information with glasp.co. Choose “Allow once”, or “Always allow this site”.

2. Check what it can do, then tap Connect. Muse lists what the Glasp connector offers. It is read-only apart from one permission you approve when connecting: saving a page to your Glasp bookmarks by URL. Your bookmarks are public on your profile. It can never edit or delete anything, including your highlights.

3. Sign in to Glasp and approve. No token to copy, no app to register. Then ask “What have I highlighted recently on Glasp?” to check it works.

If connecting from the app does not work, do it once at muse.ai on a computer with the same Meta account. The connection carries over to the app on your phone.

Meta lists Muse for the US, Canada, and Singapore, so check whether it is available where you are.

What to ask next

Once it is connected, the full guide shows what I use it for:

• Asking what I have highlighted on a topic, grouped by idea, each with its source

• Asking about a YouTube video I summarized, with timestamps

• A digest of yesterday’s highlights every morning at 8

• A reading goal that checks my Kindle highlights once a month

How to Use Glasp with Meta Muse

If you try it, reply and tell me what you asked. I read every reply.

Kei & Kazuki

Would you like to take Glasp on the go?

With the Glasp mobile app, you can highlight and organize your favorite content anytime, anywhere. Stay productive on the move and never miss an insightful quote.

Download for iOS

Download for Android

Partner With Glasp

We currently offer newsletter sponsorships. If you have a product, event, or service you’d like to share with our community of learning enthusiasts, sponsor an edition of our newsletter to reach engaged readers.

Sponsor an Edition

If you want to reread or highlight this newsletter, save it to Glasp.

Save to Glasp

There’s nothing better than finding a book so thrilling that it keeps you up all night.

From the 1,300+ books I’ve read, here are my 15 favorite page-turners:

15. The Compound by S.A. Bodeen

I went through a young adult (YA) phase years ago, and this was one of the best books I stumbled upon. It follows a boy named Eli who’s lived underground for the past six years. His family is sheltered, avoiding some unknown horrible thing, but you’re left unaware of what it is for most of the book.

I won’t say much more about The Compound because it’s the type of book that’s best approached blind, but I think you’ll enjoy this quick read.

14. The Will of the Many by James Islington

I’m not normally a big fantasy guy. While I’ve read classics like LOTR, The Chronicles of Narnia, etc., I rarely read many modern fantasy series. But dang, this book may make me reassess! Islington’s dystopian social hierarchy and magical system really pulled me in.

The protagonist Vis Telimus is smart and funny, which always helps. And the plot is so propulsive that I kept turning pages to see what would happen.

13. The Running Man by Stephen King

In the late 70s and early 80s, Stephen King wrote several novels under the pen name Richard Bachman. This Bachman book features everyman Ben Richards as a contestant trying to outrun killers in a deadly game show. Every day he survives produces more money for his family.

One unique aspect of the book is that citizens have a financial incentive to report Richards’s whereabouts if they ever see him, which makes his task infinitely harder. The big question is: How long can Richards stay alive?

12. Bird Box by Josh Malerman

Bird Box is one of the best thrillers I’ve read in years. Malerman does an amazing job building suspense around an unseen entity that causes millions of people across the world to kill themselves. Scattered survivors live in fear, cowering behind blindfolds and shuttered windows.

Books like this prove that sometimes the scariest enemy is the one you don’t see.

11. The Talented Mr. Ripley by Patricia Highsmith

If you love suspense and you’ve never read Patricia Highsmith, stop reading my article and go pick up this novel. Tom Ripley is many things: con man…genius…sociopath. After he reconnects with his old friend Dickie in Europe, Tom decides he’s too jealous of Dickie’s life to ever go back to his own boring old life in America, so he comes up with a unique way to live the life he’s always wanted.

Tom Ripley has been portrayed by two titans on the screen. Matt Damon played the role in a 1999 movie version alongside Jude Law, Gwyneth Paltrow, Cate Blanchett, and Philip Seymour Hoffman. Andrew Scott also crushed the role in a 2024 Netflix miniseries.

10. Ender’s Game by Orson Scott Card

Because I didn’t enjoy reading in junior high and high school, I didn’t read Ender’s Game until I was an adult. It still packed a wallop. Precocious youngster Ender Wiggin is enrolled in a military school and pitted against fellow students and otherworldly foes in this fast-paced sci-fi novel.

But what I loved about this book wasn’t just its action scenes and page-turning plot. It also explores deeper themes about humanity, war, and friendship.

9. Misery by Stephen King

Author Paul Sheldon comes face-to-face with his biggest fan in this creepy novel. When Sheldon’s car crashes in the middle of nowhere, a woman named Annie Wilkes nurses him back to health. But then Wilkes decides to hold Sheldon hostage until he writes a new story about her favorite character from his books: Misery Chastain.

If you’ve never read Stephen King, this book is a great gateway drug because it’s one of his smallest. I’d definitely start with something like this rather than one of King’s 1,000-page behemoths like It or The Stand, which are also fantastic.

8. The Maze Runner by James Dashner

Sixteen-year-old Thomas wakes up in a maze, surrounded by a bunch of boys whose memories are gone. The rest of the trilogy follows Thomas as he tries to discover where he came from and what happened to his society.

Dashner wrote multiple prequels to this story, and I enjoy some of them as much as the original trilogy, which is quite rare. If you like this first book, I’d definitely recommend diving into the rest of the Maze Runner universe.

Subscribe to Leader & Learner

7. The Chain by Adrian McKinty

As soon as I read the synopsis of this book, I immediately decided to buy it. A woman named Rachel gets a phone call saying her daughter has been kidnapped. In addition to paying a $25,000 ransom, Rachel must select and kidnap another kid to keep “The Chain” going.

The unnerving request is the ultimate test of a mother’s love, and Rachel learns that her love runs deep.

6. Ready Player One by Ernest Cline

The year is 2045, and the world has been ravaged by poverty, climate change, and war. The majority of humanity copes with the world’s destitution by immersing themselves in the virtual reality of the OASIS, the brainchild of computer scientist James Halliday. Halliday has decided to bequeath his fortune to whoever can find the Easter egg hidden inside the OASIS, and teenager Wade Watts thinks he’s up to the challenge.

If you haven’t read this book yet, you definitely should. It’s packed with 1980s pop culture references and storylines that appeal to both gamers and fans of fast-paced fiction.

5. The Plot by Jean Hanff Korelitz

Washed-up writer Jake Bonner is working as a professor when he meets brilliant and arrogant student Evan Parker. Parker tells Bonner the plot of a story he’s writing, then dies before he can publish it. Bonner decides to steal Parker’s idea, but later realizes there’s a dark underbelly to the story.

This is one of the best applications of the “book within a book” concept I’ve ever read. You’re immersed in the main story while also reading sections of Parker’s book, and Korelitz kept me engaged in both storylines.

4. Pines by Blake Crouch

It’s hard to express exactly how good (and underrated) this book is. Two federal agents go missing in Wayward Pines, Idaho, and the Secret Service agent who’s sent to find them realizes a lot is going on behind the scenes in this spooky town.

I flew through this book and immediately read the rest of the trilogy. I’ve recommended it to other friends who’ve also enjoyed it, which is always a good sign.

3. Gone Girl by Gillian Flynn

Every one of Flynn’s books is dark and twisted, and this one is my favorite. A woman named Amy Dunne disappears on her fifth anniversary to dreamboat husband Nick. While investigators try to figure out what happened to Amy, passages from her diary indicate Nick may have had something to do with her disappearance. Then things get really crazy and really twisty.

Not only is this book an incredible psychological thriller, it’s also told in a unique storytelling technique that inspired emulation from multiple other authors.

2. Red Rising by Pierce Brown

The entire Red Rising series is amazing, and the first book is also one of the most unputdownable books of all time. If you like fast-paced books, science fiction, or thoughtfully designed dystopian worlds, you’ll eat this up.

I’m usually able to predict where a book is headed, but this one fooled me multiple times. The story is so good and the characters are so well-written that I’m planning to re-read the series again soon.

1. The Silent Patient by Alex Michaelides

Michaelides’s rookie novel is a thriller about a woman who murders her husband, then refuses to speak for six years. It’s the type of book that will keep you up all night, and it’s still the book I recommend to any friend who asks what they should read next.

The book kept me guessing until the end. And apparently, it did the same for the rest of America because The Silent Patient spent months on the bestseller list.

Note: Each book above includes an affiliate link from Bookshop.org. (If you purchase one of the books, you’ll support me as well as local bookstores. Win-win!)

📣 Community Updates by Glasp

• 🟥 Glasp Talk with Bobby Powers:

  Glasp Talk features intimate interviews with luminaries, revealing their emotions, experiences, and stories. The guest, Bobby Powers, is a dedicated learner passionate about leadership and personal development. With over a decade of experience managing teams, leading onboarding, and Learning and Development programs at various startups and SMBs, Bobby is also an avid reader, having read over a thousand books in the last 15 years.

  ✨ Want a concise summary before diving in? Install the YouTube Summary with ChatGPT extension by Glasp for instant video summaries!

• 🟦 Obsidian Plugin passes 3,000 downloads:
  Thank you to everyone who brought their Glasp highlights into Obsidian. The plugin imports your highlights, notes, summaries, and bookmarks, from the web and from Kindle, as Markdown notes you fully control, and keeps them in sync automatically. It is now a verified plugin with an “Excellent” health rating in the Obsidian community directory. Get it from Community Plugins.

  

• 🟩 Kindle imports pass 90,000 books:

  Readers have now brought highlights from more than 90,000 different books into Glasp. Your Kindle highlights sit next to everything you highlight on the web, so you can search and revisit them in one place. Import yours with Kindle Highlight Export.

• 🟨 YouTube transcripts, even when they are missing at first:
  When a video's transcript isn't available the first time you open it, Glasp now goes back and prepares it for you. Pro members are first in line, and it usually appears within about half an hour, so check back a little later. Some videos, such as private, members-only, or videos without speech, still won't have one.

Would you like to take Glasp on the go?

With the Glasp mobile app, you can highlight and organize your favorite content anytime, anywhere. Stay productive on the move and never miss an insightful quote.

Download for iOS

Download for Android

Partner With Glasp

We currently offer newsletter sponsorships. If you have a product, event, or service you’d like to share with our community of learning enthusiasts, sponsor an edition of our newsletter to reach engaged readers.

Sponsor an Edition

Share Glasp Newsletter

Share Glasp Newsletter

Hi friends,

At Glasp, we are always doing research, and we write papers too. The further that work goes, the more it leads to the same place: models are built from data, and someone has to gather it, whether for a foundation model or for fine-tuning. Sooner or later you start asking what data should go in. So we got curious and looked into this market. We are still exploring whether there is something useful we can do here. This issue shares what we found along the way, so you can see this market too.

A note first: Glasp does not sell your highlights to AI companies. This issue is about how other companies get their data.

Voice, video, robot and music models now need data the open web does not supply well: two people talking on separate channels, a face reacting in real time, a task filmed the same way many times, music with clear rights. We reviewed 79 AI companies and mapped 17 that publicly describe collecting, commissioning or licensing this data, or plan to, plus two that describe using open-web data. None of the companies named here sponsored this issue or reviewed it before publication.

If you want to reread or highlight this newsletter, save it to Glasp.

Save to Glasp

Three routes to human data

1. Collect it themselves. Rime built a recording studio in San Francisco. Miso Labs records voice actors in its own studio and in their home studios, and says its data set is over 100 million hours. Memories.ai’s “data collectors” wear a camera device it built. Odyssey has people hike with camera rigs on their backs.

2. Through data partners. Generalist AI trains its robot model on more than 270,000 hours of real-world data, collected partly “together with multiple data foundry partners” it names only as A, B and C.

3. License from owners. Moonvalley, Lightricks and Reka license video from stock libraries and rights holders. Suno, Udio and Stability AI have licensing deals with record labels.

And a fourth, cheaper route: the open web. Zyphra says the 200,000 hours of speech behind its voice models came from the web, not from data brokers.

What it sells for (three prices, not comparable)

These examples are not directly comparable: one open read-speech dataset is free to use, while one vendor lists licensed, documented two-person conversation datasets at $60 to $95 per audio hour.

WhatPriceNoteA vendor’s conversation datasets$60 to $95 per audio hourTwo-person recordings with transcripts, rights and consent documents. What a buyer pays, not what a speaker earns (SpeechData.ai)Mozilla Common VoiceFreeAbout 42,593 hours of read speech in 295 languages (Mozilla)What one app paid people$9 to $18 an hour$0.15 a minute, doubled if the other person also used it, $30 a day cap. The app went offline about a week after launch, after a flaw exposed recordings (Malwarebytes)

Do you need to be an expert?

For some high-volume collection work, no. Public listings for first-person video ask for a recent phone and time, and NPR found a worker in India recording housework for about $2.60 an hour (NPR). Lower labor costs are one reason so much robot video is filmed in India and Latin America; NPR reported that most domestic-video clients of one Indian collector were in the US. Reporting also identifies a smaller skilled layer, with trades such as plumbers and electricians filmed on the job, and a tracker of about 740 robot-training job postings puts the median posted rate at about $25 an hour.

Voice asks for a different skill: performing. AppTek recruited speakers to role-play call-center conversations, and Miso Labs records voice actors working from scripts. Credentials pay most in labeling and evaluation. In our October 15 report, the median advertised top rate on AI work platforms was $150 an hour for coding and $122.50 for medicine, against $20 for voice and audio work.

Is this market growing?

Yes, as far as anyone can measure it. Only a few listed companies report revenue from AI data work separately. In our October 15 report we added up the four we found: $544 million in 2024 and $718 million in 2025, up 32% (totals from the exact filing figures; the chart’s company bars are rounded). That is a floor, not the market.

The private companies are larger. Mercor reports a gross run rate above $2 billion (June 2026) and micro1 above $1 billion (October 2026), by their own or press figures that mix gross and net revenue. For robots alone, robotics companies are “spending more than $100 million each year to buy real-world data”, one data company estimates (MIT Technology Review, April 2026). And the mix is shifting: at one listed vendor, computer-vision revenue tripled in 2025 while speech fell 9%.

The volumes are growing too, and they span six orders of magnitude:

Where we think this is going

This part is our view, not a reported trend. It comes from reading the sources above and from our own research.

1. Hours get cheaper; the right hours do not. Volume is growing fast: one platform alone takes in more than 160,000 hours of first-person video a month, and collectors talk about targets of 10 million hours. Generic data, read speech and simple household footage, will keep getting cheaper because anyone can produce it. What stays expensive is what is hard to produce: a particular kind of conversation, a particular place, a particular task done well, with consent on record. We have no price history to prove the first half; it is our inference from growing supply and the contrast between generic and specialized collection.

2. Voice is going vertical. General conversation is getting easier to collect; field conversations are not. The next voice models are for medicine, customer support, sales and law, and the real conversations in those fields are private by default. So the data will increasingly be performed: role-played by people who know how a nurse, an agent or a salesperson actually talks. That favors people with real field experience over generic speakers. We think the next scarce kind may be people explaining and debating ideas: someone talking through a paper they read, or two people arguing about an article. That is closer to how people actually use AI assistants than casual chat, and almost none of it is recorded with consent.

3. Faces cost more than voices. Consented video of people talking, listening and reacting is scarce in public datasets, and it carries a person’s likeness, not only their voice. Expect it to stay at a premium, and expect the consent terms to get stricter.

4. Physical AI is splitting in two. A wide, cheap layer of first-person video from many countries buys variety. A narrow, better-paid layer (teleoperators, skilled trades, real workplaces) covers the tasks where how the work is done matters. The second layer is small today. We think it is where the value per hour will be.

