Scott Wu(@ScottWu46)

Fun fact: @marvinvonhagen was on set for the original Devin launch video back in 2024. Really excite...

Scott Wu (@ScottWu46) — Thu, 23 Jul 2026 15:32:33 GMT

Fun fact: @marvinvonhagen was on set for the original Devin launch video back in 2024. Really excited to work with this team - it’s been two years in the making :)

Cognition@cognition

Cognition is acquiring @Interaction, the makers of Poke.

If you've ever used Poke, you know why we're so excited.

Your browser does not support the video tag.

🔗 View on Twitter

🔗 View Quoted Tweet

💬10🔄4❤️175👀11985📊22

⚡ Powered by xgo.ing

AI gets more inevitable every day but most companies don't have a clear path to take full advantage ...

Scott Wu (@ScottWu46) — Wed, 22 Jul 2026 01:39:19 GMT

AI gets more inevitable every day but most companies don't have a clear path to take full advantage of it. Danielle, Leah, and the whole Genius AI team are solving the practical problems that matter. Congratulations on all the progress!

DanielleCohen-Shohet@DCohenShohet

Introducing Genius AI: the platform handling scheduling, payments & admin for service businesses.

GlossGenius is now part of Genius AI. We’re nearing $200M ARR, powering over 125K businesses.

We’re also announcing our Series D at a $1.15B valuation, with $125M raised to date.

🔗 View Quoted Tweet

💬5🔄4❤️106👀20840📊14

⚡ Powered by xgo.ing

Welcome to Cognition!

Scott Wu (@ScottWu46) — Tue, 21 Jul 2026 18:03:52 GMT

Welcome to Cognition!

Jared Palmer@jaredpalmer

Personal update: I’ve joined @Cognition to lead engineering. 

I’m grateful to @asha_shar and everyone at Xbox for an amazing year. 

Excited for this next chapter with @ScottWu46 and team.

🔗 View Quoted Tweet

💬5🔄5❤️190👀23121📊17

⚡ Powered by xgo.ing

It's an honor to get to work with Anhang, Yun and the TierZero team. Excited for the next step of se...

Scott Wu (@ScottWu46) — Mon, 20 Jul 2026 21:23:59 GMT

It's an honor to get to work with Anhang, Yun and the TierZero team. Excited for the next step of self-driving software!

Anhang Zhu@anhangz

Two years ago, we set out to make software self-driving.

Today, TierZero is joining @Cognition to finish the job: make agents write code and keep it running.

@yunpark93 and I couldn't be more excited for what's next! 🚀

Your browser does not support the video tag.

🔗 View on Twitter

🔗 View Quoted Tweet

💬2🔄6❤️99👀22426📊11

⚡ Powered by xgo.ing

Great piece on my good friend @saranormous! Grateful for everything you do.

Scott Wu (@ScottWu46) — Mon, 20 Jul 2026 16:05:16 GMT

Great piece on my good friend @saranormous! Grateful for everything you do.

Colossus@colossusmag

In 2018, Sarah Guo became the youngest general partner in Greylock's 60-year history. She was 28. Four years later, she quit to launch Conviction, a firm staked entirely on AI.

Before ChatGPT shipped, she seeded Baseten and Harvey; each is now worth over $11 billion. In Conviction's first year, she wrote early checks into Sierra, Cognition, and Mistral; those three companies are now worth, together, $54 billion.

Andrej Karpathy worked out of Conviction's office until Anthropic hired him in May. Guo has been close to Jensen Huang for over a decade. Her first two calls after starting the firm were to Sam Altman and Nat Friedman.

And yet the investor closest to the AI frontier is betting against its biggest companies.

The two big frontier labs, worth close to a trillion dollars apiece, no longer just want to build the models. They also want to build every product and company on top of them, leaving nothing for anyone else.

The market is paying as though they might succeed. Of the $300 billion in venture capital deployed in the first quarter of 2026, the biggest quarter in the history of the trade, 65% went to only four companies: Anthropic, OpenAI, xAI, and Waymo.

Guo is betting the labs can't build everything, and she spends her days making sure of it. She won Harvey its first client. She flew across the country to take a single Baseten candidate to a four-hour lunch. On one wedding anniversary, she spent the whole weekend on back-to-back calls, keeping two founders on the line so they couldn't speak to rival firms. Twice a year, she flies the world's brightest young founders to San Francisco and inducts them into the fight.

