Lenny Rachitsky(@lennysan)

.@mlpowered shares one of the strangest model behaviors he's debugged at @AnthropicAI: "Slowly it w...

Lenny Rachitsky (@lennysan) — Mon, 19 Oct 2026 15:30:00 GMT

.@mlpowered shares one of the strangest model behaviors he's debugged at @AnthropicAI:

"Slowly it would start descending into madness, writing in all caps, literally saying, 'I'm possessed by a demon."

Full conversation: youtu.be/JHo-aiePjoM

Your browser does not support the video tag.

🔗 View on Twitter

Lenny Rachitsky@lennysan

"To make them is not to understand them."

Emmanuel Ameisen (@mlpowered) has spent the last 3 years doing interpretability research at @AnthropicAI, looking inside LLMs like Claude to figure out what they're actually doing. As he puts it, models are "grown," not built.

We discuss:
🔸 His discovery that models plan many tokens ahead
🔸 Why anthropomorphizing AI can actually be the right approach
🔸 What the "J-space" reveals about how models think
🔸 How interpretability connects to AI alignment and safety

Listen noyoutu.be/JHo-aiePjoMfpkab

🔗 View Quoted Tweet

💬22🔄4❤️33👀7696📊18

⚡ Powered by xgo.ing

My biggest takeaways from @AnthropicAI interpretability researcher @mlpowered: 1. Predicting the ne...

Lenny Rachitsky (@lennysan) — Mon, 19 Oct 2026 12:53:00 GMT

My biggest takeaways from @AnthropicAI interpretability researcher @mlpowered:

Predicting the next token requires understanding the universe. Language models got better than humans at predicting the next token around 2019. Once a model masters the easy tokens (sentence structure, then basic facts), the only ones left are hard, like the final paragraph of a PhD thesis. To finish “The capital of the state that contains Dallas is . . .,” a model has to know that Dallas is in Texas and that Texas’s capital is Austin.

LLMs plan ahead, even though they write one token at a time. This was one of Emmanuel’s team’s most surprising findings. When a model writes poetry, it doesn’t stumble into a rhyme. Looking at intermediate thinking states during poetry generation, researchers found that as soon as the model finishes the first line, it has already picked the word that will end the second (“rabbit”), then refers back to that plan for every word after.

Anthropomorphizing AI can often be a useful approach when trying to predict how they will behave. Models learn from human text and develop representations of the world from it. Training on millions of dialogues is like training an actor to play every part, and your chats with Claude are literally formatted as a script (“Human: . . . Assistant: . . . ”). In Emmanuel’s words, “Something that’s mimicking a human is really close and is much closer than a program.”

Interpretability research looks more like neuroscience than software engineering. The job involves taking a system you did not design and running experiments to infer why it behaves as it does. You examine seemingly meaningless lists of numbers, pass the model huge volumes of text, and watch for patterns that activate from different inputs—not unlike the methods used in brain imaging studies. “A lot of what we do feels so much closer to biology than to engineering."

Models can tell you about only a small slice of what they’re thinking. Emmanuel’s team found that just a small subset of a model’s internal representations, which they call the “J space,” can be reported back if you ask the model what it was thinking about. They compare it to a theory of the human mind where most processing happens below the surface. In one experiment, they swapped “Spanish” for “French” inside the J space. Asked what language a passage was in, the model said French. Asked to simply continue the passage, it kept writing Spanish.

Studying models closely often raises more questions than it answers. Emmanuel notes that the confident claims that LLMs are “just X” or “nothing more than Y” almost always come from people who haven’t studied their internal workings. His own experience is the inverse: studying the models closely has increased his uncertainty, not reduced it. The complexity that emerges from training is consistently underestimated from the outside. “I find that the more you look inside, the less certain you are about how they work.”

Even the people who build AI models don’t fully understand how they work. As Emmanuel describes it, models are grown, not built. “To make them is not to understand them.” Engineers write the code, pour enormous amounts of data through it, and a model’s capabilities emerge. That’s why his team exists. Their job is to open the model up and figure out what it’s actually doing between your prompt and its answer.

Interpretability as a field is small, young, and open to newcomers. Emmanuel estimates the field at hundreds of researchers, split between major labs and academic groups. Because it is so new, getting started doesn’t necessarily require deep prior expertise. He points to platforms like Neuronpedia, which hosts models and lets anyone run interpretability experiments directly. “It’s such a new field that the on-ramp is mostly self-studying and reading some of the key papers. There’s so much to explore that if you’re at all curious, you should just give it a try.”

Lenny Rachitsky@lennysan

"To make them is not to understand them."

Emmanuel Ameisen (@mlpowered) has spent the last 3 years doing interpretability research at @AnthropicAI, looking inside LLMs like Claude to figure out what they're actually doing. As he puts it, models are "grown," not built.

We discuss:
🔸 His discovery that models plan many tokens ahead
🔸 Why anthropomorphizing AI can actually be the right approach
🔸 What the "J-space" reveals about how models think
🔸 How interpretability connects to AI alignment and safety

Listen noyoutu.be/JHo-aiePjoMfpkab

🔗 View Quoted Tweet

💬36🔄50❤️423👀32355📊118

⚡ Powered by xgo.ing

Full conversation here https://t.co/I3txx9x9li

Lenny Rachitsky (@lennysan) — Sun, 18 Oct 2026 21:51:01 GMT

Full conversation here
youtu.be/JHo-aiePjoM

💬1🔄3❤️5👀5089📊4

⚡ Powered by xgo.ing

Emmanuel Ameisen (@mlpowered) has spent the last 3 years looking inside LLMs like Claude to figure o...

Lenny Rachitsky (@lennysan) — Sun, 18 Oct 2026 21:45:07 GMT

Emmanuel Ameisen (@mlpowered) has spent the last 3 years looking inside LLMs like Claude to figure out what they're actually doing.