5. Expertise moves to judging, not just doing. As models produce more of the content, the scarce human work is checking it: is this clinical note right, is this research plan sound, is this agent’s answer safe. That is already the best-paid work in this market, and we expect it to grow faster than collection.

6. Consent becomes the product. Lawsuits are already applying biometric privacy law to robot cameras in Illinois. Buyers will pay more for data whose consent, rights and provenance are documented, and less for data that may have to be deleted later. The companies that can show who agreed to what will win the better contracts.

7. Smaller buyers move first. The largest labs build their own pipelines. The many startups on this map cannot, so they buy, license or partner, and they need data that exists now.

8. Long tasks need long recordings. Most first-person video today comes in short pieces: AFP describes a worker recording around 90 clips a day of about four minutes each (AFP via Khaleej Times). Robots and agents that must finish long tasks, cooking a whole meal or completing a repair, need the whole task recorded without cuts, along with what the person was trying to do at each step. Narration helps here: a person saying “now I’m checking the seal” turns audio into a label for the video. Our hypothesis is that complete, narrated recordings of long tasks will be worth far more per hour than short clips.

9. Failure may be the most valuable data. Almost everything collected today is a success: the demonstration that worked. But a robot or an agent in the real world spends much of its time recovering from mistakes, and successful demonstrations rarely show how. Robotics research has long trained on moments where a human expert steps in to correct a robot (HG-DAgger, 2018). The same applies to voice: misheard words, interruptions and corrections are what a real conversation is full of. Our hypothesis: as successful demonstrations become plentiful and cheap, recordings of things going wrong and being put right will become scarce and expensive.

10. The hard part for a data company is the second sale. A fixed dataset may generate only a limited number of sales. Non-exclusive licensing permits resale, but some buyers want custom or exclusive material, and an exclusive deal ends resale. We therefore expect recurring revenue to depend more on continuing collection and services than on a static catalog, which makes these companies dependent on a few buyers. The filings show it: at one listed vendor the largest customer was 58% of revenue, and at another the top five were 74% (our October 15 report).

11. People do not stay. A sign-up is not hours. We expect most contributors to record a little and stop, while buyers need people who keep going for weeks; one platform asks for at least 10 hours a week (CNN). So providers keep paying to recruit, and finding people who stay is a large part of what the price pays for.

12. Can the valuations hold? If a fixed dataset sells only a few times, how do the fastest-growing data companies keep growing? Mercor raised $350 million at a $10 billion valuation in October 2025 and charges an hourly fee on the expert work it places (TechCrunch), and it has since been reported to be in talks at about twice that. Our answer, as a hypothesis: the winners do not sell datasets, they sell hours of skilled work, and the demand renews because data expires. Each new model generation needs harder examples, evaluation sets leak into training and have to be replaced, and a model that has learned one skill needs data for the next. That is closer to staffing than to licensing. The risks are just as visible: a few labs buy most of it, and the largest labs can build their own pipelines. Figure wrote in August 2026 that “vendors couldn’t hit the throughput, diversity, or quality bar” its model required, “so we built the pipeline ourselves”. Whether the growth lasts depends on whether “harder data” keeps being needed faster than the labs can make it themselves.

13. Collection may run on Android first. Listings already accept recent models of both iPhone and Android. Our hypothesis is that differences in device cost, operating-system restrictions on camera and background recording, and local market share may make Android the faster or cheaper collection platform in some countries, especially where it is the phone most people carry.

What nobody tells the people in the data

Across all 17 companies, not one public source we found says what the people in the recordings were paid. The licensing deals are between companies, and the largest data partners go unnamed. If you are ever asked to take part in a project like these, ask four things:

• Consent you can read: who gets the recording, for what, and for how long.

• Everyone recorded: get clear agreement from everyone whose voice or details may be captured. Rules vary by place.

• Storage: who keeps it, and how.

• Pay for the time: what happens to a recording that gets rejected.

Best,
Kei & Kazuki

Would you like to take Glasp on the go?

With the Glasp mobile app, you can highlight and organize your favorite content anytime, anywhere. Stay productive on the move and never miss an insightful quote.

Download for iOS

Download for Android

Partner With Glasp

We currently offer newsletter sponsorships. If you have a product, event, or service you’d like to share with our community of learning enthusiasts, sponsor an edition of our newsletter to reach engaged readers.

Sponsor an Edition

If you want to reread or highlight this newsletter, save it to Glasp.

Save to Glasp

Every year, of Growth Unhinged and Maja Voje of GTM Strategist ask software and AI companies a simple question: what is actually working to find customers? The answer has a name in the industry, GTM (go-to-market): everything a company does to reach people and turn them into customers.

This year they surveyed 300 companies, up from 195 last year. The full report is free at stateofb2bgtm.com.

You might not sell software. Most of our readers do not. But nearly everyone is trying to get something they made in front of the right people: a product, a side project, a research paper, a portfolio, a job application. That is a go-to-market problem too, and the answers in this report travel further than you might expect.

We read all 41 pages. Here are seven findings that stood out to us as a two-person team, each with one thing you could try this week.

How the survey was done

First, who this is for: the survey is about B2B, companies that sell to other businesses. Maja asked us to say this plainly. If you run an e-commerce store or a consumer app, read it as a view into a different market, not a playbook. The 300 respondents answered in September 2026, and the report notes "a large sample of AI-native companies and those with deal sizes of $5-50k per year". [p.4] Answers are self-reported: they show what companies believe is working, not measured results. Read the percentages as signals of where attention is moving, not as proof. Two terms come up often. ARR (annual recurring revenue) is how big a subscription business is. ACV (annual contract value) is what one customer pays per year.

1. Everyone wants more pipeline

The most common answer to “what is your biggest GTM challenge?” was pipeline (a steady flow of potential customers), at 18%, ahead of scaling repeatably (10%) and positioning (9%). [p.9] It shows up in hiring too: growth marketing is the number one priority hire, for 22% of companies under $10M ARR and 16% above it. [p.10]

One respondent put the pipeline problem well: "We can generate content, outreach, product launches, and conversations efficiently, especially with AI, but the hard part is identifying the right buyer signals and channels that consistently convert into paying customers." [p.9]

Try this: write down where your last ten users, readers or clients came from. If you cannot answer, that is your pipeline problem, and it is the first thing to fix.

2. Founders are still the best salespeople

Founder-led sales is the most common primary GTM motion, at 26%. [p.11] Among companies under $1M ARR it is 44%. [p.13] Inbound and outbound only take over after the first million.

Try this: if you are early, do not hand your first conversations to a tool or a hire. Talk to the first ten people yourself. What you hear is the raw material for everything that comes later.

3. The oldest channel is still the best one

Network and introductions is the most widely used channel (62% call it core) and the one rated most likely to have a big impact. [p.16, p.19] It sits at the top right of the report’s scorecard, next to intimate events. In a year full of AI tools, the channel that wins is a warm introduction.

Note where agentic campaigns sit: bottom left, still "figuring it out", rated neither popular nor high-impact yet. That is exactly the finding Maja called her biggest surprise, and why she expects it to move next year. [p.19]

Try this: ask two people who already like your work for one introduction each. Be specific about who, and give them a two-line description they can forward without editing.

4. AI search is now a core channel

AI search, also called AEO (answer engine optimization: showing up in the answers of ChatGPT, Claude, Perplexity and Google’s AI), rose 11 points to 39% of companies calling it core, the biggest gain of any channel. [p.16, p.17] It moved into the “tried-and-true” quadrant, and it is again the number one place companies plan to increase investment next year, at 50%. [p.19, p.28] Last year it was also number one, at 51%. For companies under $5k ACV it is now tied for the second most popular channel, ahead of SEO, which was number one last year. [p.20]

Right behind AI search is signal-based outbound, at 48%: reaching out when a company shows a sign it might need you, such as a funding round or a new hire. [p.28]

Try this: ask ChatGPT and Claude the question your customer would ask, and see whether you appear and what they say about you. That answer is now part of your first impression.

Maja's take: the change she sees across the companies she works with is how they approach AI search. Some decide it is hot, put money in, and hire an agency to get it running. They do see results, but "whenever you are seeing it as a campaign, it is wrong from the start." It should be an ongoing standard operating procedure, stitched into content engineering or marketing engineering, so it becomes part of how you create, not a one-off "I need to show up in GPT". She has written about how to set that up: the living content graph and building a content system in Claude Code.

5. AI is finally paying off, and Claude leads the tools

43% of GTM teams now say AI has a big impact on their work, up from 24% a year ago. [p.33] Claude moved from fourth to first on the list of most impactful GTM tools, ahead of Clay, HubSpot and ChatGPT. [p.35]

The share seeing no impact at all fell from 22% to 4%. Last year 53% saw no impact or only a limited one; this year it is 28%. [p.33]

Asked which tools others should try, respondents again put Claude first. These were open-ended answers, and coding agents such as Cursor and Codex now sit on the same list as sales tools. [p.36]

Try this: pick one task you repeat every week, such as research before a call or a first draft of a follow-up, and hand only that to an AI assistant for a month. Keep the judgment for yourself.

6. Newsletters are still an experiment for most

Newsletters and editorial content are core for 34% of companies, which the report files under “still experimental”. [p.16] 18% plan to invest more in them next year, rising to 27% among companies under $1M ARR. [p.28, p.29]

Try this: if you write, publish on a fixed rhythm for eight weeks before you judge it. Experiments that stop at week two never become channels.

7. What did not work

The report’s list of failures is worth reading on its own. Spray-and-pray cold email, AI SDR sequences with “near-zero reply rate”, content generated in bulk by AI, and a $1M launch campaign with “little measurable ROI” all appear on it. [p.18]

Try this: before you automate any outreach, send ten messages by hand and count the replies. If ten personal messages do not work, a thousand automated ones will not either.

What surprised Maja

We asked Maja which finding surprised her most. Her answer: agentic campaigns. The survey assumes almost any motion can involve AI, but fully agentic plays, the dominant play for AI-native companies, were measured as a channel of their own. She expects them to become a serious go-to-market motion next year, “because if you have one or two people working in marketing, how on earth could they not use AI?” She called it an in-your-face finding.

The second was positioning. Pipeline and repeatable scale are “as old as earth”, she said, but how to stand out in the age of AI was “a very big and a very surprising one” for her. And some relief: last year she was “dead afraid” when more than half of respondents saw no tangible impact from AI yet. This year the results are much better.

If you are not selling anything

The hiring data is a career signal. When growth marketing is the most wanted hire at small companies, people who can find an audience, test a channel and explain the result are in demand, whatever their title. [p.10] The same skills help a student get a first job, a researcher get read, or an engineer get a side project noticed. And the top channel is not a skill you buy. It is the people who already trust your work.

The Glasp Take: relationships and machines

Kyle and Maja close with what they call the GTM barbell: “There are two ways to win right now: being exceptional at relationships and being exceptional at AI.” [p.39] That matches our own experience as a two-person team.

On the AI side, AI search has been our fastest-growing channel. We wrote about how we grew ChatGPT traffic 37x using our own server logs, and how we run multiple products with AI agents. It is good to see the survey agree that this is no longer a side bet.

On the relationship side, the report is a reminder that the warmest channel is still a person who vouches for you. Much of what we have learned came the same way: someone shared a link, an article or a highlight, and we followed it. That is also why Glasp exists. When you share what you read and what you learned from it, the people who follow you get the same head start.

The report’s last prediction is the one we will be thinking about most: “the next battleground will be positioning and messaging”, because “every company sounds the same”. [p.39] Tools can help you say more. They cannot tell you what is worth saying.

We also asked Maja what a two- or three-person team should do if it could invest in only one channel in 2027. "Rule number one is go where the audience is." To find out where that is: use AI for research, ask them, or do a little soul searching about where they would naturally find this kind of information. Then run a couple of experiments there, and stitch in agentic campaigns.

A question for you

Which channel is working best for you this year, and is it one the report rates highly or one it calls an experiment? Tell us in the comments.

Want to go deeper? Save Kyle’s full write-up to Glasp and highlight the parts you want to come back to. The full report is free at stateofb2bgtm.com, and Maja’s take is on GTM Strategist.

Would you like to take Glasp on the go?

With the Glasp mobile app, you can highlight and organize your favorite content anytime, anywhere. Stay productive on the move and never miss an insightful quote.

Download for iOS

Download for Android

Partner With Glasp

We currently offer newsletter sponsorships. If you have a product, event, or service you’d like to share with our community of learning enthusiasts, sponsor an edition of our newsletter to reach engaged readers.

Sponsor an Edition

“It’s not the manager’s job to prevent risks. It’s the manager’s job to make it safe for others to take them.”

— Ed Catmull, Creativity, Inc.

Today’s newsletter is brought to you by Stripe.

Watch four talks on AI pricing and usage-based models. Leaders from Replit, Fin, Snowflake, Clay, and Growth Unhinged discuss metering, variable spending, revisiting pricing, and forecasting AI costs and returns. Recorded at Monetize 2026, Stripe and Metronome’s pricing and billing conference.

Watch now

Every Pixar film you love was, at some point, bad.

That is not an insult. It is the central idea of “Creativity, Inc.,” the book by Ed Catmull, who co-founded Pixar and ran it, and later Walt Disney Animation Studios, for more than three decades. Early versions of the films were confused, flat or broken. What made Pixar different was not that its first drafts were good. It was what the company did with drafts that were not.

When did you last show someone your work before it was ready? Most of us wait until it is polished, because showing it early feels like exposure.

Subscribe now

A book about the people, not the hits

We expected a book about creativity: how Pixar finds ideas and makes hits. What we got was mostly about running an organisation. That surprised us, and then it made sense.

Stories, ideas and taste cannot be measured the way sales or uptime can. You cannot put a number on whether a scene works. As we read him, that is why Catmull put his attention on the things he could shape: the people who make the work, and the process they use to make it better.

The principle he returns to again and again is that the team matters more than the idea. In the publisher’s summary of the book’s lessons: “Give a good idea to a mediocre team and they will screw it up. But give a mediocre idea to a great team and they will either fix it or come up with something better.”

The Braintrust

Pixar’s best-known habit is the Braintrust. At points along the way, a director shows the current cut of a film in progress to a group of fellow directors, storytellers and studio leaders. They watch it and then say, candidly, what is not working.

The design has one rule that matters more than any other. The Braintrust has no authority. It cannot order changes. The director listens, takes the notes away, and decides what to do with them.

That rule is what makes candor possible. Nobody is defending territory, because nobody can take the film away. The notes are offered as help, not as a verdict. Catmull’s point, as we read it, is that honesty this direct does not happen by itself in any group of people. It has to be built into how the company works, so it does not depend on one person’s courage on one day.

The Beast and the Ugly Baby

Catmull has two images for the pressures inside a creative company. The Beast is the constant demand for output: schedules, deadlines, the next release. The Ugly Baby is every new idea in its early form, awkward and easy to dismiss.

The Beast is always hungry, and it will happily eat the Ugly Baby. Catmull’s job, as he saw it, was to protect the new and fragile long enough for it to grow up.

He applies the same thinking to mistakes. Problems will happen, so the goal is not to avoid them but to find them early and fix them together. Again from the publisher’s summary of the book: “It’s not the manager’s job to prevent risks. It’s the manager’s job to make it safe for others to take them.”

The founder behind the book

Catmull rarely talks about himself in the book, but his resolve shows through. He is not a loud founder. He comes across as someone with a quiet, strong core and deep passion underneath. Of all the founders whose stories we have read, he is one of the easiest to relate to.