In the months @domcooke spent reporting this piece, @saranormous had her fourth child, walked the Met Gala in 45 pounds of chainmail, and still answered her founders' texts within minutes.

Guo's parents arrived from China in 1987 with $50, built a company, and took it public at $1.2 billion. Then it went bankrupt. Guo grew up inside that startup. She built its first website, did her homework in a cubicle, and slept over for bug bashes. She loved it.

If two labs build everything, no one gets to do that again.

Welcome to Sarah's Wager. Read it below.

🔗 View Quoted Tweet

💬2🔄6❤️89👀20045📊10

⚡ Powered by xgo.ing

https://t.co/ZAC0PZZdFJ

Scott Wu (@ScottWu46) — Tue, 14 Jul 2026 17:09:42 GMT

cognition.com/blog/one-year-…

💬0🔄0❤️32👀7786📊4

⚡ Powered by xgo.ing

One year ago today, Cognition acquired Windsurf, capping off one of the crazier weekends in our comp...

Scott Wu (@ScottWu46) — Tue, 14 Jul 2026 17:09:29 GMT

One year ago today, Cognition acquired Windsurf, capping off one of the crazier weekends in our company's history.

The outline has been told many times: the first call on Friday evening, the rush through Saturday and Sunday to figure out the plan together, and then the signed agreement by Monday morning. On the Cognition side, we knew we had to move fast — Windsurf had millions of users, a brand developers loved, and a great team. We printed an LOI and headed to their offices on Saturday morning ready to sign on the spot. When we got in a room with Jeff and Graham, things clicked: they had built one of the best GTM orgs in Silicon Valley and needed engineering; we had the engineering org that needed GTM. They were building an IDE, we were building a cloud agent. Everything lined up.

When we welcomed the team, we said “there's only one boat and we're all in it together.” Then we got to work. Three days after close, we shipped Wave 11: "Just Keep Shipping" – and that set the tone for everything to come. In the year since, we’ve launched our own models (including SWE1.7 just last week), Devin Review, Devin CLI, and unified everything into one brand with Devin Desktop. Along the way our team wrote 20M+ lines of code and grew revenue run rate from $73M to >$500M.

Thankful for the last year and excited to keep building the future of software engineering together!

💬58🔄42❤️1029👀152289📊173

⚡ Powered by xgo.ing

it's a pretty unique experience tbh. i grew up in louisiana where no one knew how to code and now i ...

Scott Wu (@ScottWu46) — Fri, 10 Jul 2026 23:48:52 GMT

it's a pretty unique experience tbh. i grew up in louisiana where no one knew how to code and now i just show up and get carried by the cog team all day

Chris Barber (in SF)@chrisbarber

I asked people which companies have the highest density of talented people they know:

1st. Cognition - 9 votes
Equal 1st. Anthropic - 9 votes
2nd. Modal - 5 votes
3rd. OpenAI - 4 votes
4th. Standard Intelligence - 3 votes
4th. Cursor - 3 votes

Two votes each:
- Ramp
- Flapping Airplanes
- DeepMind
- Long Lake
- Applied Compute

One vote each:
- SpaceXAI
- SpaceX (treated separately, one vote was for the AI lab subsidiary and one was for the rocket team)
- American Terawatt
- Mechanize
- Olix
- Fluidstack
- Chai Discovery
- Sail Research
- Etched
- Core Automation
- Specter
- Clay
- Applied Intuition
- Sierra
- Hivemind
- Bitrig
- Retro
- Thinking Machines
- Decagon
- Precigenetics
- Pangram
- Reflect
- Thrive Holdings
- Adaption

🔗 View Quoted Tweet

💬13🔄12❤️643👀106654📊64

⚡ Powered by xgo.ing

Benchmark scores are exciting but more importantly we are seeing incredible results so far using thi...

Scott Wu (@ScottWu46) — Wed, 08 Jul 2026 16:14:29 GMT

Benchmark scores are exciting but more importantly we are seeing incredible results so far using this model in Devin! Try it out:

Cognition@cognition

Introducing SWE-1.7, the most capable model we’ve trained yet.

It scores within a few points of the strongest frontier models at a fraction of the cost, and is now available at 1000 tok/s.

RL is not hitting its limit: after refining our recipe, we keep seeing gains as we scale

🔗 View Quoted Tweet

💬21🔄9❤️281👀30612📊35

⚡ Powered by xgo.ing

Happy 4th! Proud to be an American today and every day.