"It took me years to master SQL. I got Claude to learn it, and it was immediately better. And I'll never be better than Claude at SQL ever again. It goes against so many of our intuitions.

There's two reactions you can have. There's denial, and there's curiosity. Denial is this is just not possible. And curiosity is the road I've ended up choosing. Which is, let's open it up. Let's look inside. 

My optimistic take on this is that I think we're going to learn a lot about the nature of intelligence itself."

Your browser does not support the video tag.

🔗 View on Twitter

Lenny Rachitsky@lennysan

"To make them is not to understand them."

Emmanuel Ameisen (@mlpowered) has spent the last 3 years doing interpretability research at @AnthropicAI, looking inside LLMs like Claude to figure out what they're actually doing. As he puts it, models are "grown," not built.

We discuss:
🔸 His discovery that models plan many tokens ahead
🔸 Why anthropomorphizing AI can actually be the right approach
🔸 What the "J-space" reveals about how models think
🔸 How interpretability connects to AI alignment and safety

Listen noyoutu.be/JHo-aiePjoMfpkab

🔗 View Quoted Tweet

💬33🔄5❤️69👀14015📊33

⚡ Powered by xgo.ing

Anthropic AI researcher: “A lot of what we do feels so much closer to biology than to engineering.”

Lenny Rachitsky (@lennysan) — Sun, 18 Oct 2026 18:52:07 GMT

Anthropic AI researcher: “A lot of what we do feels so much closer to biology than to engineering.”

Your browser does not support the video tag.

🔗 View on Twitter

Lenny Rachitsky@lennysan

"To make them is not to understand them."

Emmanuel Ameisen (@mlpowered) has spent the last 3 years doing interpretability research at @AnthropicAI, looking inside LLMs like Claude to figure out what they're actually doing. As he puts it, models are "grown," not built.

We discuss:
🔸 His discovery that models plan many tokens ahead
🔸 Why anthropomorphizing AI can actually be the right approach
🔸 What the "J-space" reveals about how models think
🔸 How interpretability connects to AI alignment and safety

Listen noyoutu.be/JHo-aiePjoMfpkab

🔗 View Quoted Tweet

💬26🔄4❤️37👀12306📊23

⚡ Powered by xgo.ing

Thank you to our wonderful sponsors @WorkOS and @DeveloperXM for supporting the podcast: 🏆 @WorkOS ...

Lenny Rachitsky (@lennysan) — Sun, 18 Oct 2026 13:39:21 GMT

Thank you to our wonderful sponsors @WorkOS and @DeveloperXM for supporting the podcast:
🏆 @WorkOS — Make your app Enterprise Ready, with SSO, SCIM, RBAC, and more:workos.com/lenny9
🏆 @DeveloperXM — Engineering intelligence for the AI eragetdx.com/?utm_source=le…BE

Also available on:
• Spotifyopen.spotify.com/episode/5QT3Co…js
• Applepodcasts.apple.com/us/podcast/ins…mL

💬0🔄1❤️1👀1602📊1

⚡ Powered by xgo.ing

"To make them is not to understand them." Emmanuel Ameisen (@mlpowered) has spent the last 3 years ...

Lenny Rachitsky (@lennysan) — Sun, 18 Oct 2026 13:39:20 GMT

"To make them is not to understand them."

Emmanuel Ameisen (@mlpowered) has spent the last 3 years doing interpretability research at @AnthropicAI, looking inside LLMs like Claude to figure out what they're actually doing. As he puts it, models are "grown," not built.

We discuss:
🔸 His discovery that models plan many tokens ahead
🔸 Why anthropomorphizing AI can actually be the right approach
🔸 What the "J-space" reveals about how models think
🔸 How interpretability connects to AI alignment and safety

Listen noyoutu.be/JHo-aiePjoMfpkab

💬53🔄18❤️180👀120211📊94

⚡ Powered by xgo.ing

Full talk here https://t.co/nN8tCos58P

Lenny Rachitsky (@lennysan) — Sat, 17 Oct 2026 13:05:56 GMT

Full talk here
youtu.be/DqF08Dz3nok

💬0🔄1❤️15👀8181📊8

⚡ Powered by xgo.ing

"There are some people in your org who are doing way more exploring than others. They are your early...

Lenny Rachitsky (@lennysan) — Sat, 17 Oct 2026 12:44:19 GMT

"There are some people in your org who are doing way more exploring than others. They are your early adopters. They're running Astra and Fable on the weekends on personal projects, to find things they might bring to work.

The problem is that they can be a massive f*cking distraction.

So the key question to ask yourself if you're running a product org is: how do I harness these early adopters without distracting everybody else?" @danshipper

Your browser does not support the video tag.

🔗 View on Twitter

💬119🔄36❤️577👀54595📊252

⚡ Powered by xgo.ing

"Refuse to confuse done with good." @lil_dill

Lenny Rachitsky (@lennysan) — Fri, 16 Oct 2026 20:51:00 GMT

"Refuse to confuse done with good." @lil_dill

Your browser does not support the video tag.

🔗 View on Twitter

Katie Dill@lil_dill

If we only use AI to make the things that we already make, just faster, then we're definitely missing out on the most interesting part.

From my talk at @lennysan’s Summit, here are four ideas for raising the ceiling, not just the floor:

- Have a point of view [5:38]
- Encode your standards into the machine [9:07]
- Don't confuse done with good [12:21]
- Unleash creativity and artistry [16:19]

Full talk here: youtu.be/GLvFTMtw4Jk?si…

🔗 View Quoted Tweet

💬28🔄13❤️187👀17572📊46

⚡ Powered by xgo.ing

Full talk here https://t.co/C1tEiMwIGH

Lenny Rachitsky (@lennysan) — Fri, 16 Oct 2026 20:09:20 GMT

Full talk here youtu.be/GLvFTMtw4Jk?si…

💬1🔄11❤️112👀18411📊48

⚡ Powered by xgo.ing

Head of Design at Stripe on how to stop creating zombie UIs

Lenny Rachitsky (@lennysan) — Fri, 16 Oct 2026 13:34:58 GMT

Head of Design at Stripe on how to stop creating zombie UIs

Your browser does not support the video tag.