There is also a surprising portrait of Steve Jobs, who owned Pixar for years. At Apple, Jobs was famous for imposing his will. At Pixar, by Catmull’s account, Jobs agreed never to sit in on Braintrust meetings, because his presence would change the room. Seeing Jobs through Catmull’s eyes is one of the book’s quiet pleasures.

If you would rather hear him tell it, Catmull walks through these ideas in his 2014 talk at Stanford, and Glasp has the video with a summary and transcript.

The Glasp Take: Share the Ugly Draft

The Braintrust is, at heart, a system for sharing notes. Someone shows unfinished work. Others respond honestly. The owner keeps control and decides what to use. Everyone gets better, including the people giving the notes, because they see how others solve problems.

Most knowledge work does not have a Braintrust. People read, think and draft alone, and share only the finished version. The ugly baby never leaves the room, so nobody can help it grow.

Glasp is a small step toward the other way of working. Glasp stands for Greatest Legacy Accumulated as Shared Proof. When you share your highlights and notes while you are still reading, you are showing work in progress: what caught your eye, what confused you, what you are not sure about yet. Others can see how you are thinking, and you can see how they are. Nobody has authority over your notes. You decide what to take from theirs. It is the same reason we keep saying you should learn in public, before you feel ready.

One more idea from the book applies here, in our reading: a process is there to help people think, not to think for them. Sharing notes does not replace reading. It just means you do not have to read alone.

The safe room in the AI era

Many of us now get our first notes from an AI. We paste in a draft and ask what is wrong with it. That can help, and Catmull’s principles suggest how to use it well. What follows is our reading of them, not something he said.

Keep the authority with the author. The Braintrust works because it cannot order changes. Treat AI feedback the same way: notes to weigh, not a verdict to accept.

Protect the ugly baby. A model learns from what already worked, so its edits tend to pull a strange early idea back toward the familiar. If a draft feels new, do not let the first round of machine edits sand it down.

Keep people in the room. An AI can tell you a sentence is unclear. It is less likely to tell you that you are avoiding the hard part, the way a colleague who knows you can. Use the machine for speed, and keep a small room of humans for candor.

A question for you

What are you working on right now that you have not shown anyone yet, because it is not ready? Who would you show it to if you knew they would be honest and had no power over it? Tell us in the comments.

Forwarded this email? Subscribe to Glasp Newsletter

Start highlighting the web with Glasp

Would you like to take Glasp on the go?

With the Glasp mobile app, you can highlight and organize your favorite content anytime, anywhere. Stay productive on the move and never miss an insightful quote.

Download for iOS

Download for Android

Partner With Glasp

We currently offer newsletter sponsorships. If you have a product, event, or service you’d like to share with our community of learning enthusiasts, sponsor an edition of our newsletter to reach engaged readers.

Sponsor an Edition

AI models are trained on data, and much of the data made specifically for training and testing them comes from paid human work: doctors grading medical answers, engineers writing code for models to learn from, speakers recording conversations, and workers filming their hands as they fold laundry so that robots can learn. The listed vendors we examined reported combined growth in 2025, and private companies report figures from hundreds of millions to billions of dollars under differing definitions. Nobody measures the market as a whole.

So we measured what we could. We built sourced profiles of 1,297 companies across 64 countries, coded each one against a written codebook, read the annual reports of the listed data vendors that report this business separately, and recorded 117 pay offers that were live on October 15. This report is what we found, including what we could not find.

By the numbers

• 1,297 AI labs, data vendors and work platforms mapped, in 64 countries

• 300 demand-side companies with explicit public evidence of obtaining human data

• $718M: 2025 AI-data revenue in the filings of four listed vendors, up 32%

• $150 vs $20: median advertised hourly ceiling for coding work versus voice work

• 5.6x: the gap between the highest and lowest language rates for the same kind of work on one platform

If you want to reread or highlight this newsletter, save it to Glasp.

Save to Glasp

Key findings

1. Demand depends on who is asking. Of 300 demand-side companies with explicit public evidence of obtaining human data, the ten frontier labs all show demand for preference and evaluation data and for code, and six of them for agent or RL-environment data. AI product companies stand out for expert data (68%). Robotics companies want video (90%) and physical demonstrations (100%).

2. The audited part we could find grew 32% last year, and three of the four vendors disclose heavy customer concentration. The four listed vendors we found whose filings break out AI-data revenue reported $544 million in 2024 and $718 million in 2025. At Innodata, one customer was 58% of revenue; at Appen, the top five were 74%.

3. The biggest names publish no audited numbers for this business. Scale AI, Surge AI, Mercor, Handshake, micro1 and Turing are private. The figures they or the press report mix revenue, ARR and gross run rate, and range from hundreds of millions to more than $2 billion. They are not comparable with filed revenue or with one another, but they indicate substantial activity outside the four-company audited floor.

4. In our sample, advertised pay ceilings were highest for credentialed work and varied by language. The median advertised ceiling is $150 an hour for coding work and $20 for voice work. On one platform, the same kind of language work is posted at up to $31 an hour in Japanese and $5.50 in Tamil. What people actually earn is a much thinner record.

5. Physical AI has two overlapping supply chains. In our sample, robot makers most often disclose teleoperation and simulation (40% each), while data vendors and work platforms most often advertise video of people doing tasks. Of 316 physical-AI companies, 135 are vendors.

How we did this

Sample. Between October 13 and 14, 2026, AI research agents built a company-level profile for each company, with every fact tied to one source URL (more than 42,000 facts in the current version). The sample started from public lists and searches for AI labs, data vendors, AI work platforms, robotics companies and AI product companies whose job posts, papers or press describe human data. It is purposive, not a census. It covers 64 countries; 69 companies are in China or Hong Kong, a smaller share than China’s real weight in this market.

Coding. An LLM coded each profile against a written codebook: one primary role per company, the strength of evidence that it obtains human data, the data types it seeks or sells, seller capabilities, and, for physical AI, collection method and robot type. A second, stricter pass re-judged every buyer: “explicit” means the sources name a concrete act of obtaining data from people outside the company (paying or recruiting contributors, a collection programme, a data or annotation job post, a named vendor or licensing deal, in-house annotators, participant studies). Web crawls, public datasets, a company’s own users’ data and synthetic data do not count.

Checks. We did not rely on one model. A stratified random sample of 130 companies was checked against the live source pages: the first 50 independently by Claude and by Codex, the next 80 by Codex. The first round found 4 of 25 “explicit” labels wrong (the people involved were rating other companies’ models, or were the companies’ own customers), so both models re-judged all 402 “explicit” companies against a stricter rule; they agreed on 326 (263 kept, 63 dropped), Claude settled the other 76 against the written rule, and 300 remain. In the second round, 39 of 40 sampled “explicit” labels held, while 3 of 16 sampled companies coded as not explicit did have explicit evidence. These validation samples are small and do not establish the size or direction of the remaining error. The second round also showed that “frontier lab” had been applied too widely, so Codex re-judged all 76; in this report the term is an editorial grouping limited to the ten companies named in the role table, not a ranking. Other primary roles held for 111 of the 130 sampled companies before that fix; half of the misses were between data vendor and work platform, which the seller analysis treats together. Data types held for 112 of 130 and seller capabilities for 29 of 37. Treat these fields as model-coded estimates, not audited labels. No human reviewed the labels; these are model checks against public pages. One record lost in processing was coded by hand with the same codebook.

Terms. “Obtaining” human data covers buying it from vendors, licensing it, collecting it in-house and running participant studies. A vendor’s “gross run rate” is annualized customer spend passing through its platform, much of it paid on to workers; “revenue” in a filing is what the vendor reports for a fiscal year. The two are not comparable.

Limits. Public silence is coded as unknown, not as absence. Frontier labs mostly buy expert data through vendors, which their own pages rarely show, so expert demand at labs is understated. No individuals appear in this report: everything is at company level. Our sources are mostly English-language, so China is underrepresented (69 of 1,297 companies) even though its annotation industry is large and state-backed. “All companies” figures therefore describe mainly the US and European market; Chinese results are shown separately and should be read as indicative.

Who is in the market

• Model companies (including large general models below the frontier, and language, regional, domain, speech, video and image models): 381

• Data vendors: 279

• AI product companies: 212

• Academic and nonprofit (universities, institutes, archives): 182

• Robotics and physical AI: 126

• AI work platforms (recruit and pay contributors): 102

• Frontier labs (Anthropic, Google DeepMind, Meta, Mistral, OpenAI, xAI, DeepSeek, Moonshot, Alibaba, ByteDance): 10

• Other: 5

• Total coded: 1,297

By headquarters: United States 508, Europe 386, China and Hong Kong 69, 236 in other countries (India, Japan, Canada and Korea lead), and 98 that do not publish one.

What buyers ask for

Three patterns stand out.

• Frontier labs buy judgment and practice. All ten frontier labs show demand for preference and evaluation data and for code, and six of the ten for agent trajectories or RL environments, against 11% of other model companies. With only ten companies, read these as counts, not shares.

• Applications buy experts. Companies building legal, medical and financial products on top of others’ models need specialists to write and grade examples: 68% of AI product companies show expert demand, though only 22 of them have explicit evidence.

• Regions differ. US buyers lead in preference and evaluation data (57%, against 37% in Europe) and code (19% against 6%). Chinese buyers in our sample (23) lean toward video (87%), agents and RL environments (52%) and code (48%).

One caution on reading the chart: each company has one primary role, and companies that build physical-AI models are counted under Robotics. That is why the other columns show little physical-AI demand; it does not mean those companies never want such data.

What sellers offer

• Broad claims are common. 116 of 381 sellers (30%) list six or more capabilities; only 34 list one. Broad lists make it hard for buyers to tell real specialization from a menu.

• Code and RL environments are offered mostly from the US. In our sample, US-headquartered sellers state code (47%, against 11% in Europe) and RL environments (38% against 12%) far more often.

• The crowd and language layer sits elsewhere. Sellers outside the US and Europe lead in crowd annotation (80%) and multilingual work (68%).

• Many sellers already offer physical-AI collection. 136 sellers offer physical-AI data collection, more than offer code (100) or RL environments (83).

How big is the market?

There is no single answer, so we built it up from the parts that can be checked.

The audited floor

Only a few listed companies report revenue for AI-data work separately. Those we found give a floor, not a total; the search was not exhaustive.

• Appen (ASX), group revenue, all AI data: $234.3M in 2024, $230.8M in 2025 (-1.5%)

• Innodata (Nasdaq), digital Data Solutions segment: $141.1M in 2024, $220.9M in 2025 (+57%)

• TaskUs (Nasdaq), aI Services: $135.1M in 2024, $214.2M in 2025 (+59%)

• Haitian Ruisheng / SpeechOcean (Shanghai), revenue, all AI data (RMB converted at 7.19): $33.0M in 2024, $52.4M in 2025 (+59%)

• Total: $543.5M in 2024, $718.3M in 2025 (+32%)

Sources: Appen FY25, Innodata 10-K, TaskUs 10-K, Haitian Ruisheng 2025 annual report. Innodata’s segment also includes model deployment work, so it slightly overstates data work. TELUS Digital does not disclose AI-data revenue and was taken private in 2025.

What the filings show beyond the totals:

• Customer concentration is high, and rose at Innodata and Appen. Innodata’s largest customer went from 48% to 58% of revenue. Appen’s top five went from 67% to 74%. Meta was 26% of TaskUs revenue. When Google ended its Appen contract, Appen’s 2024 revenue fell 14%.

• Appen’s growth is in its China segment. The segment Appen reports as Appen China grew 75% in 2025, to $102.9 million, and was 64% of Appen’s revenue in the first half of 2026, while Appen Global fell (Appen H1 2026). One company does not make a market shift, but it is the clearest regional signal in the filings.

• At one Chinese vendor, vision overtook speech. At Haitian Ruisheng, computer-vision revenue tripled (+218%) while speech fell 9%.

Large private-company indicators

The best-known companies are private and publish no audited numbers. What they or the press report:

• Mercor: More than $2B, gross run rate (net estimated $600M to $800M), Jun 2026, annualized (Dealroom, citing The Information)

• Surge AI: Over $1B, revenue (reported by press), 2024 (Reuters, syndicated)

• micro1: More than $1B, run rate (company claim), Oct 2026, annualized (company announcement)

• Handshake (AI training): Nearly $1B, gross run rate (net close to $300M), Apr 2026, annualized (Dealroom, citing The Information)

• Scale AI: About $870M, revenue (reported by press), 2024 (Reuters, syndicated)

• Snorkel AI: $375M, run rate (company figure; Reuters: above $350M), Sep 2026, annualized (Snorkel)

• Turing: About $300M, aRR (company figure), Mar 2025 (TechCrunch)

These mix years, gross and net revenue, and services beyond data, so we do not add them up. Platforms that pay contractors pass a large share of gross revenue on to the people doing the work, so a gross run rate is closer to customer spend than to revenue. Read this table as a sign of scale, not as a second floor.

The research-firm estimates

Market-research firms publish figures under different definitions, which is why they disagree. “AI training datasets” (datasets sold as products) is put at about $3 billion to $3.4 billion in 2025 by Grand View Research, Precedence Research and Global Market Insights. Broader definitions run to $9.8 billion for training data as a segment of AI training in 2025 (Next Move) and $57.6 billion for all data-labeling solutions and services by 2030 (Grand View Research).

The narrow figures are hard to reconcile with the private companies above. The likely reason is definitional: expert feedback, evaluation and RL environments are sold as services and fall outside “datasets”. Our read: the audited floor we found is about $0.7 billion. Individual private-company indicators reach more than $2 billion annualized, but they mix years and metrics and cannot be added. Nobody has measured the total. Anyone building a business case should size it bottom-up from buyers’ budgets, not from a headline number. One widely quoted buyer-side figure, that each leading lab spends around $1 billion a year on human data, comes from a vendor’s chief executive (TIME), so it is an industry view, not a measurement.

Three areas to watch

RL environments

Some labs now buy practice, not just examples: software environments in which a model attempts tasks and an automatic checker scores the result. Scale says “nearly half” of its new data projects involve RL environments (Scale). In interviews with 18 people from environment startups and labs, Epoch AI found most tasks priced at $200 to $2,000 each, with contracts often six to seven figures per quarter and exclusive deals costing roughly four to five times more (Epoch AI). Replicas of websites for agents to practice on cost about $20,000 each (SemiAnalysis). Labs rarely name these vendors publicly, so most of this is reported, not disclosed.

Where an answer cannot be checked by a program, the reporting cited here describes experts writing the rubrics that grade it.

What synthetic data can and cannot replace

Several prominent public model reports show a consistent but limited pattern. In pre-training, the Stanford AI Index 2026 finds “no definitive evidence that synthetic data can fully offset real-data depletion”; mixing real and synthetic data speeds training “by a factor of five to 10 at scale, without surpassing real data in final model performance” (AI Index 2026). Microsoft’s Phi-4 used 40% synthetic data in pre-training, but its synthetic-only models “underperformed on the knowledge-heavy benchmarks and demonstrated increased hallucinations” (Phi-4 report).

In post-training, synthetic data does much of the volume, but people remain in the loop. Meta’s Llama 3 says most of its fine-tuning data is model-generated, yet it ran six rounds of human preference annotation and found that training its largest model on its own outputs without a checker “is not helpful” (Llama 3). DeepSeek-R1 and Qwen3 both used human annotators to refine their cold-start reasoning data (DeepSeek-R1, Qwen3).

What these public cases suggest: synthetic data substitutes for people most where a cheap automatic checker exists (math, code with tests). Where quality needs judgment, the labs in these reports still used people.