Scott Wu (@ScottWu46) — Sat, 04 Jul 2026 16:45:27 GMT

Happy 4th! Proud to be an American today and every day.

💬21🔄14❤️661👀28600📊75

⚡ Powered by xgo.ing

Demand for inference compute will grow exponentially at least as fast as the METR curve grows (and a...

Scott Wu (@ScottWu46) — Tue, 30 Jun 2026 19:13:47 GMT

Demand for inference compute will grow exponentially at least as fast as the METR curve grows (and arguably faster). Excited to have more great chips on the market - congrats to the @Etched team!

Etched@Etched

We're coming out of stealth.

We've built our first racks after a successful A0 tapeout, $1B+ in customer contracts, and $800m raised.

Early customer tests show us achieving SOTA throughput, latency, and power efficiency on inference workloads.

Our first racks ship this summer.

🔗 View Quoted Tweet

💬9🔄12❤️314👀32281📊34

⚡ Powered by xgo.ing

Model routing is a hot topic but there are two challenges to doing routing for coding agents: 1) Eve...

Scott Wu (@ScottWu46) — Mon, 29 Jun 2026 22:13:19 GMT

Model routing is a hot topic but there are two challenges to doing routing for coding agents:
1) Even if different models can pass the task, there are subtle differences in behavior & style that mean they aren't perfectly interchangeable.
2) The initial agent prompt isn't enough to know the difficulty of the task. "Fix xyz bug" could be a one-line edge case or could require rearchitecting your entire product; you can't know until you've actually investigated the code.

How do you solve these problems? Well, you need evals that account for style and behavior, not just pass/fail. And you need the agent to be able to dynamically update and re-route. We built Devin Fusion with both of these points in mind and found that it reduces costs by 30-40% while still maintaining the frontier intelligence "smell":

Cognition@cognition

Conventional model routing sucks. It passes benchmarks but fails to write code you'd actually merge. 

Introducing Devin Fusion, a new hybrid-model harness for agentic coding. 

In testing, it reduces the cost of Fable-level intelligence by 35% and still feels good to use.

🔗 View Quoted Tweet

💬36🔄54❤️603👀105091📊113

⚡ Powered by xgo.ing

huge shoutout to @ryanbai who spearheaded our AI Productivity Guarantee work!

Scott Wu (@ScottWu46) — Fri, 12 Jun 2026 19:58:32 GMT

huge shoutout to @ryanbai who spearheaded our AI Productivity Guarantee work!

Ryan Bai@ryanbai1412

x.com/i/article/2065…

🔗 View Quoted Tweet

💬3🔄5❤️77👀16861📊12

⚡ Powered by xgo.ing

we are going to need up our own human orchestration as much as we're upping our agent orchestration:

Scott Wu (@ScottWu46) — Wed, 10 Jun 2026 18:44:28 GMT

we are going to need up our own human orchestration as much as we're upping our agent orchestration:

Walden@walden_yan

My take 24 hours after Fable 5:

Your organization will likely not scale with the exponential curve of AI.

I'l just come out to say: This should be a wakeup call for engineering teams.

Set up your cloud software factories. Now.

Models can now fix impossible bugs, UI-test the hardest flows, writing extremely good code, etc. I have't opened Datadog manually as far as I can remember.

AI should be the first-line defense for bugs and feedback. Humans should only look at PRs after an AI has already reviewed it. AI should generate screen recordings of any PR before a human eye even reaches it. The agent should just prompt itself most of the time.

Ex. (pictured) our ui feedback channel manages itself, creates tickets, assigns itself automatically

You might also be worried about cost. Anthropic, OpenAI, and other labs will likely continue to put out bigger and more expensive models. But, we will also continue to get more capable small models. Not everything will need the smartest models. It's about having the organizational harness in place to continue taking advantage of this rising tide.

Moreover, if you use Devin, we've already optimized our harness a bit, and Fable is actually only ~40% more expensive in practice (vs the 2x people assume). I'm honestly pleasantly surprised - it might be higher ROI than you think.

Anyway, if you take anything away, engineers shouldn't be manually picking up tickets, humans shouldn't be digging into logs themselves, rethink what you do with your time that shouldn't just be an AI. We need to rethink what humans spend their time going.

🔗 View Quoted Tweet

💬7🔄7❤️165👀38675📊26

⚡ Powered by xgo.ing

Really thoughtful piece by @saranormous !