🔗 View on Twitter

Katie Dill@lil_dill

If we only use AI to make the things that we already make, just faster, then we're definitely missing out on the most interesting part.

From my talk at @lennysan’s Summit, here are four ideas for raising the ceiling, not just the floor:

- Have a point of view [5:38]
- Encode your standards into the machine [9:07]
- Don't confuse done with good [12:21]
- Unleash creativity and artistry [16:19]

Full talk here: youtu.be/GLvFTMtw4Jk?si…

🔗 View Quoted Tweet

💬107🔄467❤️5153👀344346📊1385

⚡ Powered by xgo.ing

@lil_dill Beating her former boss @bchesky! Here's the talk: https://t.co/dHFnvQVLuu

Lenny Rachitsky (@lennysan) — Thu, 15 Oct 2026 18:20:45 GMT

@lil_dill Beating her former boss @bchesky!

Here's the talk: youtube.com/watch?v=GLvFTM…

💬4🔄1❤️6👀3107📊5

⚡ Powered by xgo.ing

Katie Dill (@lil_dill)'s talk from the Lenny & Friends Summit is now my third most viewed YT vid...

Lenny Rachitsky (@lennysan) — Thu, 15 Oct 2026 18:19:59 GMT

Katie Dill (@lil_dill)'s talk from the Lenny & Friends Summit is now my third most viewed YT video of all-time 👏👏👏

Katie Dill@lil_dill

If we only use AI to make the things that we already make, just faster, then we're definitely missing out on the most interesting part.

From my talk at @lennysan’s Summit, here are four ideas for raising the ceiling, not just the floor:

- Have a point of view [5:38]
- Encode your standards into the machine [9:07]
- Don't confuse done with good [12:21]
- Unleash creativity and artistry [16:19]

Full talk here: youtu.be/GLvFTMtw4Jk?si…

🔗 View Quoted Tweet

💬31🔄2❤️95👀13749📊32

⚡ Powered by xgo.ing

@danhockenmaier Full post https://t.co/0vHBsikTTa

Lenny Rachitsky (@lennysan) — Thu, 15 Oct 2026 13:26:43 GMT

@danhockenmaier Full post lennysnewsletter.com/p/how-to-tell-…

💬1🔄1❤️8👀3052📊3

⚡ Powered by xgo.ing

The only four real moats btw

Lenny Rachitsky (@lennysan) — Thu, 15 Oct 2026 13:17:06 GMT

The only four real moats btw

Lenny Rachitsky@lennysan

Moats! Does anyone have one anymore?

Yes. Yes, they do.

But only a small subset of what people call moats create real long-term value and defensibility.

Not actually a moat:
1. Distribution: This does matter more than ever. Building products has gotten easier, but getting them to customers has not. Distribution, however, does not create durable defensibility. It helps create the raw ingredients for one, but it is not itself a moat.

2. Execution: Yes, speed is critical for startups. But a company that has to sprint forever is not a good business. Ultimately, it will misstep, slow down, and competition will catch up. Speed gives a company more shots on goal as it tries to build a great product and a great business, but it must eventually find something more durable.

3. Brand: The idea is that when everyone is building the same thing, the most resonant brand will win. But for startups, brand works more like customer acquisition spend: Put money in, get growth out. It’s not something that will effectively keep others from competing long-term. Brand can become a true moat over time (as with Coke or Nike or Apple), but it just takes too long to really matter for most startups.

What you're looking for is compounding advantages: a self-reinforcing flywheel that gets stronger and stronger over time, ultimately making a company very hard to compete with long-term.

There are only four compounding advantages that really matter. Every major tech company has at least one:

1. Switching costs: The more deeply embedded a product is in the way a customer works, the harder it is for that customer to switch. This increases retention and revenue per customer, which can be reinvested into the product.

2. Network effects: The more users a product has, the better it gets for everyone. This increases engagement and attracts more users.

3. Scale economies: The more volume a company has, the larger and more efficient its operations can be. This lowers unit costs, funding lower prices or a better product at the same price, further driving up volume.

4. Data effects: The more usage a product has, the more data it generates. This data can be used to improve the product and increase usage.

If you can’t see the potential for any in the startup you’re considering joining, that’s a red flag.

In today’s incredible post, @danhockenmaier shares a simple framework you can run in about an hour after getting an offer:
1. Identify a startup’s compounding advantage
2. Gauge how strong it could get
3. Measure whether it’s working

Inside: dozens of real company examples, the specific metrics to track, and benchmarks for what good and great look like.

Read it here: lennysnewsletter.com/p/how-to-tell-…

🔗 View Quoted Tweet

💬53🔄26❤️351👀25321📊107

⚡ Powered by xgo.ing

There’s way too much AI hype right now. So @noamseg and I are launching "Hype(less)," a research pr...

Lenny Rachitsky (@lennysan) — Wed, 14 Oct 2026 18:50:23 GMT

There’s way too much AI hype right now.

So @noamseg and I are launching "Hype(less)," a research project to showcase the most genuine, impactful, and repeatable AI workflows that have created real value for you and your teams in product, design, and engineering.

We’ll showcase the best in a special three-part newsletter series in the coming months.

Here's our ask: If you're a PM, engineer, or designer, please share your most valuable AI workflow here: tally.so/r/dWL6Xd (by 10/23).