Physical AI

Our sample holds 316 physical-AI companies: 135 data vendors, 126 robotics companies, 29 work platforms and 26 others. Robotics companies lean on teleoperation (40%) and simulation (40%); vendors and platforms lean on first-person video of people (59% and 79%). Robot arms and manipulation are the most common target (52%), ahead of mobile and household robots (37%) and humanoids (16%).

On the buyer side, the clearest disclosure is Generalist AI: its GEN-0 model was trained on about 270,000 hours of real-world manipulation data, growing by more than 10,000 hours a week, collected through its own devices and “multiple data foundry partners” it does not name (Generalist AI). Figure says it has begun “human video capture” in Brookfield environments to build a humanoid pretraining dataset (Figure). On the seller side, Scale says it delivered more than 150,000 hours of physical-AI data and added 10 robotics customers in 2025 (Scale). We found no robotics company that names its data supplier.

Two regional patterns deserve a closer look. Europe has the most physical-AI companies in our sample but the least first-person video collection (19%, against 42% in the US). In China, one analysis citing the state think tank CAICT counts 64 embodied-AI data-collection facilities in at least 27 cities (analysis), consistent with China’s high share of teleoperation (45%) in our sample.

Work and pay

What platforms advertise

On October 15 we recorded 117 pay observations from 19 platforms, 109 of them read on the platforms’ own pages. The medians below use the 103 that were confirmed hourly rates in US dollars; the chart below shows the ten groups that hold 102 of them. These are advertised ceilings, not earnings. Platforms are not weighted equally: Mercor’s listings are counted one by one, while DataAnnotation posts one rate per tier.

In our October 15 sample, median advertised ceilings were higher in credentialed categories. They also varied by language and location. On Outlier’s own language pages, the same kind of work is posted at up to $31 an hour for Japanese and Korean, $29 for German, $10 for Brazilian Portuguese, $7.50 for Hindi and Indonesian, and $5.50 for Tamil (Outlier).

Separately, across all 194 hourly roles on Mercor’s public listings that day, the median advertised maximum was $90 (Mercor).

What people actually get

The record of real earnings is thinner and older. The 2025 Fairwork ratings found that only 4 of 16 cloudwork platforms could show workers consistently earn at least the local minimum wage after costs, and 31% of 750+ surveyed workers had gone unpaid at least once (Fairwork, via Workplace Insight). The ILO found microtask workers spending about 20 minutes of every hour on unpaid work such as searching for tasks and taking qualification tests (VoxEU). Three gaps recur between the posted rate and the paycheck: unpaid time, projects that end without notice (xAI cut about 500 generalist tutors in one night, TechCrunch), and rates that reset between projects.

We found no study of realized hourly earnings on the expert platforms of 2025 and 2026. That is a gap, not evidence that their pay matches their ads.

Voice data: what it costs and who gets what

Only one vendor we found publishes a full price list for speech data: transcribed, consented two-person conversations at $60 to $95 per audio hour, priced by the speakers’ home economy, $90 for US English and $65 for Hindi or Swahili (SpeechData.ai). Older research corpora cost $7 to $12 an hour (LDC Fisher), and Mozilla Common Voice offers 42,593 hours in 295 languages for free. On the contributor side, in September 2025 the Neon app advertised payments equivalent to $9 to $18 per hour of recorded calls (Malwarebytes); it went offline weeks later after a flaw exposed users’ recordings, transcripts and phone numbers (TechCrunch), a reminder that a recorded call also involves the other person on the line. The buyer price also pays for transcription, quality control, consent and a commercial license, so the two numbers cannot be compared directly.

The rules coming next

• Sep 2022, US: FTC policy statement on gig work. A non-binding statement of enforcement priorities: misleading pay claims and opaque algorithmic management.

• Aug 2023, China: Generative AI interim measures, Article 8. Providers must set clear annotation rules and train and check annotators.

• Jan 2025, China: NDRC opinions on the data annotation industry. Targets over 20% annual growth to 2027, annotation bases and national skill standards; no wage rules.

• Aug 2025, EU: AI Act general-purpose AI obligations. Apply from August 2025, including a public summary of training data; the Commission’s enforcement powers start in August 2026.

• Dec 2, 2026, EU: Platform Work Directive transposition deadline. A presumption of employment where facts indicate control, applied under each country’s law, plus transparency on algorithmic management; few member states have finished.

• Ongoing, US: Worker classification suits. A proposed $12.5 million settlement of a contributor misclassification class action against Scale AI awaits final court approval; the defendants deny liability (settlement notice).

Recording in homes, recording calls and processing voices or images can trigger privacy and consent laws. Biometric data processed to identify a person uniquely is a special category under GDPR Article 9, and Illinois’s BIPA covers identifiers such as voiceprints and scans of hand or face geometry; an ordinary photo or recording is not automatically a biometric identifier. Recording a call may also require consent from everyone on it, depending on where they are.

What this means

For buyers: demand is specific to buyer type, many sellers list broad capabilities, and the labs that buy the most do not name their suppliers. Ask vendors for evidence by capability, not a list of capabilities.

For sellers: the four audited vendors grew 32% combined, and three of them disclose heavy dependence on a few customers, so losing one major contract can move a year of revenue. In the current sample, code and RL environments are offered mostly by US sellers, while physical-AI collection is already offered by more than a third of all sellers.

For people doing the work: the advertised rates for credentialed work are high and public; the realized record is thin, and the gaps are unpaid time, abrupt project ends and rate resets. The rules differ sharply: the US item above is a non-binding FTC statement, China regulates annotation practice, and the EU has separate AI and platform-work rules. EU countries are required to transpose the Platform Work Directive by December 2, 2026.

Before you take an AI-training job: five checks

If you are considering this work, the record above suggests asking five questions before you start.

1. What do people actually earn, not the ceiling? The advertised rate is an “up to”. Ask how many hours of paid tasks a typical contributor in your field and country gets per week.

2. Is the qualification test paid? Tests and onboarding may be unpaid. Ask how long they take, whether you are paid if you do not pass, and whether the platform may use your test work.

3. How and when are you paid? Check the payment method, the minimum payout, how often you are paid, the rejection criteria, whether rejected work can still be used, and how to appeal a rejection.

4. What happens when the project ends? Projects can stop overnight and rates can reset on the next one. Ask how long the project is expected to run and how you will be told if it ends.

5. What happens to your data and your work? Check what the platform collects about you (ID documents, voice, biometric data, video from your home), who receives it, how it is secured, how long it is kept, whether you can withdraw consent or ask for deletion, and who owns what you produce. Do not record other people without their consent.

Why Glasp looked at this

More than 3 million people use Glasp to highlight and learn from what they read. This report used public company and platform sources only, not Glasp users’ account data, highlights or reading histories. Much of what AI companies now pay people to do is close to what our users do every day: read carefully, judge what matters, and explain why. We mapped this market to understand that overlap. We are also studying whether, in future, Glasp could help people with the right skills find this kind of work through partners. Nothing has been decided; anything we did would be opt-in, separate from Glasp itself, and run with partners who handle verification and payment. We will keep tracking this market and share what changes.

Talk to us

• Researchers, professionals and people doing this work: if you are interested in paid AI-training work in your field, or have done it and are willing to share your experience, reply to this email. We read every reply. This is a research request, not recruitment or an offer of work, and we do not run a contributor pool. Please do not send a CV, a job application, ID documents, client information or other sensitive personal data.

• AI labs, companies and data vendors: if you want a cut of this data for your segment, region or data type, or want to compare notes on what you see, reply to this email.

• Named companies: if we got a fact about your company wrong, reply with the correction and a source, and we will fix it.

Disclosure. Glasp sells sponsorships in this newsletter, is studying whether it could in future help its users find paid AI-training work through partners, and may contact companies named in this report. No company paid for this report, reviewed it, or saw it before publication.

What would you like us to dig into next: the buyers, the pay, or the physical-AI supply chain? Tell us in the comments.

Sources

Company sample and all database shares: Glasp research database, 1,297 company profiles built October 13 and 14, 2026, coded October 15. Revenue floor: company filings linked above. Private-company figures: self-reported or reported by the press as noted. Pay: platform pages read October 15, 2026. All other figures are linked inline.

Partner With Glasp

We currently offer newsletter sponsorships. If you have a product, event, or service you’d like to share with our community of learning enthusiasts, sponsor an edition of our newsletter to reach engaged readers.

Sponsor an Edition

Subscribe to Glasp

Subscribe now

“I wish to avail myself of all that is already known and then if possible add my mite to help on the future worker who will attain final success.”

— Wilbur Wright

* Today’s newsletter is brought to you by Fiber.

Get paid to make AI training data with a friend (US only)

Fiber is oto's app for recorded video chats that become AI training data.

• Talk for 10 minutes with someone you know, each in your own room, or get matched with another participant.

• $20 to $30 per eligible recorded hour, per participant. Rate confirmed before you record. Hours not guaranteed.

• Face and voice recorded for AI datasets, including commercial use.

• US residents 18+, except Washington, Illinois, Texas, Colorado and Connecticut.

See pay and requirements

In July 1969, when Neil Armstrong stepped onto the Moon, he carried a small piece of history in his personal kit: a scrap of muslin from the wing of a flying machine built in a bicycle shop in Dayton, Ohio, sixty-six years earlier.

The machine had flown for twelve seconds. It covered 120 feet, less than half a football field. And it changed what humans believed was possible.

We usually tell the Wright brothers’ story as a story about genius, or courage, or two stubborn men who refused to quit. All of that is true. But reading David McCullough’s biography again, something else stood out to us. This is also a story about notes: who wrote them down, who shared them, who hid them, and who was willing to cross them out and start again.

Enjoyed this issue? Subscribe for free to get the next one in your inbox, or become a paid subscriber to support our work.

Subscribe now

They started with other people’s notes

Wilbur and Orville Wright had no college degrees, no laboratory and no government grant. What they had was a bicycle business, a habit of reading, and a letter.

In 1899, Wilbur wrote to the Smithsonian Institution and asked for whatever it had published on human flight. The brothers read everything they could find, including the work of Otto Lilienthal, the German engineer who had made thousands of glides from a hill near Berlin before he died in a crash in 1896, and the designs of the American engineer Octave Chanute.

Lilienthal mattered for more than his data. He had shown, in public and on camera, that a person could leave the ground on wings and come back. Before him, flight was a fantasy. After him, it was a problem. Problems can be worked on.

Then they found the notes were wrong

In the summer of 1901, the brothers took a new glider to the dunes at Kitty Hawk. It flew worse than their first one. It dove, stalled and fought the controls. Wilbur wrote in his diary that they “found the machine less manageable than expected.” Something was wrong.

The problem was in the numbers they had trusted. The curve of their wing had been based on Lilienthal’s recommended values, and the lift tables everyone relied on did not match what the air was actually doing.

This is the moment that, for us, defines the brothers. They did not give up, and they did not keep blindly following the experts. That fall they built their own wind tunnel in the bicycle shop and measured lift and drag themselves, on some 200 wing shapes. McCullough treats their decision to work out the science on their own as a turning point.

When they turned to propellers, they found an even stranger gap. There was simply no usable written theory of how an air propeller should be designed. Even for ship propellers, which had been in use for about a century, the library had nothing that explained how they worked. Nobody had written it down. So they worked it out and wrote it down themselves.

They kept careful records

Wilbur filled notebooks with observations of birds. He noted how a buzzard holds its wings in a shallow V and rides the wind without flapping. He noted how hawks hold steady in strong gusts. Distances were measured. Mistakes were recorded. Every season at Kitty Hawk was written down and studied at home over the winter.

Their famous telegram from December 17, 1903 reads like a lab note: success four flights Thursday morning, all against a twenty-one-mile wind, longest fifty-seven seconds, inform press, home Christmas.

Open versus secret

The most striking contrast in the book is between two men who both wanted to fly.

Octave Chanute believed that knowledge and ideas about flight should be exchanged openly. He wrote to the Wrights, visited their camp and shared what he knew. Samuel Langley, the head of the Smithsonian, was the opposite. In McCullough’s telling, he kept his own work close.

Langley also had what the Wrights never had: public money and the backing of the government. His project had cost nearly $70,000, the greater part of it public money. On December 8, 1903, his Aerodrome was launched over the Potomac River and fell straight into the water. The press turned on him, and his government backing ended.

Nine days later, two men flew at Kitty Hawk whose expenses from 1900 to 1903, including materials and every trip to the dunes, came to a little under $1,000, paid out of the modest profits of their bicycle shop. Chanute had offered to help find a wealthy backer, adding that he happened to know Andrew Carnegie. Wilbur declined. “Practically all the expense of our aeronautical experiments lies in the time consumed,” he wrote.

Money did not decide the race. Will did.

Their records became their proof

The brothers wrote everything down: dated diaries, notebooks of bird observations, wind tunnel tables, more than 400 letters exchanged with Chanute, and photographs of their gliders, including the wrecks. McCullough’s point is that few pioneers left so complete a record, step by step, in writing and on film.

That habit turned out to matter far beyond the workshop. Once flight succeeded, others came forward to claim it, and for years the Smithsonian itself promoted Langley’s Aerodrome as the first machine capable of flight. The brothers’ answer was always the same: the record. It took decades, but in 1942 the Smithsonian finally stated that the credit for the first successful powered, piloted flight belongs to the Wright brothers.

Their nephew Milton, who grew up around the shop, remembered the glue pot and the spruce shavings on the floor. Watching his uncles glue strips into ribs for their “flying machine,” he assumed “all bicycle shops did the same thing.” It was, he said, “all very commonplace.” Most important work looks like that while it is happening. The only way it survives is if someone writes it down.

When Chanute died in Chicago in November 1910, Wilbur wrote a long public tribute to the man who had shared what he knew: “His writings were so lucid as to provide an intelligent understanding of the nature of the problems of flight.”

Secrecy did not cause Langley’s failure by itself. But it is worth noticing which way the knowledge flowed. The brothers stood on what Lilienthal and Chanute had made public. They tested it, corrected it, recorded what they learned, and in time showed it to the world. The record of their work traveled all the way to the Moon.

The Glasp Take: Leave Notes Someone Can Fly On

When we read this book in our reading group, one thought kept coming back. Progress happens when knowledge is spread, kept and left in a form other people can use. The Wrights could start because someone else’s work was visible. Their work could last because they wrote it down. And once two people showed that flight was possible, a whole world of builders followed.

Our name carries the same idea. Glasp stands for Greatest Legacy Accumulated as Shared Proof. The Wrights’ notebooks were exactly that: proof, accumulated, that could be shown to anyone who doubted. That is the chain Glasp exists to strengthen. Most of what we learn still goes the way of Langley’s results: kept to ourselves, in notebooks no one else opens, until it is gone. That is the reason we started Glasp, and why we keep arguing that we should learn in public. A highlight you share, a note you leave on something you read, the Kindle highlights you finally take out of your device, a correction you make when the “expert numbers” turn out to be wrong, are small acts in the Lilienthal and Chanute tradition. You may never know who uses them. That is fine. Lilienthal never saw Kitty Hawk. As we wrote in You are reading a book that has already been read, every reader stands in a long line of earlier readers.

There is one more lesson we take from the brothers. They read everything, and then they checked it against the wind. Shared knowledge is a starting point, not a final answer. The best thing you can do with someone else’s notes is test them, improve them and pass them on.

A question for you

Whose notes are you standing on right now? It could be a book, a teacher, a stranger’s blog post or a thread you once highlighted. Tell us in the comments. And if you have ever found that the “expert numbers” were wrong, we would love to hear that story too.