Scott Wu (@ScottWu46) — Wed, 10 Jun 2026 03:02:33 GMT

Really thoughtful piece by @saranormous !

sarah guo@saranormous

x.com/i/article/2064…

🔗 View Quoted Tweet

💬4🔄9❤️371👀155645📊99

⚡ Powered by xgo.ing

A new top scorer just one day after our benchmark released! Especially strong on the hardest tasks: ...

Scott Wu (@ScottWu46) — Tue, 09 Jun 2026 19:40:25 GMT

A new top scorer just one day after our benchmark released! Especially strong on the hardest tasks: 13.4% -> 29.3% on FrontierCode Diamond compared to Opus 4.8.

Cognition@cognition

Claude Fable 5 is now available in Devin.

Fable 5 earns the #1 spot on FrontierCode, our benchmark for real-world engineering tasks that grades mergeability and quality:

🔗 View Quoted Tweet

💬11🔄10❤️224👀23399📊26

⚡ Powered by xgo.ing

@NickADobos I hope that it gets solved in the next 6 months and then we can move on to even more cha...

Scott Wu (@ScottWu46) — Mon, 08 Jun 2026 23:48:28 GMT

@NickADobos I hope that it gets solved in the next 6 months and then we can move on to even more challenging tasks!

💬0🔄0❤️53👀34229📊6

⚡ Powered by xgo.ing

SWE-Bench style grading has been the standard for years now - you ask the agent to solve an issue an...

Scott Wu (@ScottWu46) — Mon, 08 Jun 2026 19:54:53 GMT

SWE-Bench style grading has been the standard for years now - you ask the agent to solve an issue and then run its code on a pre-constructed unit test.

The problem is that passing a unit test is only one part of writing production-ready code. You also want to evaluate agents on a number of other axes, including scope, coding style, and unintended side effects.

The result is our new benchmark FrontierCode - which has ~80% fewer false positives and for which the best model (Opus 4.8) only scores 13%!

"Where others grade like a CI, FrontierCode grades like a tech lead."

Cognition@cognition

Introducing FrontierCode: a coding eval that raises the bar for difficulty & quality. Each task took 40+ hrs of work by leading open-source maintainers.

Models write sloppy code that works but isn’t maintainable. Our eval is first to measure: would you actually merge this code?

🔗 View Quoted Tweet

💬43🔄42❤️631👀94758📊96

⚡ Powered by xgo.ing

Measuring someone's productivity by their token usage is a horrible idea. Giving everyone the same f...

Scott Wu (@ScottWu46) — Thu, 04 Jun 2026 18:23:13 GMT

Measuring someone's productivity by their token usage is a horrible idea. Giving everyone the same fixed token budget isn't much better. So what's the right way to roll out AI across your org?

We built a system to measure how many productive engineering hours every Devin task is worth, validated against a dataset of real engineers’ times estimates. The goal is to answer the fundamental question that companies are grappling with: how much real value are you getting from each of your agent sessions?

On top of that, we're giving an AI productivity guarantee! Now if Devin delivers less engineering value than you're paying for, we fund your usage until it does.

The whole industry needs to move from measuring activity to measuring output. We hope to see more AI companies taking this approach.

Cognition@cognition

AI should earn its keep. Introducing the AI Productivity Guarantee.

If Devin delivers less engineering value than you’re paying for, Cognition will fund your usage until it does, up to $10 million.

It’s time for the AI industry to stop maximizing tokens and start maximizing productive output.

🔗 View Quoted Tweet

💬59🔄59❤️896👀162879📊154

⚡ Powered by xgo.ing

Devin pushing the frontier on ECDSA circuits!

Scott Wu (@ScottWu46) — Thu, 04 Jun 2026 00:44:46 GMT

Devin pushing the frontier on ECDSA circuits!

gautham@bbuddha_xyz

Looks like @DevinAI the best research agent out there?

Nikhil from Devin is moving the frontier forward by reducing our the number of qubits our circuit requires.

🔗 View Quoted Tweet

💬5🔄9❤️121👀17753📊14

⚡ Powered by xgo.ing

Excited to partner with @Carahsoft! An important step for more of the public sector to have access t...

Scott Wu (@ScottWu46) — Wed, 03 Jun 2026 19:25:09 GMT

Excited to partner with @Carahsoft! An important step for more of the public sector to have access to the same tools and technology that the private sector is adopting en masse today.