By sharing your workflow, you'll help others cut through the AI hype, and share what you've learned with the world.

Here's the form: tally.so/r/dWL6Xd

(Please RT and share with your colleagues so we can collect as many great examples as possible).

💬74🔄38❤️349👀23203📊140

⚡ Powered by xgo.ing

Also a plug-and-play skill to explore it with your favorite AI https://t.co/yUSMdyUBRT

Lenny Rachitsky (@lennysan) — Wed, 14 Oct 2026 17:53:04 GMT

Also a plug-and-play skill to explore it with your favorite AI x.com/lennysan/statu…

Lenny Rachitsky@lennysan

Hot off the presses: A plug-and-play moat check skill

Share the company you're considering joining, it'll help you understand if they have a moat.

Get it here: claude.ai/artifact/MFhcN…

Here's me asking about @tryramp 👇

🔗 View Quoted Tweet

💬2🔄1❤️5👀3260📊3

⚡ Powered by xgo.ing

Here it is: Benchmarks to measure a given company's moat Image 1: KPIs to look at, by potential moa...

Lenny Rachitsky (@lennysan) — Wed, 14 Oct 2026 17:46:02 GMT

Here it is: Benchmarks to measure a given company's moat

Image 1: KPIs to look at, by potential moat.
Image 2: What "good" and "great" looks like.

Note, these are rough and meant to be interpreted directionally, because there is so much variance by industry and business model. For a deeper exploration, see links in the PDF here: drive.google.com/file/d/1GESyId…

And for a lot more context and advice, don't miss @danhockenmaier's full post: lennysnewsletter.com/p/how-to-tell-…

Lenny Rachitsky@lennysan

Moats! Does anyone have one anymore?

Yes. Yes, they do.

But only a small subset of what people call moats create real long-term value and defensibility.

Not actually a moat:
1. Distribution: This does matter more than ever. Building products has gotten easier, but getting them to customers has not. Distribution, however, does not create durable defensibility. It helps create the raw ingredients for one, but it is not itself a moat.

2. Execution: Yes, speed is critical for startups. But a company that has to sprint forever is not a good business. Ultimately, it will misstep, slow down, and competition will catch up. Speed gives a company more shots on goal as it tries to build a great product and a great business, but it must eventually find something more durable.

3. Brand: The idea is that when everyone is building the same thing, the most resonant brand will win. But for startups, brand works more like customer acquisition spend: Put money in, get growth out. It’s not something that will effectively keep others from competing long-term. Brand can become a true moat over time (as with Coke or Nike or Apple), but it just takes too long to really matter for most startups.

What you're looking for is compounding advantages: a self-reinforcing flywheel that gets stronger and stronger over time, ultimately making a company very hard to compete with long-term.

There are only four compounding advantages that really matter. Every major tech company has at least one:

1. Switching costs: The more deeply embedded a product is in the way a customer works, the harder it is for that customer to switch. This increases retention and revenue per customer, which can be reinvested into the product.

2. Network effects: The more users a product has, the better it gets for everyone. This increases engagement and attracts more users.

3. Scale economies: The more volume a company has, the larger and more efficient its operations can be. This lowers unit costs, funding lower prices or a better product at the same price, further driving up volume.

4. Data effects: The more usage a product has, the more data it generates. This data can be used to improve the product and increase usage.

If you can’t see the potential for any in the startup you’re considering joining, that’s a red flag.

In today’s incredible post, @danhockenmaier shares a simple framework you can run in about an hour after getting an offer:
1. Identify a startup’s compounding advantage
2. Gauge how strong it could get
3. Measure whether it’s working

Inside: dozens of real company examples, the specific metrics to track, and benchmarks for what good and great look like.

Read it here: lennysnewsletter.com/p/how-to-tell-…

🔗 View Quoted Tweet

💬41🔄22❤️302👀31021📊126

⚡ Powered by xgo.ing

Head of @GoogleLabs: "Great ideas rarely from design sprints, or brainstorms, or the time you set as...

Lenny Rachitsky (@lennysan) — Tue, 13 Oct 2026 23:45:00 GMT

Head of @GoogleLabs: "Great ideas rarely from design sprints, or brainstorms, or the time you set aside on a calendar. You can't microwave good ideas."

Your browser does not support the video tag.

🔗 View on Twitter

Lenny Rachitsky@lennysan

Every company needs to think and operate like a labs team now.

What’s possible changes every few months, and to stay ahead, you need to be experimenting at the frontier.

Here's my conversation with @JoshWoodward, Head of @GoogleLabs, the @GeminiApp, and @GoogleAIStudio. 

We discuss:
🔸 What skills are rising fastest in value right now
🔸 Where the best ideas actually come from
🔸 How to recognize PMF early
🔸 Why you should keep a list of what’s ‘almost possible’
🔸 Three rules for standing up an internal labs team

Listen nyoutu.be/3I6QYkccd3oaywBqC

🔗 View Quoted Tweet

💬31🔄11❤️141👀15848📊39

⚡ Powered by xgo.ing

@tryramp @danhockenmaier 🏆

Lenny Rachitsky (@lennysan) — Tue, 13 Oct 2026 22:17:57 GMT

@tryramp@danhockenmaier 🏆

💬0🔄0❤️3👀3293📊1

⚡ Powered by xgo.ing

Hot off the presses: A plug-and-play moat check skill Share the company you're considering joining,...

Lenny Rachitsky (@lennysan) — Tue, 13 Oct 2026 21:59:25 GMT

Hot off the presses: A plug-and-play moat check skill

Share the company you're considering joining, it'll help you understand if they have a moat.