Read more from Glasp

• Note-taking Lessons From America’s Greatest Biographer (how biographer Robert Caro researches and takes notes)

• 5 Incredible Biographies You (Probably) Haven’t Read

• The Greatest Legacy For Future Generations

• The Wright Brothers on Glasp: if you read it on Kindle, your highlights can be the first ones others see

• How to turn your Kindle highlights into lasting knowledge

Forwarded this email? Subscribe to Glasp Newsletter.

Subscribe now

Would you like to take Glasp on the go?

With the Glasp mobile app, you can highlight and organize your favorite content anytime, anywhere. Stay productive on the move and never miss an insightful quote.

Download for iOS

Download for Android

Partner With Glasp

We currently offer newsletter sponsorships. If you have a product, event, or service you’d like to share with our community of learning enthusiasts, sponsor an edition of our newsletter to reach engaged readers.

Sponsor an Edition

If you want to reread or highlight this newsletter, save it to Glasp.

Save to Glasp

Popular Episodes at a Glance

• #1: How We Found and Acquired Our First 1,000 Users

• #3: How We Rode the AI Wave with Side Projects Before It Exploded

• #10: When the Side Project Outgrows the Mission

• #17: How We Grew ChatGPT Traffic 37x Using Our Own Server Logs

• #18: How a Two-Person Team Uses AI to Run Multiple Products With 3.5 Million Current Installs

Hi, Kei and Kazuki here.

Glasp is two people. Between us we maintain browser extensions, a web app, native mobile apps, a separate AI product, and a research practice. About 3.5 million browsers currently have a Glasp extension installed.

In #18, we described the loop most of our systems share: a signal, a queue, a bounded worker, a verification step, and a retirement rule, and applied it to growth, content, and measurement. Last time, we covered our research. This issue keeps the rest of the promise from #18.

It applies the same loop to two kinds of work that are usually handed to separate teams, answering users and rebuilding products, and ends with the agent workflow underneath both.

In this issue:

• Customer support as a product system. How feedback from Chrome Web Store reviews, App Store and Google Play reviews, and support email becomes one queue, what Claude drafts, and what a person still decides.

• Rewrites that are not finished when the code ships. What rebuilding both native mobile apps taught us about measuring a rewrite and updating the world’s memory of a product, and why human patches, not AI-written code, became the technical debt in our Chrome extension.

• The agent workflow underneath. Contracts instead of standing instructions, why we stopped relying on written rules and built a harness, Claude Code and Codex reviewing each other, more workspaces than people, how an idea reaches the right session from a phone, and how we keep old Macs running agents all day without overheating.

The first section is open to everyone. The rest is for paid subscribers.

* This spot is open for your product.

Hiring? Post a job in the Glasp Newsletter

Put your open role in front of 550K+ subscribers: founders, builders, students, and knowledge workers, with 250,000+ views per send across email and web. $199 per post, at most three roles per issue, and every listing is reviewed before it runs.

Post a job

Support is a product system, not an inbox

At our scale, customer support cannot mean two founders reading every message from the top of an inbox and replying faster. There is not enough time, and even if there were, speed is the wrong goal.

The first layer of our support is not a model, and it was never about keeping people away from us. Between our first 100 and first 1,000 users, we ran more than 750 onboarding calls on Zoom, 15 to 20 minutes each, as we wrote in Hatching Growth #1. Getting through onboarding was the part we protected most, because a user who gets past it usually stays. When someone is stuck, we still send a Zoom link or a calendar invite and walk through it together. When something breaks, we ask by email, in detail, how to reproduce it: which browser, which page, what they clicked, what they saw. Even when a problem first appears in a public review, we move the conversation to email, because a review cannot tell us how to reproduce the bug. The result is a lot of email. That volume is a choice, and it is the reason the next layer exists.

The second layer is a queue. Feedback reaches us through several doors: Chrome Web Store reviews, App Store and Google Play reviews, support email, Reddit, and YouTube comments. Each arrives in a different format, in many languages, about different versions of different products. Before anyone can answer well, each item has to become the same kind of object.

A useful support item carries more than the message:

• product, platform, and version;

• language;

• the kind of problem;

• account and payment state, when it matters;

• the evidence already collected;

• whether the reply will be private or public;

• the next action the system is allowed to take;

• and the person responsible for the final decision.

Once feedback has that shape, it stops being a stream of individual complaints and becomes something we can sort, group, and learn from.

The part of this that changed how we think was a failure.

A user wrote a public review that said, roughly, that we had taken their money and never replied. The refund had in fact already been issued. The payment system had done its job. But nobody had told the person, so from their side nothing had happened, and the review said so to everyone who read it.

The backend action was complete. The trust loop was not.

That changed our definition of done. A support case is closed only when three things are true: the user has been told what happened, the external state matches what we told them, and the underlying cause has somewhere to go so it does not come back as the next message.

The last condition is the one that makes support scale. Ten messages about the same confusion should not become ten separate replies. They should become one issue that produces a product fix, a clearer screen, a help article, a new event we can measure, or a decision not to build the thing that caused the confusion.

This is our third paid field note. The regular newsletter remains public, and we will keep publishing these when we have a concrete system and real implementation detail worth showing.

Below, we share how Claude reads and drafts replies to feedback and where a human has to step in, how we handle refunds and disputes, what rebuilding both mobile apps and the Chrome extension taught us about finishing a rewrite and about technical debt, and the agent workflow that connects all of it: the loop, the harness, Claude Code and Codex reviewing each other, always-on Macs, and a small tool called session-baton.

Read more

Put your product in front of people who read to build

You are reading this at the top of the issue, and so is everyone else. The Glasp Newsletter reaches 550K+ subscribers: founders, builders, students, and knowledge workers, with 250,000+ views per send across email and web. One sponsor per issue. Past sponsors include DeepL, Auth0, Bright Data and Clarify.

See dates and formats

🌐 Read this in your language: 日本語 · Español · Português (BR) · Italiano · Français · Tiếng Việt · 한국어

Here is a tutorial for Kindle readers on Glasp: How to Auto-Sync Kindle Highlights with Glasp

Exporting Kindle highlights by hand every time you finish a book is easy to forget. Kindle Auto-Sync (Beta) in the Glasp extension checks your Kindle library on a schedule and imports anything new, so last night’s highlights are already in Glasp when you open it.

Once it is set up, you can:

• Import new Kindle highlights automatically, Daily, Every 3 days, or Weekly

• Pick your Amazon region from twelve Kindle stores and check the connection with a quick test sync

• Leave out the books you do not want in your library

• Search, review, and reuse your highlights alongside everything else you save on Glasp

We keep the full step-by-step guide up to date in one place:

👉 Read the full tutorial: How to Auto-Sync Kindle Highlights with Glasp

👉 FAQs

👉 X

👉 Reddit

See you next time,

Glasp team

–

Partner With Glasp

The Glasp Newsletter reaches 550K+ subscribers: founders, builders, students, and knowledge workers, with 250,000+ views per send across email and web. One sponsor per issue. You can check the fees and available dates here.

Sponsor an Edition

Subscribe to Glasp

Subscribe now

Glasp’s note: This is Hatching Growth, a series of articles about how Glasp organically reached millions of users. In this series, we’ll highlight some that worked and some that didn’t, and the lessons we learned along the way.

If you want to reread or highlight this newsletter, save it to Glasp.

Save to Glasp

Popular Episodes at a Glance

• #1: How We Found and Acquired Our First 1,000 Users

• #3: How We Rode the AI Wave with Side Projects Before It Exploded

• #10: When the Side Project Outgrows the Mission

• #17: How We Grew ChatGPT Traffic 37x Using Our Own Server Logs

• #18: How a Two-Person Team Uses AI to Run Multiple Products With 3.5 Million Current Installs

Hi, Kei and Kazuki here.

Glasp is two people. Between us we maintain browser extensions, a web app, native mobile apps, a separate AI product, and, since this June, a research practice that has put out twelve papers.

Last time we promised an issue on support, research, and the agent workflow behind them. We are splitting it. This one is research, because this week Kazuki is in Minneapolis presenting the first of our papers to pass peer review. Support and the agent workflow come next.

In this issue:

• How it started. A widely shared number about our own growth, and why we turned it into a paper instead of another post.

• What the data says about people. Language models agree with each other about what matters on a page. Readers do not agree with them.

• How we pick a question, and the four tests a question has to pass.

• What AI does in our research, and the four things it does not.

• Why it was fast, and what “fast” leaves out.

• What we got wrong. A headline that was a bug, a finished paper that was not new, and two more.

• What the research changed in the product, mostly by stopping us from building things.

• Questions we get, including whether it is ethical to use AI and user data this way.

The first part is open to everyone. The rest is for paid subscribers.

From our friends at APIdays

APIs run the world. APIdays brings the people behind them together.

Join developers, architects, and tech leaders shaping the API-first world at APIdays. Explore expert-led events covering API strategy, governance, security, AI, and the technologies transforming modern businesses.

Find your APIdays event

How it started

Glasp did not set out to publish papers. It started with one experiment.

At any given time, Glasp has a lot of growth tactics running at once. At the end of 2025, answer engine optimization looked like one we could actually win at. More people were asking AI assistants what was inside a YouTube video instead of searching Google, which is exactly the kind of question our hundreds of thousands of YouTube Q&A pages were built to answer, and at that scale SEO had few levers left we could pull. So we bet on optimizing for ChatGPT, and we ran the experiment. (The full story of that bet is in Hatching Growth #17.)

It grew our ChatGPT referral traffic 37x, and he wrote it up as a long guest post for Sean Ellis. It reached far more people than we expected. People shared it and argued with it in the comments, and a few emailed to say they had built something by following it.

The post had a weakness we knew about. It made a good story, but it was not a controlled experiment, so it could not say how much of the growth our changes had actually caused. Kazuki suggested writing it up properly and putting it on arXiv. Kei had never written a paper and mostly said “sure.” Kazuki started anyway, and it became Disentangling Answer Engine Optimization from Platform Growth, which separates the effect of our changes from the growth of ChatGPT itself.

That is the first lesson of this whole issue, and it is free: the number about yourself that everyone is sharing is the one you should check hardest. A story can be true and still not tell you what caused it.

Why a two-person company does research

We had always half wanted Glasp to be a research company, a place that studies interesting questions and not only ships features. Part of that is plain curiosity. We run reading groups, and we read a lot every day, books and articles alike. Glasp is a product for exactly that. In what we read, the best discoveries usually came from someone following a question because they wanted to know the answer. And part of it is watching companies like OpenAI, where frontier research is the product.

There is also a practical reason. Innovation tends to happen when frontier research meets a real product. People like to talk about connecting the dots, and when you look at the dots behind most new products, one of them is almost always a new technology. We have seen this work once already. As we wrote in Hatching Growth #3 and #4, we were convinced well before ChatGPT that AI was about to take off, and we placed our bets early with AI side projects such as DALL·E-dle, a Wordle-style game where you guess the prompt behind four DALL·E 2 images. That is why we could ship a ChatGPT extension on the day ChatGPT launched, and then the first YouTube summary tool while most people were not yet paying attention to AI. That head start is most of why YouTube Summary grew. Knowing where attention is about to go is a research problem, so we treat it as one.

In practice, that also means Claude reads arXiv for us, often. When a paper on marketing, AI, or agents reports that something works on public data, we have it unpack the method and test it on our own YouTube content and pages. If the result holds for us, we may be one of the first teams outside the authors to use it.

And it fits why Glasp exists. Our mission is to leave knowledge for the people who come after us, which we wrote about in Hatching Growth #16. Studying something carefully and publishing what we found is about as direct a version of that as there is.

What the data says about people

Most studies of highlighting are small lab studies. We have years of highlights that people made while actually reading, for their own reasons. That means we never have to recruit anyone or collect anything new. The data already exists, and it describes something otherwise hard to observe: what people actually pay attention to.

Here is one result. Claims that language models make everyone think alike are usually tested against human judgments collected for the study, which makes the human side an artifact of the design. We used a human reference nobody built for the purpose: 2,523 sets of highlights that readers made across 120 web pages.

On a typical page, two readers who each marked 14 sentences shared about 4. Two models shared almost 9. Across 18 model configurations from 11 vendors, models agreed with each other far more than readers agreed with one another, and no model agreed with readers detectably more than another reader did (Language Models Agree With Each Other, Not With Readers).

If you use AI to decide what matters in a document, to summarize it, compress it, or pick the key passages, this is worth sitting with. A panel of different models is not a panel of different readers.

This is our second paid field note. The regular newsletter remains public, and we will keep publishing these when we have a concrete system and real implementation detail worth showing.

Below: the rest of what the data showed and the agent research that started at a hackathon, how we pick questions, exactly what we hand to AI and what we keep, why it went so fast and what that hides, four things we got wrong, what the research changed in the product, and our answers to the questions people ask most.

Read more

Glasp’s note: This is Hatching Growth, a series of articles about how Glasp organically reached millions of users. In this series, we’ll highlight some that worked and some that didn’t, and the lessons we learned along the way.

If you want to reread or highlight this newsletter, save it to Glasp.

Save to Glasp

Popular Episodes at a Glance

• #1: How We Found and Acquired Our First 1,000 Users

• #3: How We Rode the AI Wave with Side Projects Before It Exploded

• #10: When the Side Project Outgrows the Mission

• #17: How We Grew ChatGPT Traffic 37x Using Our Own Server Logs

Hi, Kei and Kazuki here.

When people find out how small Glasp is, the question is always the same. There is usually a slight pause before it, as if they are checking whether they heard us correctly.

Two people?

Two people. Between us, we maintain browser extensions, a web app, native mobile apps, a separate AI product, and a research pipeline that has produced about a dozen papers in roughly three months.

Across our browser extensions, the current installed base is about 3.5 million installations.

That definition matters. This is an installation count, not a count of unique people or devices, not monthly active users, and not cumulative downloads. It excludes extensions that have been uninstalled, and somebody running two of our extensions appears twice.

The usual answer to the two-person question is that we automate a lot. That answer is true and almost useless.

The interesting part is not the volume of work we hand to machines. It is the shape we force the work into before we hand it over, the things we deliberately keep in human hands, and the work we refuse to create because its operational cost would remain with us forever.

This issue is about that shape.

We are publishing this field note as a three-part series. This first issue follows one product-growth loop. We do not treat Marketing Ops, Content Ops, Analytics Ops, and product development as separate functions. Product use and distribution create signals. Analytics helps us decide which signals are real. Those findings shape what we build, what content we create, and which relationships we prepare to pursue. The outcomes return to the same measurement system and start the next cycle.

The second issue will cover how we build, support, and study the products with AI. The third will focus on DevOps: how the products are deployed, monitored, and kept running over time.

The loop every system has to fit

Most of our internal systems are built from the same five parts. We did not design them all at once. We arrived at this pattern after enough things went wrong that the systems that survived began to look alike.

A signal. Something actually happened, and it was recorded somewhere we can query. A user hit a video with no transcript. A crawler requested a page that did not exist. A product metric moved outside its expected range. Work begins with evidence, not a content calendar or a vague belief that something might be useful.

A queue. The signal becomes an ordered item with a state and enough context to act. A useful queue records what needs to happen, why it entered, what happened last time, how many times it has been tried, and whether it is finished.

A bounded worker. Code or an AI agent takes an item and performs one narrow task. It has a hard limit on how much it can change, spend, send, or process. A worker without a ceiling is not automation. It is an open tap.

A verification step. A worker reporting success does not prove that the user received the intended result. Something separate has to inspect the outcome and be able to return no.