Carahsoft@Carahsoft

We have partnered with @Cognition to bring their AI coding platform, Devin, to the Public Sector. Together, we’re helping organizations modernize software development and accelerate secure AI adoption. Learn more: carah.io/CognitionAIPR

🔗 View Quoted Tweet

💬1🔄1❤️70👀8836📊7

⚡ Powered by xgo.ing

Standalone IDEs have about 6 months left to live. An interface for manually editing and refactoring ...

Scott Wu (@ScottWu46) — Wed, 03 Jun 2026 02:28:14 GMT

Standalone IDEs have about 6 months left to live. An interface for manually editing and refactoring doesn’t need to exist if you're not manually editing and refactoring anymore.

So what's the right interface for a dev to be working in for 8h / day? Some parts are obvious: you want to be able to spin up agents (either local or cloud agents) and to have a clean interface to keep up with all of your parallel running agents. Then you want to be able to get into the weeds whenever needed for last-mile fixes and review.

But as software engineering continues to evolve we will see more and more of the lifecycle get reinvented. How do you build a single surface that allows you to plan, spec, prototype, debug, review, QA?

Bringing Devin and Windsurf together has been our vision ever since the acquisition. Devin Desktop is our first shot at what this looks like. Excited to make this a reality today!

Windsurf is now Devin Desktop@windsurf

Introducing Devin Desktop: the next generation of Windsurf

Manage fleets of local and cloud agents from one surface

Support for any ACP-compatible agent

With a full IDE for when you need to jump into the code

Your browser does not support the video tag.

🔗 View on Twitter

🔗 View Quoted Tweet

💬69🔄50❤️810👀249765📊168

⚡ Powered by xgo.ing

from the goat himself

Scott Wu (@ScottWu46) — Fri, 29 May 2026 19:07:51 GMT

from the goat himself

Ido Pesok@ido_pesok

x.com/i/article/2060…

🔗 View Quoted Tweet

💬7🔄31❤️494👀123998📊152

⚡ Powered by xgo.ing

@wolfejosh @cognition thank you for the kind words and for believing in us early!

Scott Wu (@ScottWu46) — Wed, 27 May 2026 23:28:14 GMT

@wolfejosh@cognition thank you for the kind words and for believing in us early!

💬1🔄5❤️31👀7569📊6

⚡ Powered by xgo.ing

Grateful for our customers, investors, partners, and most of all for our incredible team. Hard to be...

Scott Wu (@ScottWu46) — Wed, 27 May 2026 16:39:08 GMT

Grateful for our customers, investors, partners, and most of all for our incredible team. Hard to believe it's only been two years.

Still a long way to go to build the future of software engineering - we are hard at work every day on making Devin even better! If you haven’t tried it in a bit, check it out and let us know what you think: devin.ai

Cognition@cognition

1/ We’ve raised over $1B at a $26B valuation, led by @Lux_Capital, @generalcatalyst, and @8vc.

Our enterprise usage has grown >10x since the start of this year, and our run-rate revenue grew to $492 M.

We launched Devin two years ago as the first AI software engineer. Since then, cloud agents have gone from niche to mainstream, and today they are the fastest growing way to create software.

🔗 View Quoted Tweet

💬43🔄41❤️391👀65180📊81

⚡ Powered by xgo.ing

I beat Minesweeper in 77 seconds! Devin now supports Windows VMs https://t.co/zVDGMjAox9 #DevinGot...

Scott Wu (@ScottWu46) — Thu, 21 May 2026 17:33:09 GMT

I beat Minesweeper in 77 seconds!

Devin now supports Windows VMs
devin.ai/windows/minesw…#DevinGotAWindowsPC

💬6🔄6❤️112👀11479📊24

⚡ Powered by xgo.ing

grateful for our deep partnership with the whole Anthropic team! seems i will never get to live down...

Scott Wu (@ScottWu46) — Tue, 19 May 2026 20:20:45 GMT

grateful for our deep partnership with the whole Anthropic team! seems i will never get to live down my middle school math videos though

Claude@claudeai

Scott Wu (@ScottWu46) runs @cognition, the team behind Devin, an AI software engineer built on Claude. 

He wants to make building software 10x faster for every engineering team:

Your browser does not support the video tag.

🔗 View on Twitter

🔗 View Quoted Tweet

💬16🔄20❤️582👀75683📊79

⚡ Powered by xgo.ing

thank you @colossusmag and thank you to @JeremySternLA for coming to visit the Cog office! to give ...