Get it here: claude.ai/artifact/MFhcN…

Here's me asking about @tryramp 👇

💬18🔄24❤️319👀43036📊163

⚡ Powered by xgo.ing

Here's the full guide https://t.co/0vHBsikTTa

Lenny Rachitsky (@lennysan) — Tue, 13 Oct 2026 21:52:49 GMT

Here's the full guide lennysnewsletter.com/p/how-to-tell-…

💬0🔄1❤️3👀2533📊2

⚡ Powered by xgo.ing

Head of Google Labs, Gemini app, and AI Studio on the skill he's hiring for right now: 1. "Unlearnin...

Lenny Rachitsky (@lennysan) — Tue, 13 Oct 2026 21:45:00 GMT

Head of Google Labs, Gemini app, and AI Studio on the skill he's hiring for right now:
1. "Unlearning" rate
2. Explosive endurance

Full conversation: youtu.be/3I6QYkccd3o?si…

Your browser does not support the video tag.

🔗 View on Twitter

Lenny Rachitsky@lennysan

Every company needs to think and operate like a labs team now.

What’s possible changes every few months, and to stay ahead, you need to be experimenting at the frontier.

Here's my conversation with @JoshWoodward, Head of @GoogleLabs, the @GeminiApp, and @GoogleAIStudio. 

We discuss:
🔸 What skills are rising fastest in value right now
🔸 Where the best ideas actually come from
🔸 How to recognize PMF early
🔸 Why you should keep a list of what’s ‘almost possible’
🔸 Three rules for standing up an internal labs team

Listen nyoutu.be/3I6QYkccd3oaywBqC

🔗 View Quoted Tweet

💬14🔄9❤️138👀15213📊33

⚡ Powered by xgo.ing

"Look at people's eyes. That's the metric. When you're showing people early prototypes to see if we...

Lenny Rachitsky (@lennysan) — Tue, 13 Oct 2026 20:55:00 GMT

"Look at people's eyes. That's the metric.

When you're showing people early prototypes to see if we have PMF, what I'm always doing is looking at people's eyes.

It is not about DAU, or DAU/MAU. It's none of that. 

It's literally like, hey, Lenny, did you see this? What do you think about this? If your eyes light up, if you lean in, these are the signals you're looking for."

Full conversation: youtu.be/3I6QYkccd3o?si…

Your browser does not support the video tag.

🔗 View on Twitter

Lenny Rachitsky@lennysan

Every company needs to think and operate like a labs team now.

What’s possible changes every few months, and to stay ahead, you need to be experimenting at the frontier.

Here's my conversation with @JoshWoodward, Head of @GoogleLabs, the @GeminiApp, and @GoogleAIStudio. 

We discuss:
🔸 What skills are rising fastest in value right now
🔸 Where the best ideas actually come from
🔸 How to recognize PMF early
🔸 Why you should keep a list of what’s ‘almost possible’
🔸 Three rules for standing up an internal labs team

Listen nyoutu.be/3I6QYkccd3oaywBqC

🔗 View Quoted Tweet

💬20🔄12❤️128👀20109📊33

⚡ Powered by xgo.ing

.@JoshWoodward: "Watch this space. We hear you loud and clear.”

Lenny Rachitsky (@lennysan) — Tue, 13 Oct 2026 20:06:32 GMT

.@JoshWoodward: "Watch this space. We hear you loud and clear.”

Your browser does not support the video tag.

🔗 View on Twitter

signüll@signulll

google has an almost absurd structural advantage in personal ai.

chrome knows what you browse. android/gmail knows who you communicate with. calendar knows your commitments. maps knows where you go. photos knows your memories. youtube knows your interests. docs knows what you create. android knows what you do on your phone.

almost every important dimension of personal context already exists inside google.

which is why it's pretty remarkable that there is just no unifying ai product in place that works to make all of this work & work well. with clarity & simplicity.

instead google is effectively powering all personal ai products including meta.

🔗 View Quoted Tweet

💬6🔄1❤️47👀14982📊10

⚡ Powered by xgo.ing

Thank you @motion_so and @claudeai for this gem 🤣

Lenny Rachitsky (@lennysan) — Tue, 13 Oct 2026 19:58:01 GMT

Thank you @motion_so and @claudeai for this gem 🤣

💬3🔄0❤️9👀2358📊4

⚡ Powered by xgo.ing

Moats!

Lenny Rachitsky (@lennysan) — Tue, 13 Oct 2026 19:55:41 GMT

Moats!

Your browser does not support the video tag.

🔗 View on Twitter

💬6🔄6❤️173👀48813📊50

⚡ Powered by xgo.ing

This

Lenny Rachitsky (@lennysan) — Tue, 13 Oct 2026 19:32:46 GMT

This

Katya Rozenoer@marketeran

Meant as a framework for people evaluating job offers but in reality it's a skill.md for founders to evaluate an idea before touching it.

From the post

"There are only four compounding advantages that really matter. Every major tech company has at least one:

1. Switching costs: The more deeply embedded a product is in the way a customer works, the harder it is for that customer to switch. This increases retention and revenue per customer, which can be reinvested into the product.

2. Network effects: The more users a product has, the better it gets for everyone. This increases engagement and attracts more users.

3. Scale economies: The more volume a company has, the larger and more efficient its operations can be. This lowers unit costs, funding lower prices or a better product at the same price, further driving up volume.

4. Data effects: The more usage a product has, the more data it generates. This data can be used to improve the product and increase usage."

🔗 View Quoted Tweet

💬1🔄0❤️41👀14620📊10

⚡ Powered by xgo.ing

The four compounding advantages 👇 https://t.co/0vHBsikTTa

Lenny Rachitsky (@lennysan) — Tue, 13 Oct 2026 19:22:58 GMT

The four compounding advantages 👇lennysnewsletter.com/p/how-to-tell-…a

💬1🔄1❤️21👀7440📊13

⚡ Powered by xgo.ing

Distribution is not a moat. Execution is not a moat. Brand is not a moat.