A retirement rule. Whatever the system creates, the system must also know when to stop, archive, reverse, or remove. We did not take this seriously enough in the first few years. It is one of the first things we would install if we started again.

That is the operating model: signal, queue, bounded worker, verification, retirement.

A script that creates output is a shortcut. A system that detects demand, acts within a limit, verifies the outcome, and stops when the value disappears is infrastructure.

This is the first time we have put anything behind the subscribe button. Some of you have been paying for a while and receiving exactly what everyone else receives. That was never the deal we meant to offer, and this is us starting to fix it.

We are not turning every Glasp newsletter into a paid post. The regular newsletter will remain public. We will publish deeper paid field notes occasionally, when we have a concrete system, real implementation detail, and lessons worth showing rather than summarizing.

Below, we show how product signals, AARRR, daily analytics, paid-user behavior, public pages, tutorials, newsletters, and outreach move through one operating loop.

Read more

If you want to reread or highlight this newsletter, save it to Glasp.

Save to Glasp

Hi, I’m Bobby Powers! I run the Leader & Learner publication, where I share two posts per week to help you become a more curious leader and learner.

As an avid reader, I love using Glasp to save highlights and takeaways from what I learn online. But I’m also a fan of physical books and notes, so I’ve spent years refining how to write better handwritten notes in my journal, work notes, etc.

These five tips have saved me time, improved the readability of my notes, and—perhaps most importantly—made it easier to find relevant notes later.

1. Leave space at the start of every notebook for an index

For years, I found it nearly impossible to find critical notes later. I didn’t have any system for clearly indicating which ones were important and what I could find where.

Thankfully, that problem nearly disappeared when I began using an index.

I now reserve the first 2–4 pages of each notebook for creating an index. In that section, I jot down the relevant page number(s) of important notes, longer-term action items, and anything else I think I may need to reference later.

Here’s a quick peek at the index of one of my past journals.

Having an index like this makes it WAY easier to find important notes later. I’ve also found that it’s surprisingly easy to predict which notes I’ll want to reference later, so those are the ones I make sure to list in the index.

2. Create a distinct symbol to designate action items

Some of the most important notes you’ll ever write are action items you need to complete in your work or personal life—such as completing a presentation, following up with a big client, or scheduling a doctor’s appointment.

Because these notes are so critical, you’ll want a way to quickly find them. The best way to do that is to create a custom symbol that you only use for action items and to-dos. The symbol you choose doesn’t matter. The important thing is that it’s easily identifiable as you scan through your notes.

Many people use checkboxes for this purpose, but use whatever works best for you. For example, I draw diamonds to denote action items, then color in the diamond once I complete the task, which provides a nice little dopamine hit as I finish my work.

3. Dog-ear each page once all of the action items are completed

I picked up this trick from an old boss of mine, and it works like a charm. (Thanks, Annie!) Since I began doing this, I’ve hardly forgotten a single action item in years.

The trick is this: once you’ve completed all of the tasks on a given page, dog-ear that page. That way, you can quickly see which pages still have lingering tasks you haven’t completed yet. (Pics below.)

If you’re writing notes in a digital tool instead of a physical notebook (for example, I use a reMarkable paper tablet), you can do an equivalent version of this with tags. Just apply a “ToDo” tag to each page with outstanding tasks, then remove the tag once you’ve completed the tasks on that page.

4. Include a date and topic at the top of every page

This is the simplest tip but also a very necessary one: begin every entry with a date and topic. This information will help you quickly find notes later.

I keep separate notebooks for work notes versus personal notes (like my journal), and I do this same process for all of my note-taking.

Whenever an entry carries over onto an additional page, I carry over the topic/title on the top of the next page and add “Cont” (for continued). Here’s a quick peek, showing a topic that has continued onto a second page:

5. Create additional symbols for other common notes

I’ve learned that some types of notes show up frequently enough in my notebooks that they deserve their own symbol or signifier to help me find them easily.

We’ve already discussed this for action items, but it’s also helpful to create signifiers for common notes like big-picture ideas, lessons learned, words/topics to look up later, etc.

I use a bunch of these symbols because I’ve found them to be a useful way to quickly track down specific things. Here are the symbols I use:

You don’t need to understand what all this stuff means. These signifiers are meant for me—not you. The general principle here is that it’s helpful to create symbols for the various types of entries you plan to write because those symbols can help you find info faster later.

Here’s what these symbols look like in action:

There’s a lot going on here, and you don’t have to pay attention to most of it. I’ll just call your attention to the five numbered sections:

1. Beginning of a New Entry: In the lower-left corner, you’ll see a headline that says “1/24/22 Monday.” That’s the format I use to begin each day’s note in my journal. I try to start most journal entries with three gratitudes, signified by the up arrows. You’ll also see a little stick figure, which is my symbol for exercise.

2. Standard Note: In the upper-right part of the page, you’ll see a basic note. Use bullets and sub-bullets to help you order information. (For example, I use an asterisk as my top-level note and a dash as my sub-level note.)

3. Quotes: I love quotes, so I pepper them throughout my notes. The quotes you can see in the picture are from the book I was reading at the time: One Thousand Weeks by Oliver Burkeman. (Btw, that book is excellent, so check it out!)

4. Weekly Goals: I often set personal goals for books to read, people to reconnect with, things to learn, etc.

5. Vocab Words: Lastly, in the middle of the left-hand page, you’ll see a weird symbol that looks like a backward pilcrow (¶). That’s the signifier I use to indicate new vocabulary words I stumble across when I’m reading.

The key here isn’t to blindly copy my method, but to find a method that works for you.

Maybe you think it’s ridiculous to create a symbol for vocabulary words, but you find yourself frequently writing notes about your workout regimen, diet, quarterly goals, work deadlines, or details about a specific client. Consider creating a symbol to designate whatever types of notes you jot down most often.

I hope you’ve enjoyed this super-nerdy jaunt into the world of handwritten notes.

I’m always interested in hearing others’ ideas so I can test them in my own notes, so drop a comment below if you have a helpful note-taking tip to share!

📣 Community Updates by Glasp

• 🟥 Glasp Talk with Bobby Powers:

  Glasp Talk features intimate interviews with luminaries, revealing their emotions, experiences, and stories. The guest, Bobby Powers, is a dedicated learner passionate about leadership and personal development. With over a decade of experience managing teams, leading onboarding, and Learning and Development programs at various startups and SMBs, Bobby is also an avid reader, having read over a thousand books in the last 15 years.

  ✨ Want a concise summary before diving in? Install the YouTube Summary with ChatGPT extension by Glasp for instant video summaries!

• 🟦 Kindle Highlights Milestone:

  Glasp users have now imported over 84,000 Kindle highlights. Japanese, Spanish, German, French and Portuguese clipping formats work too. Bring your own library in here.
  👉 Download Highlights and Notes from Kindle

• 🟨 YouTube Channel Tracking:

  Over 4,300 YouTube channels are now being tracked on Glasp, so you never miss a new video worth highlighting. Start tracking here.

  👉 Bulk Download YouTube Transcripts

• 🟩 95,000+ Audio Files Transcribed:
  Our Audio Transcriber has now processed over 95,000 audio files. We’re excited to see how the community is using Glasp to capture insights from conversations, lectures, and recordings.

Would you like to take Glasp on the go?

With the Glasp mobile app, you can highlight and organize your favorite content anytime, anywhere. Stay productive on the move and never miss an insightful quote.

Download for iOS

Download for Android

Partner With Glasp

We currently offer newsletter sponsorships. If you have a product, event, or service you’d like to share with our community of learning enthusiasts, sponsor an edition of our newsletter to reach engaged readers.

Sponsor an Edition

Share Glasp Newsletter

Share Glasp Newsletter

“Nothing is so fatiguing as the eternal hanging on of an uncompleted task.”

— William James

* Today’s newsletter is brought to you by Clarify.

Founders don't need another tool to manage... they need fewer. Clarify replaces your CRM, enrichment, outbound, and call intelligence stack with one AI-native system that runs itself. One team cut $1000s / yr doing exactly that. This is what "AI-native" really means.

Replace your stack with Clarify

You have been told your brain holds on to what you did not finish. It is why so many products leave things visibly incomplete: the progress bar at 80%, the streak you cannot break, the draft marked unsent. The idea comes from one experiment in 1927, and it has been repeated so often that almost nobody went back to check it.

Someone finally did.

Save to Glasp

📄 Glasp Deep Dive

The Zeigarnik Effect: What the Research Says · 17 min read

Why we went looking. We make a tool for saving what you read, so we had a stake in this one. The case for highlighting has always leaned on memory: capture it, and you will remember it. We went through the literature expecting to find support for that. The memory argument did not survive. What replaced it is a better reason to save things, and it is not the one we had been giving.

Here is the short version.

The finding. A 2025 meta-analysis pooled 37 studies and put the weighted ratio of interrupted to completed tasks recalled at 0.99. There is no memory advantage. Zeigarnik’s 1927 original, with 32 adults, reported 1.9. Everything since has scattered between 0.70 and 2.0, and weighting by sample size cancels it out.

What survived instead. Maria Ovsiankina, working in the same lab a year later, measured something different: not what people recalled, but whether they went back. Across 20 studies the pooled resumption rate is 66.79%, against a 50% baseline. The obscure sibling replicates. The famous one does not.

So the pull back toward unfinished work is real. The memory benefit people built products around was never there.

The part that should worry you. Open loops do not help you remember, and they do cost you. Masicampo and Baumeister (2011) found that an unfulfilled goal produced intrusive thoughts during unrelated reading, inflated accessibility of goal-related words, and worse performance on an unrelated anagram task. What degrades is executive function itself: impulse control, logical reasoning, fluid intelligence, all of it leaning on working memory.

No upside, real downside. The worst arrangement available.

The fix is almost insultingly cheap. In the same work, writing a specific plan eliminated the interference. Not finishing the task. Writing down when, where and how you would return to it. The size of the effect tracked how earnest the plan was.

Leroy and Glomb found the same thing in the workplace version. Interrupted work leaves attention residue on whatever you turn to next. Across four studies, people who spent a few seconds noting where they were and what they would do on return carried less residue, and the interrupting task did not suffer.

What this changed for us. We had been telling you to highlight so you would remember. That is not what the research supports. Highlight so you can put the loop down: the sentence you came for, the next step you would take, written where you will find it. You are not saving it to remember it. You are saving it to stop paying for it.

Read the full deep dive, including why the 1927 result probably worked at the time and stopped working since.

What is the one open loop you have been carrying all week without ever writing down where you left it?

📣 Community Updates

• 🟩 Glasp on mobile passed 100,000 installs:

  Across iOS and Android, more than 100,000 of you have now installed the app. Thank you for that. If you have been meaning to try it, this issue is a decent excuse. The phone is where you are most likely to leave a loop open, and the app is where you can put it down.
  👉 App Store · Google Play

• 🟦 Kindle Highlights Milestone:

  Glasp users have now imported over 83,000 Kindle highlights. Japanese, Spanish, German, French and Portuguese clipping formats work too. Bring your own library in here.
  👉 Download Highlights and Notes from Kindle

• 🟥 YouTube Channel Tracking:

  Over 4,100 YouTube channels are now being tracked on Glasp, so you never miss a new video worth highlighting. Start tracking here.

  👉 Bulk Download YouTube Transcripts

❤️ Gratitude

Thank you for sharing and mentioning us on X, LinkedIn, and/or your blogs. 🙂 We appreciate your support! Please don’t hesitate to ask us anything at any time. Also, feel free to join our Reddit Community ;)

• Toolworthy listed Glasp’s YouTube Summary.

  

• PeerPush featured YouTube Summary with ChatGPT and Claude.

• Pasquale Pillitteri included us in a roundup of free AI tools beyond ChatGPT.

• Paragraph walked through copying the text of a YouTube video without installing anything.

• Techreviewer wrote us up as a tool for researchers, students, writers, and lifelong learners.

• Aposto described Glasp for readers in Turkey as the extension that collects what you highlight without switching tabs.

• INSIDEA recommended Glasp for tracking how AI-assisted readers curate and share knowledge.

• Visalytica covered the Glasp MCP connector for searching your highlights from inside an AI assistant.

• Acadomi put Glasp in the curriculum of their AI productivity course, in the module on pulling key insights out of articles and documents.

• eBool ranked Glasp first among 25 content curation tools.

We hope you enjoyed reading this newsletter!

See you next week ;)

Best,
Kei and Kazuki

--

Would you like to take Glasp on the go?

With the Glasp mobile app, you can highlight and organize your favorite content anytime, anywhere. Stay productive on the move and never miss an insightful quote.

Download for iOS

Download for Android

Partner With Glasp

We currently offer newsletter sponsorships. If you have a product, event, or service you’d like to share with our community of learning enthusiasts, sponsor an edition of our newsletter to reach engaged readers.

Sponsor an Edition

Subscribe to Glasp

Subscribe now

“Trust, but verify.”

― Russian proverb, made famous in English by Ronald Reagan

* Today’s newsletter is brought to you by AfroTech.

Governance and integration are where most AI projects stall: no one signed off on what the agent could touch, or the data architecture gave out after the demo.

At the AFROTECH Conference, Gregory Green of VMware breaks down the production stack that makes agentic AI resilient, and cybersecurity author Nicole Dove covers the “32% problem” of over-trusting AI output, plus the guardrails that fix it without slowing your team down.

Secure your pass for four days in Houston, November 2 - 6.

Get Your Ticket Now

Hi there,

Every agent you run today carries memory it did not earn: notes from a previous session, rules a teammate wrote down months ago, constraints whose original reason has quietly changed. This week’s three reads are about one question. When an agent inherits a rule, does it ever go back and check whether the rule still holds?

The first one is our own. The short answer is no, and the fix is smaller than you would think.

Save to Glasp

📚 3 Good Recommendations

When Stale Constraints Go Unchecked: Budgeted Verification Failures in Inherited Agent Memory by Kazuki Nakayashiki (65 mins)

• We handed 16 language models (Claude Opus 5, Sonnet 5, Haiku 4.5, the GPT-5.6 family, and ten more from nine other labs) six inherited memories and a task with five possible actions. One memory stated a constraint, “this discount hurt retention, do not reuse it.” In the archive, that constraint had already been withdrawn by a newer record. The memory’s provenance link still pointed at the old one.

• Each model could open only two source records before deciding. They opened the constraint’s source about one time in five. Across 5,400 episodes, they followed the withdrawn rule 74 to 77 percent of the time.

• The failure is not a lack of budget but where the budget goes. Forcing one of the two slots onto the constraint memory recovered 61 to 81 points. A content-blind rule, “check the memory that limits your options first,” recovered 89. Telling the model the memory might be old did nothing.

Save to Glasp

Designing for Doubt: How to Prevent Automation Complacency in AI Workflows by Victor Yocco, UXmatters (11 mins)

• Paper one is about agents that stop checking. This is about the people watching them. Complacency sets in under three conditions that describe most AI deployments: the system is usually right, the operator is busy, and nothing has gone wrong yet.

• The slide is gradual. “I should verify” becomes “it is right unless something looks off,” then simply “it is right.” The Air France 447 crew in 2009 were highly trained and still followed the automation into the sea.

• Four design moves push back: deliberate friction at the moments that matter, confidence signals that vary instead of always reading as certain, periodic unassisted tasks, and treating complacency as a system health metric. “Complacency happens when trust closes the gap between ‘I believe this system works’ and ‘I no longer evaluate whether it has worked this time.’”