Scott Wu (@ScottWu46) — Tue, 12 May 2026 06:15:01 GMT

thank you @colossusmag and thank you to @JeremySternLA for coming to visit the Cog office!

to give my side of what seems to be the most controversial story of the article: I went to a Founders Fund game night and Mango showed up because Brian from FF is also on the board of Cloud9. I have played a lot of melee (probably my most played game after, like, competitive programming) so was pretty excited to get to play him.

the setup was mega laggy but we played for a while. I took one game (he played falcon and wasn't trying that hard). would not use the phrase "beat him to a pulp," i'd obviously get washed in a real tournament set lol.

not sure how Jeremy even heard this story but seems his sources run deep :)

Colossus@colossusmag

Scott Wu is the co-founder of Cognition AI, one of the fastest-growing companies in history. He’s also the greatest competitive programmer the US has ever produced. You may have seen him doing impossible card tricks and mental math.

You’ve never seen him asked about weed, Michael Jordan, cancer, and human consciousness over a punnet of strawberries. That is what Colossus editor-in-chief Jeremy Stern did on a recent visit to San Francisco.

For those less familiar with @ScottWu46: In 2nd grade, he entered a math competition for 7th graders, lost, and was so furious he still fumes about it 20 years later. The next year he entered the 9th-grade division as a 3rd-grader and got a perfect score.

Then he won first place at the US national middle-school math competition and three straight gold medals at the International Olympiad in Informatics, where he became the greatest American gold-medalist and coach in history.

Most of the people running the biggest AI companies met as teenagers, competing for their countries on international math and science teams. OpenAI’s Greg Brockman, Anthropic’s Dario Amodei, Meta’s Alexandr Wang, to name just a few.

Most agree that the von Neumann among them was Scott Wu.

In November 2023, a few weeks after his mother died of lung cancer, on the day Sam Altman was fired from OpenAI, Wu founded his own AI company: Cognition.

He was 26 and saw earlier than almost anyone that AI would converge on agents that work in the background, 24/7, like coworkers. He shipped Cognition’s AI software engineer Devin in March 2024. It worked poorly, and he took intense public criticism for it.

Now, in its first 18 months of service, Devin has generated $445 million of revenue run rate and usage has doubled every eight weeks. The US Army, Goldman Sachs, and Mercedes-Benz are all customers. Cognition is raising at a valuation around $25 billion.

@JeremySternLA sat down with Wu, the emperor of the nerds, to ask the questions we’d all ask one of the smartest people in America—building the most consequential technology of our generation—if we ever got the chance.

As well as MJ and weed, they talk about the cluster of competitive math prodigies behind so much of AI, what makes us human when AGI arrives, and why Wu believes he was put on this earth to teach AI how to code.

Read the piece below.

🔗 View Quoted Tweet

💬34🔄14❤️1091👀171929📊139

⚡ Powered by xgo.ing

Katie has always had a unique way of seeing around corners - would highly recommend working with her...

Scott Wu (@ScottWu46) — Mon, 04 May 2026 17:57:54 GMT

Katie has always had a unique way of seeing around corners - would highly recommend working with her! Congrats @katie_haun :)

Katie Haun@katie_haun

x.com/i/article/2050…

🔗 View Quoted Tweet

💬7🔄2❤️228👀42585📊39

⚡ Powered by xgo.ing

honored to work with you, @imjaredz!

Scott Wu (@ScottWu46) — Thu, 30 Apr 2026 18:01:16 GMT

honored to work with you, @imjaredz!

Jared Zoneraich@imjaredz

Huge update.. I joined @Cognition

I’m so incredibly proud of what we built at @PromptLayer. Although it feels a little surreal, I know the team is left in amazing hands. I’ll be staying on as an advisor and can’t wait to see where they take it.

We started PromptLayer four years ago, and we practically coined the term “prompt engineering”. ChatGPT had come out weeks prior. Nobody knew exactly how LLMs would evolve, but we did know it will completely change the way we build.

We were the first developer platform for this new type of builder.

Cognition feels like an extension of this vision. Devin is how this new builder will imagine, invent, and create. We're entering the age of software abundance, to steal a term from @scottwu46

Everything is Coding Agents. It’s the single biggest problem in AI. Cognition is one of the fastest growing companies ever.

But I think the team is what really sold it for me- it's packed with so many former founders

Excited to share more soon...

Your browser does not support the video tag.

🔗 View on Twitter

🔗 View Quoted Tweet

💬4🔄8❤️120👀19727📊17

⚡ Powered by xgo.ing

Katrin, Ola, and the whole @MercedesBenz team have been incredible partners already - excited to do...