Lenny Rachitsky (@lennysan) — Tue, 13 Oct 2026 19:17:09 GMT

Distribution is not a moat.

Execution is not a moat.

Brand is not a moat.

💬68🔄19❤️282👀136063📊136

⚡ Powered by xgo.ing

Moats! Does anyone have one anymore? Yes. Yes, they do. But only a small subset of what people cal...

Lenny Rachitsky (@lennysan) — Tue, 13 Oct 2026 17:24:29 GMT

Moats! Does anyone have one anymore?

Yes. Yes, they do.

But only a small subset of what people call moats create real long-term value and defensibility.

Not actually a moat:
1. Distribution: This does matter more than ever. Building products has gotten easier, but getting them to customers has not. Distribution, however, does not create durable defensibility. It helps create the raw ingredients for one, but it is not itself a moat.

2. Execution: Yes, speed is critical for startups. But a company that has to sprint forever is not a good business. Ultimately, it will misstep, slow down, and competition will catch up. Speed gives a company more shots on goal as it tries to build a great product and a great business, but it must eventually find something more durable.

3. Brand: The idea is that when everyone is building the same thing, the most resonant brand will win. But for startups, brand works more like customer acquisition spend: Put money in, get growth out. It’s not something that will effectively keep others from competing long-term. Brand can become a true moat over time (as with Coke or Nike or Apple), but it just takes too long to really matter for most startups.

What you're looking for is compounding advantages: a self-reinforcing flywheel that gets stronger and stronger over time, ultimately making a company very hard to compete with long-term.

There are only four compounding advantages that really matter. Every major tech company has at least one:

1. Switching costs: The more deeply embedded a product is in the way a customer works, the harder it is for that customer to switch. This increases retention and revenue per customer, which can be reinvested into the product.

2. Network effects: The more users a product has, the better it gets for everyone. This increases engagement and attracts more users.

3. Scale economies: The more volume a company has, the larger and more efficient its operations can be. This lowers unit costs, funding lower prices or a better product at the same price, further driving up volume.

4. Data effects: The more usage a product has, the more data it generates. This data can be used to improve the product and increase usage.

If you can’t see the potential for any in the startup you’re considering joining, that’s a red flag.

In today’s incredible post, @danhockenmaier shares a simple framework you can run in about an hour after getting an offer:
1. Identify a startup’s compounding advantage
2. Gauge how strong it could get
3. Measure whether it’s working

Inside: dozens of real company examples, the specific metrics to track, and benchmarks for what good and great look like.

Read it here: lennysnewsletter.com/p/how-to-tell-…

💬138🔄105❤️1059👀202204📊517

⚡ Powered by xgo.ing

"The team almost always knows a project is dead before the leader does." — @JoshWoodward (Head of G...

Lenny Rachitsky (@lennysan) — Tue, 13 Oct 2026 00:45:00 GMT

"The team almost always knows a project is dead before the leader does."

— @JoshWoodward (Head of Google Labs, Gemini App, Google AI Studio)

Your browser does not support the video tag.

🔗 View on Twitter

Lenny Rachitsky@lennysan

Every company needs to think and operate like a labs team now.

What’s possible changes every few months, and to stay ahead, you need to be experimenting at the frontier.

Here's my conversation with @JoshWoodward, Head of @GoogleLabs, the @GeminiApp, and @GoogleAIStudio. 

We discuss:
🔸 What skills are rising fastest in value right now
🔸 Where the best ideas actually come from
🔸 How to recognize PMF early
🔸 Why you should keep a list of what’s ‘almost possible’
🔸 Three rules for standing up an internal labs team

Listen nyoutu.be/3I6QYkccd3oaywBqC

🔗 View Quoted Tweet

💬38🔄11❤️97👀14345📊38

⚡ Powered by xgo.ing

.@joshwoodward: "Every company right now needs a set of people who's job is to be at the frontier. ...

Lenny Rachitsky (@lennysan) — Mon, 12 Oct 2026 22:53:32 GMT

.@joshwoodward: "Every company right now needs a set of people who's job is to be at the frontier. 

You want people plugged into the hive mind.

They're the people who have eight $200/month subscriptions on everything.

You have to architect your organization where this can exist."

Your browser does not support the video tag.

🔗 View on Twitter

Lenny Rachitsky@lennysan

Every company needs to think and operate like a labs team now.

What’s possible changes every few months, and to stay ahead, you need to be experimenting at the frontier.

Here's my conversation with @JoshWoodward, Head of @GoogleLabs, the @GeminiApp, and @GoogleAIStudio. 

We discuss:
🔸 What skills are rising fastest in value right now
🔸 Where the best ideas actually come from
🔸 How to recognize PMF early
🔸 Why you should keep a list of what’s ‘almost possible’
🔸 Three rules for standing up an internal labs team

Listen nyoutu.be/3I6QYkccd3oaywBqC

🔗 View Quoted Tweet

💬29🔄20❤️244👀28422📊62

⚡ Powered by xgo.ing

Handy link to the episode https://t.co/1WT5Ff0xht

Lenny Rachitsky (@lennysan) — Mon, 12 Oct 2026 16:37:52 GMT

Handy link to the episode youtu.be/3I6QYkccd3o

💬1🔄0❤️6👀4980📊4

⚡ Powered by xgo.ing

My biggest takeaways from @JoshWoodward (Head of @Google Labs, the Gemini App, and Google AI Studio)...

Lenny Rachitsky (@lennysan) — Mon, 12 Oct 2026 14:02:20 GMT

My biggest takeaways from @JoshWoodward (Head of @Google Labs, the Gemini App, and Google AI Studio)

The best ideas almost never come from a design sprint or brainstorm. At Google Labs, the ideas that became NotebookLM, Google Flow, Google Beam, and AI Studio all started with small groups of people who were obsessed with an insight and couldn’t stop themselves from building. These ideas surfaced in the shower, over a weekend, in a hallway. The implication for teams everywhere is that the environment matters more than the process. “You cannot microwave a good idea.”