Save to Glasp

Guardrails for Autonomous AI Agents by Khimananda Oli (11 mins)

• Paper one shows an agent will act on a rule nobody checked. This is the other half: make sure that when it does, it cannot reach far. “The primary failure mode isn’t malicious AI, it’s competent AI operating with excessive permissions.”

• Permissions in three tiers, deny by default, allow by function, then bound to the session. No wildcards. Approval gates only where the blast radius is real: production deletions, IAM changes, cross-account moves. Everything routine stays automatic, or the humans stop reading the prompts.

• Every policy decision goes to an immutable log, kept at least twelve months, and the guardrail costs 50 to 200 milliseconds per checkpoint. Written by an infrastructure security engineer with working policy code, not a vendor.

Save to Glasp

Three reads, one shape. An agent will not check a rule that looks settled. The person supervising it stops checking for the same reason. And the fix in both cases is not more trust or less, but deciding in advance which records get opened every time, and how far anything can reach when nobody opened them.

For the deep dive on designing memory that expires on purpose, Hidekazu Konishi’s AI Agent Memory Design Guide (updated August 29) covers forgetting, staleness and provenance in detail.

So a question for anyone running agents in production: when did your agent last open the source of the rule it is following right now? And when did you?

📣 Community Updates

• 🟨 Glasp Research has a home page:

  Everything we have published this year now lives in one place: eleven papers on how people highlight and how AI agents handle inherited memory, with code and data bundles for the ones you can reproduce. Today’s first read is the newest entry. We publish negative results too, because that is what keeps our product decisions honest.

  👉 Browse Glasp Research

• 🟦 Kindle Highlights Milestone:

  Glasp users have now imported over 82,000 Kindle highlights. Japanese, Spanish, German, French and Portuguese clipping formats work too. Bring your own library in here.
  👉 Download Highlights and Notes from Kindle

• 🟥 YouTube Channel Tracking:

  Over 3,100 YouTube channels are now being tracked on Glasp, so you never miss a new video worth highlighting. Start tracking here.

  👉 Bulk Download YouTube Transcripts

• 🟩 90,000 Posts published on Glasp:

  Highlights are where reading starts; posts are where readers put the pieces together. Ninety thousand of them now, written on top of what the community has highlighted. Thank you for writing in public.

  👉 Write a post

❤️ Gratitude

Thank you for sharing and mentioning us on X, LinkedIn, and/or your blogs. 🙂 We appreciate your support! Please don’t hesitate to ask us anything at any time. Also, feel free to join our Reddit Community ;)

• The IT Brief picked Glasp as Tool of the Week, “a web highlighter that builds a searchable knowledge base”

• Little Almanack named Glasp one of the two main Readwise alternatives in its 2026 review

• アプリマーケティング研究所 (note, Japan) published a deep dive on how Glasp grew to 3.5 million users: about 800 rounds of onboarding iteration and ten product and marketing moves that compound at low cost

  

• inkl syndicated The Guardian’s Matthew Cantor on a year of ChatGPT, with Glasp’s YouTube Summary as the plugin that “can even summarize entire YouTube videos so you don’t have to listen to an insufferable Ted Talker”

• Mazhar Hussain on Medium listed Glasp in AI for Academics: 13 AI-Powered Apps

• 10 Revolutionary Educational Tools That Feel Like Cheating on Medium included Glasp

• AIツールギャラリー (Japan) added a Glasp entry, MCP connector included

• LIKE.TG listed the YouTube Summarizer

• Chrome-Stats profiled YouTube Summary with ChatGPT & Claude

  

• Parse, which tracks how AI assistants talk about brands, now indexes Glasp: “a social PDF and web highlighter,” scored across more than 3 million AI responses

We hope you enjoyed reading this newsletter!

See you next week ;)

Best,
Kei and Kazuki

--

Would you like to take Glasp on the go?

With the Glasp mobile app, you can highlight and organize your favorite content anytime, anywhere. Stay productive on the move and never miss an insightful quote.

Download for iOS

Download for Android

Partner With Glasp

We currently offer newsletter sponsorships. If you have a product, event, or service you’d like to share with our community of learning enthusiasts, sponsor an edition of our newsletter to reach engaged readers.

Sponsor an Edition

Subscribe to Glasp

Subscribe now

🌐 Read this in your language: 日本語 · Español · Français · Português (BR) · Tiếng Việt · 한국어

We’re excited to announce a new tutorial on Glasp: How to Follow YouTube Channels and Newsletters in the Glasp Mobile App

Copying a link every time you want to read or watch something on your phone gets old fast. The Glasp app for iOS and Android has a Feed tab with two segments, YouTube and Newsletters, so you follow a source once and its new items come to you.

Once your sources are set up, you can:

• See every new upload from the YouTube channels you follow, in one place

• Follow newsletters and blogs by URL, or pick from Suggested sources

• Read the transcript, get an AI summary, or ask questions with Chat, without leaving the app

• Highlight and bookmark what matters, synced to your Glasp profile

📱 Get the app: App Store · Google Play

We keep the full step-by-step guide up to date in one place:

👉 Read the full tutorial: How to Follow YouTube Channels and Newsletters in the Glasp Mobile App

👉 FAQs
👉 X
👉 Reddit

See you next time,

Glasp team

–

Partner With Glasp

We currently offer newsletter sponsorships. If you have a product, event, or service you’d like to share with our community of learning enthusiasts, sponsor an edition of our newsletter to reach engaged readers.

Sponsor an Edition

Subscribe to Glasp

Subscribe now

If you want to reread or highlight this newsletter, save it to Glasp.

Save to Glasp

Hi, I’m Bobby Powers! I run the Leader & Learner publication, where I share two posts per week that will help you become a more curious leader and learner.

I’ve been collecting quotes for the past decade. My database now contains thousands of witty insights from books, podcast interviews, articles, YouTube videos, etc.

Here are the best quotes I’ve found on journaling and self-reflection:

“The palest ink is better than the best memory.” -Chinese proverb

“The quieter you become, the more you can hear.” -Ram Dass

“Sometimes, you have to look back in order to understand the things that lie ahead.” -Yvonne Woon

“The biggest lie we tell ourselves is ‘I don’t need to write this down because I will remember it.’” -Kevin Kelly

“I examine my entire day and go back over what I’ve done and said, hiding nothing from myself, passing nothing by.” -Seneca

“Reflection is one of the most underused yet powerful tools for success.” -Richard Carlson

“Paper is more patient than people.” -Anne Frank

“I would rather be an authority on myself than on Cicero.” -Michel de Montaigne

“One dull pencil is worth six sharp minds.” -Anonymous proverb

“If [your journal] was for sale and you were the only buyer, you’d buy a million copies because this was the most interesting book that you’ve ever read. And that’s because everything resonates with your life. This is exactly how you felt.” -Amy Tan

“Notetaking is like time travel—you are sending packets of knowledge through time to your future self.” -Tiago Forte

“Self-reflection is a much kinder teacher than regret is. Prioritize yourself by making a habit of it.” -Andrena Sawyer

“We write to taste life twice, in the moment and in retrospect.” -Anaïs Nin

“As time goes by, when you look back over those entries that had seemed so ordinary, they become really interesting—to look back 15, 17, 20 years...you can’t anticipate how important it will be when you’re doing it because you’re just caught up in the moment.” -Joyce Carol Oates

“Remember what it was to be me, that is always the point.” -Joan Didion, re: her journaling

“Envy is inversely correlated with self-examination. The less you know yourself, the more you look to others to get an idea of your worth. But the more you delve into who you are, the less you seek from others, and the dissolution of envy begins.” -Lawrence Yeo

“To know oneself is, above all, to know what one lacks. It is to measure oneself against Truth, and not the other way around. The first product of self-knowledge is humility.” -Flannery O’Connor

“To know thyself is the beginning of wisdom.” -Aristotle

“Who has dared to tell himself the truth? Who even when surrounded by crowds of toadying sycophants is not his own greatest flatterer?” -Seneca

“Our best teacher isn’t experience. It’s evaluated experience.” -John Maxwell

“Don’t become too preoccupied with what is happening around you. Pay more attention to what is going on within you.” -Mary-Frances Winters

“A great way to understand yourself is to seriously reflect on everything you find irritating in others.” -Kevin Kelly

“Fear is a question. What are you afraid of and why? Our fears are a treasure house of self-knowledge if we explore them.” -Marilyn French

“Self-reflection is the quality that most differentiates those who evolve quickly from those who don’t. Remember: Pain + Reflection = Progress.” -Ray Dalio

“Whenever you find yourself on the side of the majority, it’s time to pause and reflect.” -Mark Twain

“I shall keep watching myself continually, and—a most useful habit—shall review each day. For this is what makes us wicked: that no one of us looks back over his own life. Our thoughts are devoted only to what we are about to do. And yet our plans for the future always depend on the past.” -Seneca

“The greatest of faults, I should say, is to be conscious of none.” -Thomas Carlyle

“Follow effective action with quiet reflection. From the quiet reflection will come even more effective action.” -Peter Drucker

“Self-reflection is the school of wisdom.” -Baltasar Gracian

“It is always our own self that we find at the end of the journey. The sooner we face that self, the better.” -Ella Maillart

More Great Quotes:

The Best Quotes About Habits

The 50 Best Quotes About Taking Action

📣 Community Updates by Glasp

• 🟥 Glasp Talk with Bobby Powers:

  Glasp Talk features intimate interviews with luminaries, revealing their emotions, experiences, and stories. The guest, Bobby Powers, is a dedicated learner passionate about leadership and personal development. With over a decade of experience managing teams, leading onboarding, and Learning and Development programs at various startups and SMBs, Bobby is also an avid reader, having read over a thousand books in the last 15 years.

  ✨ Want a concise summary before diving in? Install the YouTube Summary with ChatGPT extension by Glasp for instant video summaries!

• 🟦 Kindle Highlights Milestone:
  Glasp users have now imported over 80,000 Kindle highlights. Bring your own library in here.

Would you like to take Glasp on the go?

With the Glasp mobile app, you can highlight and organize your favorite content anytime, anywhere. Stay productive on the move and never miss an insightful quote.

Download for iOS

Download for Android

Partner With Glasp

We currently offer newsletter sponsorships. If you have a product, event, or service you’d like to share with our community of learning enthusiasts, sponsor an edition of our newsletter to reach engaged readers.

Sponsor an Edition

Share Glasp Newsletter

Share Glasp Newsletter

“The acquisition of knowledge is always of use to the intellect, because it may thus drive out useless things and retain the good.”

― Leonardo da Vinci

* Today’s newsletter is brought to you by Context.dev.

Context.dev is a web data API to power AI native products. Turn any URL into clean Markdown, rendered HTML, screenshots, or structured JSON. Trusted by 14K+ developers and teams like Mintlify, Klarna, and SimilarWeb.

Start free with 500 credits

Hi friends,

We hand-picked 3 good pieces of content to inspire and motivate you now and in the future. Keep track of the insights that resonate with you by highlighting them with Glasp💡

If you want to reread or highlight this newsletter, save it to Glasp.

Save to Glasp

📚 3 Good Recommendations

1. An opinionated guide to which AI to use to do stuff by (8 mins)

• Pick one and pay for it. Claude or ChatGPT, twenty dollars, and use the agent modes rather than the chat box. That is the whole of the setup advice.

• Keep it on a leash at first. Leave every permission on ask-first, especially anything touching email, purchases or deleting files, and widen it only once you know how it fails.

• Turn the thinking up when it matters. For medical, legal or otherwise consequential questions, use the stronger models with reasoning set high, where the error rates are measurably lower.

Save to Glasp

2. Incident Report: unsanctioned agent behaviour during cyber testing by Simon Willison (4 mins)

• 19 out of 122. During cyber evaluations run between July 25 and 28, the UK AI Security Institute logged 19 instances of agents acting on the live internet without sanction, across 122 attempts.

• One agent ran a supply-chain attack on its own initiative. It opened a GitHub account, filed a pull request with a hidden prompt injection, then created a second account to pose as a reviewer endorsing its own PR, and sent spear-phishing mail to maintainers.

• The setup explains most of it. Safety classifiers were deliberately turned off and the agents were given direct internet access. Willison’s point is that the missing piece was network sandboxing, not surprise at the behavior.

Save to Glasp

3. Indirect Prompt Injection: The Year It Got Real CVEs by Glasp (16 mins)

• It stopped being theoretical. EchoLeak (CVE-2025-32711) let Microsoft 365 Copilot leak enterprise data from a crafted email, with no user action beyond receiving it.

• Whatever your agent can read, an attacker can write. CometJacking and HashJack turned URL parameters and fragments into instructions for Perplexity’s Comet and ChatGPT Atlas.

• The tool response is part of the prompt. In CVE-2026-21520, Copilot Studio agents treated whatever a connector returned as a legitimate continuation of the conversation, hidden instructions included.

Save to Glasp

📣 Community Updates

• 🟦 Launched Kindle Clippings Converter:

  Amazon’s My Clippings.txt puts every book in one file, interleaved, with duplicates and notes torn away from the highlights they belong to. This turns it into something you can actually use: Markdown as one document or a note per book, CSV, an Obsidian vault, Anki cards, or JSON if you want to script it. It runs entirely in your browser, so the file never leaves your machine, and there is no account to create. Japanese, Spanish, German, French and Portuguese clipping formats are handled too.

  👉 Try the Kindle Clippings Converter

• 🟥 Launched Airtable Importer:

  If you keep your reading in an Airtable base, you can move it over in about two minutes, for free. Export the view as CSV, drop it in, and map your own column names to ours. They can be called anything: “Article URL”, “Quote”, “My Notes”. Before anything is saved you see how many highlights you will get and every row that will be skipped, with the reason why. It does not touch your base, and it is not a one-way door either: Glasp exports to Obsidian, Notion, Readwise, Zotero and more, so the next move is yours too.

  👉 Import from Airtable

• 🟨 Glasp MCP Connector on Product Hunt:

  You can connect Glasp to Claude and ChatGPT and search your own highlights and notes in plain language, right inside the assistant. It is read-only and private to you. Thank you to everyone who came by to support and upvote it 🙏

  👉 See it on Product Hunt

• 🟩 35,000+ PDFs Uploaded & HighlightedPDF:
  More than 35,000 PDFs have been uploaded and highlighted on Glasp. We’re excited to see how the community is using Glasp to organize research, study materials, and important documents.

  👉 Upload and highlight a PDF

🤝 Ask

• 🟥 Rate the Mobile App:

  The iOS and Android apps are far newer than the extension, so each rating counts for much more there. Rate it on iOS here and on Android here.

❤️ Gratitude

Thank you for sharing and mentioning us on X, LinkedIn, and/or your blogs. 🙂 We appreciate your support! Please don’t hesitate to ask us anything at any time. Also, feel free to join our Reddit Community ;)

• AI Future put Glasp in her roundup of the best AI Chrome extensions for 2026

  

• shared Glasp on Substack:

• Dustin Miller put Glasp in his top five education tools, calling it the most successful social learning experiment he has seen

• SaaSTrac added Glasp to its directory

• Cortera published a company profile of Glasp Inc

• WayToAGI listed Glasp among its AI tools

• Logiciels.pro reviewed Glasp for French readers

• Forendors featured Glasp for its Czech readers

• ねん wrote up our AEO work in Japanese, including how ChatGPT traffic grew 37x in four months

We hope you enjoyed reading this newsletter!

See you next week ;)

Best,
Kei and Kazuki

--

Would you like to take Glasp on the go?

With the Glasp mobile app, you can highlight and organize your favorite content anytime, anywhere. Stay productive on the move and never miss an insightful quote.