Scott Wu (@ScottWu46) — Mon, 27 Apr 2026 18:02:15 GMT

Katrin, Ola, and the whole @MercedesBenz  team have been incredible partners already - excited to do much more together. Thanks for stopping by the office!

Cognition@cognition

Cognition is partnering with @MercedesBenz to accelerate software engineering across their global engineering teams, representing one of the most extensive deployments of AI software engineering in the automotive industry to date.

@ScottWu46 sat down with Katrin Lehmann, Mercedes-Benz CIO, to discuss the work:

Your browser does not support the video tag.

🔗 View on Twitter

🔗 View Quoted Tweet

💬4🔄5❤️208👀25543📊24

⚡ Powered by xgo.ing

great deep dive from @walden_yan!

Scott Wu (@ScottWu46) — Wed, 22 Apr 2026 21:04:58 GMT

great deep dive from @walden_yan!

Walden@walden_yan

x.com/i/article/2046…

🔗 View Quoted Tweet

💬7🔄30❤️508👀139573📊177

⚡ Powered by xgo.ing

here we go again

Scott Wu (@ScottWu46) — Tue, 21 Apr 2026 21:31:55 GMT

here we go again

💬44🔄45❤️1340👀570742📊214

⚡ Powered by xgo.ing

Craziest part is we all knew each other already in high school! Along with @randomjohnnyh (Perplexit...

Scott Wu (@ScottWu46) — Tue, 14 Apr 2026 06:17:10 GMT

Craziest part is we all knew each other already in high school! Along with @randomjohnnyh (Perplexity cofounder), @demi_guo_ (Pika CEO), @stevenkplus1 and Andrew (Cognition), and many others. We all grew up in different states but met thru the olympiad scene.

Vividly remember this line from @alexandr_wang when we were around 19: "I hear people saying they want to find the next Paypal mafia. Why shouldn't it just be us?"

Glad to see @chameleon_jeff get the recognition he deserves :)

Tanay Jaipuria@tanayj

HRT’s first ever intern class of 10 included:

• Jesse Zhang, cofounder/CEO of Decagon
• Alexandr Wang, cofounder/CEO of Scale AI
• Scott Wu, cofounder/CEO of Cognition
• Jeffrey Yan, founder/CEO of Hyperliquid

Insane!⁠

🔗 View Quoted Tweet

💬57🔄136❤️2669👀434120📊462

⚡ Powered by xgo.ing

h/t @alexatallah :)

Scott Wu (@ScottWu46) — Wed, 08 Apr 2026 23:18:31 GMT

h/t @alexatallah :)

💬0🔄0❤️16👀9233📊2

⚡ Powered by xgo.ing

btw you can see this effect live on OpenRouter: total # tokens has gone from 1.78T / wk one year ag...

Scott Wu (@ScottWu46) — Wed, 08 Apr 2026 23:17:27 GMT

btw you can see this effect live on OpenRouter:

total # tokens has gone from 1.78T / wk one year ago to 27T / wk today (15.2x).

but % usage of the frontier / most expensive model has gone from 22% one year ago (Sonnet 3.7) to just 4% today (Opus 4.6).

economics works!

Scott Wu@ScottWu46

Total amt of flops across all the GPUs in the world has grown about 3x per year for the last few years. Total amt of inference demand has probably grown ~10x per year. What happens when those lines cross?

The econ answer is: when demand > supply, price goes up. That might be true (even H100s are more expensive than they have ever been...) but doesn't actually solve the problem on its own here - demand continues to grow as we unlock new use cases and there is only so much additional supply coming.

To get to a healthy equilibrium, we also need to shift much more usage to smaller, targeted models. This will happen naturally as the incentives make sense for it: it's much easier to 10x your agent usage if you know that you can now solve 90% of your tasks with good cheap, fast models.

SWE 1.6 is not a general model. But we have trained it to specifically be good at the kinds of standard coding tasks that our users run. And thanks to its small size & a little magic from Cerebras it can run very cheaply at ~1000 tokens / sec. Give it a try and let us know what you think!

🔗 View Quoted Tweet

💬37🔄70❤️861👀189299📊185

⚡ Powered by xgo.ing

Total amt of flops across all the GPUs in the world has grown about 3x per year for the last few yea...