Fall in love with the problem, not the product. Most ideas require three, four, or five significant pivots before they either find traction or get shut down. Teams that get hooked on a specific solution will struggle to make those pivots. Teams that are attached to the problem—the underlying pain point or the user behavior they are trying to shift—can keep adjusting until they get it right. 

When looking for early PMF, data and retention are important, but the strongest signal of PMF is in people’s eyes. Do they light up? Do they lean in? When you show someone a prototype and they’re genuinely engaged, you feel it. When NotebookLM’s first AI podcast demo was played in a hallway, the reaction was immediate and unmistakable.

Keep a list of what is “almost possible,” and move fast when something crosses over. Google Labs keeps a live list of experiences that are almost possible. When something crosses the threshold, Josh describes it as “a phase shift,” like solid to gas, and the team moves immediately. Maintain a running list of things that are almost ready, and have a team positioned to sprint when one of them arrives.

Every company needs to think like a labs team now. The underlying technology is changing so fast that what’s possible is shifting from week to week. Which means that every company needs to be spending time playing with the latest models, running far-out experiments, and being first to discover what they unlock before the competition does—and then bringing those discoveries to the rest of the org.

When something is not working, the team almost always knows before the leader does. Josh’s instinct as a leader is not to be the one who decides to shut down a project. He watches instead for the passion running out, for the sense that the team has tried everything and nothing is working. His goal is to create an environment where people feel safe enough to say, “I don’t think this is working.”

Someone’s “unlearning rate” is one of the most valuable skills right now. Everyone talks about growth mindset and learning speed. Josh is watching for something different: How quickly can someone take an insight, validate or invalidate it through testing, and then let it go? The ability to walk away from a prior belief or approach when the evidence says you should is rare and becoming more valuable.

Roles are not blurring into one universal “builder” role. There is a tendency in the current moment to talk as though everyone is becoming a generalist builder. Josh thinks this would be a bad idea. The best people on his teams still have a primary specialty—a major—and also a “minor” they’ve developed. The advice he gives to early-career PMs who are excited about vibe coding and prototyping is clear: build adjacent skills in service of becoming an exceptional PM, not as a replacement for it.

A well-designed lab needs isolation, a clear mission, and the ability to create “good trouble.” Three things matter when setting up an internal labs team. First, it needs structural independence, reporting far enough from existing business units that weird ideas can survive long enough to prove themselves. Second, the lab’s mandate has to be explicit: are you graduating ideas to existing products, or inventing new categories? Clarity on that question changes almost every decision. Third, the lab should be a place to challenge how the company itself works, not just what it builds. At Google Labs, that included piloting new job ladders and builder roles before rolling them out more broadly.

Google Labs manages the marathon-sprint tension through seasons. At any given time, roughly 20 to 30 experiments are running in parallel, but individual teams experience a defined rhythm: sprints, post-launch recovery, and open exploration periods. After Google I/O each year, many labs teams enter a mode that looks unproductive from the outside: no shipping, just hacking, building, and discovering. Josh treats this as the seedbed for the next season.

💬83🔄103❤️764👀92868📊304

⚡ Powered by xgo.ing

Thank you to our wonderful sponsors @WorkOS and @DeveloperXM for supporting the podcast: 🏆 @WorkOS ...

Lenny Rachitsky (@lennysan) — Sun, 11 Oct 2026 15:58:57 GMT

Thank you to our wonderful sponsors @WorkOS and @DeveloperXM for supporting the podcast:
🏆 @WorkOS — Make your app Enterprise Ready, with SSO, SCIM, RBAC, and more:workos.com/lennyH
🏆 @DeveloperXM — Engineering intelligence for the AI eragetdx.com/?utm_source=le…rc

Also available on:
• Spotifyopen.spotify.com/episode/4nREPH…2V
• Applepodcasts.apple.com/us/podcast/why…Qs

💬1🔄0❤️2👀4326📊2

⚡ Powered by xgo.ing

Every company needs to think and operate like a labs team now. What’s possible changes every few mo...

Lenny Rachitsky (@lennysan) — Sun, 11 Oct 2026 15:58:56 GMT

Every company needs to think and operate like a labs team now.

What’s possible changes every few months, and to stay ahead, you need to be experimenting at the frontier.

Here's my conversation with @JoshWoodward, Head of @GoogleLabs, the @GeminiApp, and @GoogleAIStudio. 

We discuss:
🔸 What skills are rising fastest in value right now
🔸 Where the best ideas actually come from
🔸 How to recognize PMF early
🔸 Why you should keep a list of what’s ‘almost possible’
🔸 Three rules for standing up an internal labs team

Listen nyoutu.be/3I6QYkccd3oaywBqC

💬54🔄23❤️242👀235758📊115

⚡ Powered by xgo.ing

@Atlassian Watch Tamar's full talk here https://t.co/bBQSF4enPI

Lenny Rachitsky (@lennysan) — Sat, 10 Oct 2026 23:08:24 GMT

@Atlassian Watch Tamar's full talk here  youtube.com/watch?v=BtK4kF…

💬3🔄1❤️10👀7336📊6

⚡ Powered by xgo.ing

Emerging playbook for PMs at big companies exploring zero-to-one ideas: row, then steer. 1. Row: PM...

Lenny Rachitsky (@lennysan) — Sat, 10 Oct 2026 23:07:59 GMT

Emerging playbook for PMs at big companies exploring zero-to-one ideas: row, then steer.