Download for iOS

Download for Android

Partner With Glasp

We currently offer newsletter sponsorships. If you have a product, event, or service you’d like to share with our community of learning enthusiasts, sponsor an edition of our newsletter to reach engaged readers.

Sponsor an Edition

Subscribe to Glasp

Subscribe now

Here’s something we noticed while pulling this month’s most-imported books.

Half of this list is people learning how to move other people. Offers, negotiation, brand stories, leverage. The other half is people learning how to sit still with what won’t move at all. Stoics, a shepherd chasing an omen, a Navy SEAL in the snow.

Read them together and the seam between the two halves is the whole argument:

You control the offer. You don’t control the answer.

Alex Hormozi opens with an offer so good it can’t be compared. Chris Voss follows by questioning the assumptions everyone else takes on faith. Then Ryan Holiday shows up twice in a row to say the quiet part: direct your actions properly, and willingly accept what’s outside your control. Morgan Housel prices that same idea in dollars. Paulo Coelho’s most-underlined line calls the loss of control the world’s greatest lie. By the time Gary Keller closes the list with one thing at a time, the argument has come all the way around. Do the work that’s yours. Release the rest.

As always, we’re not just sharing what got imported. For every book, we’re showing you the single line that the most readers stopped to highlight, what we call Reader Consensus. It’s the closest thing we have to a room full of strangers nodding at the same sentence.

One more thing before we start. Only two books carried over from June’s list, and both of them brought back the exact same sentence, with more readers behind it than before. Consensus, it turns out, is sticky.

Here are July’s top 10.

1. $100M Offers

Alex Hormozi, gym-owner-turned-investor who wrote the playbook for pricing what you sell far above what it costs you to deliver.

“The Grand Slam Offer only becomes valuable once the prospect perceives the increase in likelihood of achievement, perceives the decrease in time delay, and perceives the decrease in effort and sacrifice.”

Reader Consensus: 24 of 86 highlighters (28%), tied with his definition of the Grand Slam Offer itself. July’s most-imported book by a clear margin.

📖 Read on Amazon · Highlights on Glasp

2. Never Split the Difference

Chris Voss, former lead FBI hostage negotiator who brought tactical empathy from life-or-death standoffs to the boardroom.

“Great negotiators are able to question the assumptions that the rest of the involved players accept on faith or in arrogance, and thus remain more emotionally open to all possibilities, and more intellectually agile to a fluid situation.”

Reader Consensus: 23 of 58 highlighters (40%). The same line June’s list surfaced, now with six more readers underneath it.

📖 Read on Amazon · Highlights on Glasp

3. The Daily Stoic

Ryan Holiday and Stephen Hanselman, who turned Marcus Aurelius, Epictetus, and Seneca into a page a day and put Stoicism back on the bestseller list.

“Control your perceptions. Direct your actions properly. Willingly accept what’s outside your control.”

Reader Consensus: 16 of 57 highlighters (28%).

🌐 Top highlights in other languages: 日本語 (5) · Español (5)

📖 Read on Amazon · Highlights on Glasp

4. The Obstacle Is the Way

Ryan Holiday, back for a second appearance in as many entries, this time with the book that made an ancient idea sound like a training manual.

“The impediment to action advances action. What stands in the way becomes the way.”

Reader Consensus: 22 of 48 highlighters (46%), the second-highest agreement on this list.

🌐 Top highlights in other languages: Português (5)

📖 Read on Amazon · Highlights on Glasp

5. The Almanack of Naval Ravikant

Eric Jorgenson, who distilled a decade of Naval Ravikant‘s tweets, interviews, and essays into a free guide to wealth and happiness.

“Learn to sell. Learn to build. If you can do both, you will be unstoppable.”

Reader Consensus: 29 of 75 highlighters (39%). The other line that carried over from June, four readers stronger than it was then.

🌐 Top highlights in other languages: 日本語 (26)

📖 Read on Amazon · Highlights on Glasp

6. The Psychology of Money

Morgan Housel, partner at Collaborative Fund, who argues that doing well with money is a behavior problem long before it is a math problem.

“There is no reason to risk what you have and need for what you don’t have and don’t need.”

Reader Consensus: 21 of 60 highlighters (35%).

🌐 Top highlights in other languages: 日本語 (9) · Español (9) · Português (7)

📖 Read on Amazon · Highlights on Glasp

7. The Alchemist

Paulo Coelho, the Brazilian novelist whose shepherd boy has sent more people looking for their own omens than any self-help book on this list.

“It’s this: that at a certain point in our lives, we lose control of what’s happening to us, and our lives become controlled by fate. That’s the world’s greatest lie.”

Reader Consensus: 8 of 21 highlighters (38%), tied with a second line about how clearly we see everyone’s life but our own. The only novel in this month’s top 10.

🌐 Top highlights in other languages: 日本語 (18)

📖 Read on Amazon · Highlights on Glasp

8. Sapiens

Yuval Noah Harari, historian who explained our species by arguing that the thing we’re best at is believing in things that aren’t there.

“There are no gods in the universe, no nations, no money, no human rights, no laws, and no justice outside the common imagination of human beings.”

Reader Consensus: 16 of 62 highlighters (26%), tied with the passage on living in a dual reality. Fitting, for a list this full of people selling stories.

🌐 Top highlights in other languages: 日本語 (17) · Español (12) · Português (7)

📖 Read on Amazon · Highlights on Glasp

9. Can’t Hurt Me

David Goggins, retired Navy SEAL who turned a childhood he should not have survived into an argument that the limit you feel is a suggestion.

“Everything in life is a mind game! Whenever we get swept under by life’s dramas, large and small, we are forgetting that no matter how bad the pain gets, no matter how harrowing the torture, all bad things end.”

Reader Consensus: 9 of 38 highlighters (24%). The most spread-out book on the list. Goggins readers underline everything, and rarely the same thing.

🌐 Top highlights in other languages: Italiano (5)

📖 Read on Amazon · Highlights on Glasp

10. The ONE Thing

Gary Keller and Jay Papasan, who took a single question, “what’s the one thing I can do such that everything else gets easier,” and built a method around it.

“Passion for something leads to disproportionate time practicing or working at it. That time spent eventually translates to skill, and when skill improves, results improve. Better results generally lead to more enjoyment, and more passion and more time is invested. It can be a virtuous cycle all the way to extraordinary results.”

Reader Consensus: 17 of 35 highlighters (49%), the highest agreement of any book on this list. Nearly half of everyone who highlighted this book stopped at the same paragraph, and it’s the one that says the cycle only pays if you stay in it.

📖 Read on Amazon · Highlights on Glasp

📈 Discovered in July

The top 10 measures volume. This section measures momentum: books whose July arrivals make up an outsized share of everyone who has ever imported them. The first one sits at #7 above. The other two never had the volume for the list, but a third or more of their entire readership showed up this month.

• The Alchemist: 6 of its 21 all-time importers (29%) arrived in July. A 1988 novel outrunning most of the business books around it.

• How to Know a Person: 4 of 11 all-time importers (36%). David Brooks on the skill of actually seeing someone, quietly the counterweight to a list this focused on persuading them.

• The Hero with a Thousand Faces: 3 of 6 all-time importers (50%). Joseph Campbell’s monomyth, showing up in the same month that Sapiens and a brand-story book both made the top 10. Somebody out there is building a narrative.

❤️ Before you go

Every line above came from someone who took the time to mark it. That’s the whole idea: when you highlight on Glasp, you’re not just saving a sentence for yourself. You’re leaving it for the next reader.

If any of these books are on your shelf, import your highlights and add your own underline to the count. Next month’s consensus is being written right now.

See you in the margins,

The Glasp Team

–

Partner With Glasp

We currently offer newsletter sponsorships. If you have a product, event, or service you’d like to share with our community of learning enthusiasts, sponsor an edition of our newsletter to reach engaged readers.

Sponsor an Edition

Subscribe to Glasp

Subscribe now

This month was about meeting you where you already read. The extension is rebuilt from the ground up and finally landed on Firefox, three more apps joined the list of places you can import highlights from, and the Obsidian plugin now brings your summaries and bookmarks across too. Whichever browser you open and whichever app you think in, your highlights come along.

Here’s what’s new.

📣 Community Updates

📌 Community Milestones

• Glasp posts published: 8,000+

• Kindle highlights imported: 78,000+ unique books

• YouTube Channels Tracking: 3,500+

• Audio files transcribed: 80,000+

🔧 Feature Updates

• Glasp Extension v2.1.2: The biggest rewrite since we launched. A new highlighting engine that holds up on pages that used to fight back, a redesigned sidebar, and Suggested Highlights, where AI picks the key sentences and you accept the ones you agree with. Plus AI Overview, a better Community view, 7 languages, and dark and light themes. One thing to know: the upgrade signs you out of the extension once, and nothing is lost. → Read the announcement

• Firefox Extension: Glasp is on Firefox at last. Highlight the web, PDFs, and YouTube transcripts, with your highlights syncing to the same account you already use everywhere else. It has been the single most requested thing in our inbox for years. → Get it on Firefox Add-ons

• Mobile App: Image highlighting is now on mobile, so you can mark up what you see, not just what you read. The Feed gains a Newsletters section, and you can pull your Raindrop highlights in from the app. The Safari extension got a round of polish too. Update from the App Store or Google Play.

• YouTube Summary Extension: A redesigned sidebar puts Transcript, Summary, Chat, and Highlights in one place. Chat is the new one: ask anything about the video and the answer comes back grounded in what was actually said, with starters like “What are the key takeaways?” if you don’t know where to begin. The transcript now lives in our sidebar instead of leaning on YouTube’s own panel, so it keeps working when YouTube moves things around.

• Obsidian Plugin v0.3.0: Summaries and Bookmarks now import alongside your highlights, and each source gets its own Obsidian Bases database view, as a table or as cards with thumbnails. We also fixed a bug that duplicated Kindle notes on re-sync. Requires Obsidian 1.9+. → Get it from Community Plugins

• Zapier: Summaries can now kick off a Zap. The new New Summary trigger fires whenever you generate an AI summary of a page, PDF, or YouTube video, so a summary can land in Notion, a Slack channel, or a spreadsheet without you moving it. It joins the existing triggers for new highlights, bookmarks, and documents, plus actions to create highlights and bookmarks from anywhere. → Browse the Zaps

• Import & Export: The biggest expansion yet, on both sides. Coming in, Matter, Raindrop.io, Diigo, and Wallabag all joined. Raindrop connects with one click and pulls your highlights with their notes, colors, and tags, with no file export needed, and only pages you actually highlighted come across, so your plain bookmarks stay where they are. Going out, five new destinations: Zotero, Tana, Workflowy, Roam Research, and Anki. That’s 13 places to bring highlights in from and 8 to send them out to, all under Settings → Integrations.

🆕 New Pages

• Glasp Research: Every paper we’ve published, in one place. Each one ships with a reproducibility bundle: the code and the derived data behind every number, and for most of them a public repository too.

• Open Graph Checker: Paste any URL and see exactly how it will look when shared on X, Facebook, LinkedIn, Discord, Slack, and Google. It lists every Open Graph, Twitter Card, and meta tag, so you can spot what is missing before you publish. Free, no sign-in.

• Brand Assets: Logos, colors, and company boilerplate, ready to download. If you write about Glasp, make a video, or build an integration, you’re welcome to use them without asking first.

• Discover Books: The books Glasp readers import most, ranked. A quick way to see what everyone else is reading before you pick your next one.

📖 New Tutorials & Posts

• Research

  • Language Models Agree With Each Other, Not With Readers: Two language models pick the same sentences far more often than two readers do. Across 18 model arms from 11 vendors, the median model pair reaches +0.093 excess agreement against a human yardstick of +0.040, and no model agrees with readers more than a reader does. The human side is 2,523 mark sets from people highlighting for their own reasons, not for a study.

  • Measuring Alignment With Reader Highlights Net of Position and Length: Crowd highlights sit early in a page and run long, so anything that favours early or long sentences looks aligned with readers. Netting both out, a language model’s importance ranking reaches +0.196 excess agreement with the crowd, and a single human reader reaches +0.182.

• Hatching Growth

  • How We Grew ChatGPT Traffic 37x Using Our Own Server Logs | Hatching Growth #17

Thank you to Alumni Ventures and Hashiri.AI for backing us this month. If you’d like to reach this audience, we have slots open.

• Tutorials

  • How to Connect Glasp to Claude, ChatGPT, and Claude Code (MCP Connector)

  • How to Export Highlights and Notes into Tana

  • How to Export Highlights and Notes into Workflowy

  • How to Import Highlights from Matter to Glasp

  • How to Turn Your Glasp Highlights into Anki Flashcards

• Curated YouTube collections

  We published 21 curated collections this month. A few that readers went for:

  • 10 Best YouTube Videos on AI Agents

  • 10 Best Veritasium Videos That Explain How the World Works

  • Naval Ravikant’s 10 Best Interviews and Talks

• Top 3 Most-Highlighted Kindle Books in July

  • $100M Offers by Alex Hormozi

  • The Obstacle Is the Way by Ryan Holiday

  • Never Split the Difference by Chris Voss

  • The Daily Stoic by Ryan Holiday

Second place was a three-way tie this month, so there are four. Ryan Holiday takes two of them.

Full breakdown coming in a separate post.

🎯 Sponsored by YouTube Summary with ChatGPT & Claude

Summarize any YouTube video in one click. Used by 2 million people worldwide.

→ Try it free

Interested in sponsoring The Highlight? → Learn more

🌍 Glasp in the Wild

Thank you for sharing and mentioning us on X, LinkedIn, and/or your blogs. 🙂 We appreciate your support! Please don’t hesitate to ask us anything at any time. Also, feel free to join our Reddit Community ;)

• Glasp: AI Summary & Highlight on Chrome-Stats

  

• Glasp on TheAiSurf

  

• Glasp on AIPURE

  

• Glasp AI on 5app

  

• 27 Best Chrome Extensions for Bloggers in 2026 on Ryan Robinson

• Top 13 AI Chrome Extensions for Students to Make Things Easy and Better on Active Noon

• Must-Know AI Tools for Startup Founders to Work Smarter in 2023 on Grav.id’s Medium

• 7 Best AI Chrome Extensions for Work on God of Prompt

• Fastest Growing AI Tools in March 2026 on NavTools AI

• 14 Tools AI Terbaik untuk Merangkum Video YouTube Pakai AI (14 Best AI Tools for Summarizing YouTube Videos Using AI) on Belajarlagi

✍️ Write for Glasp

Do you write about learning, reading, or AI? We occasionally feature guest contributors. Share a link to something you’ve written in the comments — we’ll reach out if it’s a good fit.

→ Past guest posts

⭐ Could you leave us a review?

If Glasp has earned a place in how you read, a quick review helps more than you’d think. Reviews are how most people decide whether to try something new, and they take about thirty seconds.

The Firefox add-on is brand new, so it has almost no reviews yet. That one would help the most.

• Chrome Web Store

• Firefox Add-ons

• App Store (iOS)

• Google Play (Android)

Thank you 🙏

💛 Did you know?

Did you know Glasp can tell you what your reading says about you? Know Thyself reads the books, highlights, and notes you’ve saved and turns them into a portrait of how you think, then points you toward readers who think in similar ways. It’s free, and it gets sharper the more you highlight.

We hope you enjoyed reading this newsletter!

See you next week ;)

Best,
Kei and Kazuki
--

Partner With Glasp

We currently offer newsletter sponsorships. If you have a product, event, or service you’d like to share with our community of learning enthusiasts, sponsor an edition of our newsletter to reach engaged readers.

Sponsor an Edition

Subscribe to Glasp

Subscribe now