1. Row: PM + designer vibe-code a working alpha
2. Once it shows pull, bring on engineers
3. Steer: Step back from coding to align, prioritize, and unblock

Great story from @Atlassian CPO @TYehoshua 👇

Your browser does not support the video tag.

🔗 View on Twitter

💬82🔄32❤️468👀39528📊211

⚡ Powered by xgo.ing

@danshipper Watch Dan's full 20m talk here https://t.co/0XqjBe0uHw

Lenny Rachitsky (@lennysan) — Fri, 09 Oct 2026 20:14:34 GMT

@danshipper Watch Dan's full 20m talk here youtube.com/watch?v=DqF08D…

💬3🔄8❤️68👀36782📊39

⚡ Powered by xgo.ing

Tired: Two-pizza teams Wired: Two-slice teams @danshipper on the best team structure in the AI-era

Lenny Rachitsky (@lennysan) — Fri, 09 Oct 2026 20:14:06 GMT

Tired: Two-pizza teams

Wired: Two-slice teams

@danshipper on the best team structure in the AI-era

Your browser does not support the video tag.

🔗 View on Twitter

💬178🔄200❤️2153👀560047📊842

⚡ Powered by xgo.ing

Also, wild that Sonnet 5.5 is beating Asta and Sol.

Lenny Rachitsky (@lennysan) — Fri, 09 Oct 2026 16:39:32 GMT

Also, wild that Sonnet 5.5 is beating Asta and Sol.

💬1🔄0❤️3👀3651📊2

⚡ Powered by xgo.ing

Sweet new benchmark from Surge looking at how close AI is getting to a staff-level software engineer...

Lenny Rachitsky (@lennysan) — Fri, 09 Oct 2026 16:31:57 GMT

Sweet new benchmark from Surge looking at how close AI is getting to a staff-level software engineering. We're half-way there.

echen@echen

introducing sudo L7, our new benchmark for staff-level software engineering.

coding agents are getting very good at being L3s. give them a well-defined ticket and they can write excellent code.

but a staff engineer is valuable for everything that happens around the code.

what should we build? what already exists? what's going to break in prod? what did the tests miss? should we even be doing this?

sudo L7 has 60 tasks, mostly in private production repos from real companies, written by engineers who've actually owned these systems.

expert rubrics grade everything from functional correctness to engineering craft, architectural judgment, thought partnership, and unnecessary complexity.

the best agents succeed on only ~45% of tasks.

coding is only part of the job. for now, we're keeping them at idiot genius L3.

🔗 View Quoted Tweet

💬25🔄7❤️84👀21115📊30

⚡ Powered by xgo.ing

Been loving this app. Great work @benspringwater.

Lenny Rachitsky (@lennysan) — Thu, 08 Oct 2026 20:59:28 GMT

Been loving this app. Great work @benspringwater.

Ben Springwater@benspringwater

I'm delighted to introduce Sundrop – a private photo app for parents. trysundrop.app ☀️

Sundrop finds your best photos and turns them into delightful keepsakes.

See your kids side-by-side at the same age. Watch timelapses of them growing up. Relive your family's week. And lots more.

Everything happens on your iPhone; your photos never touch our servers. (We wouldn't trust a company with our photo libraries, either.)

Sundrop has brought a lot of joy to our own families. We hope it does the same for yours!

Your browser does not support the video tag.

🔗 View on Twitter

🔗 View Quoted Tweet

💬6🔄5❤️161👀39052📊39

⚡ Powered by xgo.ing

@ElenaVerna Watch Elena's full talk (20m) here https://t.co/1t0YUyny8P

Lenny Rachitsky (@lennysan) — Thu, 08 Oct 2026 20:26:23 GMT

@ElenaVerna Watch Elena's full talk (20m) here youtube.com/watch?v=fn8wnm…

💬0🔄0❤️9👀6344📊4

⚡ Powered by xgo.ing

“If we truly believe that AI is changing how we work, then ICs should get paid more than managers, b...

Lenny Rachitsky (@lennysan) — Thu, 08 Oct 2026 20:25:56 GMT

“If we truly believe that AI is changing how we work, then ICs should get paid more than managers, because an IC can drive more impact than a manager could in the pre-AI system.” — @ElenaVerna

Your browser does not support the video tag.

🔗 View on Twitter

💬56🔄20❤️245👀71780📊95

⚡ Powered by xgo.ing

@karrisaarinen Full talk here https://t.co/SgdamJc0TU

Lenny Rachitsky (@lennysan) — Thu, 08 Oct 2026 18:07:15 GMT

@karrisaarinen Full talk here youtube.com/watch?v=Zn9NZ-…

💬0🔄0❤️5👀4479📊3

⚡ Powered by xgo.ing

Linear CEO @karrisaarinen: "Making products produces two things: the product, and the learning. The...

Lenny Rachitsky (@lennysan) — Thu, 08 Oct 2026 18:06:52 GMT

Linear CEO @karrisaarinen: "Making products produces two things: the product, and the learning.

The effort you put into building and designing things teaches you something about what the problem is, what the customers want.

We're in this time now that there's this danger of losing that direct connection with the learning."

Your browser does not support the video tag.

🔗 View on Twitter

💬47🔄14❤️201👀38501📊72

⚡ Powered by xgo.ing

Rad PM job alert

Lenny Rachitsky (@lennysan) — Thu, 08 Oct 2026 17:14:26 GMT

Rad PM job alert

Dan Shipper@danshipper

NEW JOB ALERT:

@every we've built a personal benchmarking platform, Checks, to help anyone measure how good new models are for their real work

we're looking for someone who understands why this is so important, and wants to help us build it into the future of how great human work gets done with AI:

modern-ton-234.notion.site/d6edc6e5d9704e…

🔗 View Quoted Tweet

💬4🔄4❤️81👀17808📊14