Scott Wu (@ScottWu46) — Wed, 08 Apr 2026 06:36:33 GMT

Total amt of flops across all the GPUs in the world has grown about 3x per year for the last few years. Total amt of inference demand has probably grown ~10x per year. What happens when those lines cross?

The econ answer is: when demand > supply, price goes up. That might be true (even H100s are more expensive than they have ever been...) but doesn't actually solve the problem on its own here - demand continues to grow as we unlock new use cases and there is only so much additional supply coming.

To get to a healthy equilibrium, we also need to shift much more usage to smaller, targeted models. This will happen naturally as the incentives make sense for it: it's much easier to 10x your agent usage if you know that you can now solve 90% of your tasks with good cheap, fast models.

SWE 1.6 is not a general model. But we have trained it to specifically be good at the kinds of standard coding tasks that our users run. And thanks to its small size & a little magic from Cerebras it can run very cheaply at ~1000 tokens / sec. Give it a try and let us know what you think!

💬18🔄31❤️394👀243049📊67

⚡ Powered by xgo.ing

lucky to get to work with so many great ex-founders and so many great future-founders!

Scott Wu (@ScottWu46) — Tue, 07 Apr 2026 18:13:56 GMT

lucky to get to work with so many great ex-founders and so many great future-founders!

8VC@8vc

There's a reason @cognition keeps winning deals against companies 10x their size.

@vvkgopalan sat down with @theodormarcu, @bryceehunt and @moritz_stephan (all former founders!) to discuss hiring philosophy, infinity stories, enterprise coding agents, and what it actually takes to compete.

(00:00) Introduction

(03:45) Cognition team & hiring philosophy

(06:00) Infinity stories

(13:15) Product surface area and future goals

(17:20) Activation energy, experimentation, and customer-driven innovation

(19:00) Adoption, operational changes, and ownership

(22:05) Vibe coding hot takes

(24:15) AI-first engineering organizations and managing AI agents

(28:40) Future of software engineering roles and misconceptions about Devin

(30:00) Model development, learnings, and Cognition's journey in vertical integration

(38:10) What should you do as a new grad?

Your browser does not support the video tag.

🔗 View on Twitter

🔗 View Quoted Tweet

💬2🔄8❤️115👀17934📊17

⚡ Powered by xgo.ing

Devin Review caught the axios supply chain attack for multiple Cognition customers before the attack...

Scott Wu (@ScottWu46) — Tue, 31 Mar 2026 06:26:13 GMT

Devin Review caught the axios supply chain attack for multiple Cognition customers before the attack was publicly known.

These attacks will be 10x more frequent in the age of AI; it is critical that repo maintainers start using AI for defense as well.

(showing one example below where Devin Review caught the attack within an hour of its release - text minorly edited for anonymization)

💬95🔄145❤️1655👀326019📊404

⚡ Powered by xgo.ing

AW is goated :) such a pleasure to work with @rick_gerson and the whole team!

Scott Wu (@ScottWu46) — Sun, 29 Mar 2026 17:06:08 GMT

AW is goated :) such a pleasure to work with @rick_gerson and the whole team!

Abhinav Kukreja@kukreja_abhinav

maybe the most insane (appreciative) barbell portfolio of all time. respect.

🔗 View Quoted Tweet

💬2🔄3❤️112👀22222📊15

⚡ Powered by xgo.ing

new podcast with @JTLonsdale! a real shame the part of us playing Ricochet Robots got cut - guess we...

Scott Wu (@ScottWu46) — Fri, 27 Mar 2026 17:20:46 GMT

new podcast with @JTLonsdale! a real shame the part of us playing Ricochet Robots got cut - guess we'll never know who won

American Optimist@AmOptimistShow

EPISODE 147: The New Era of Software Abundance

@JTLonsdale visits @ScottWu46 & @russelljkaplan at @cognition HQ

00 Episode intro

35 Why technical talent & execution matters in AI

10 Do young people have an edge in the AI era?

26 Cognition's rapid growth

55 The new era of software abundance

30 Cognition engineers don't type code anymore

20 "Never sleep while Devin is idling"

25 The case for AI disinflation

50 How Devin generates 12X productivity gains

25 Cognition for government / taking on complex, broken systems

40 The AI race / competition with Anthropic

00 Forward deployed engineers?

40 How fast are LLMs improving?

10 The AI-led small business explosion

Your browser does not support the video tag.

🔗 View on Twitter

🔗 View Quoted Tweet

💬5🔄11❤️92👀42312📊20

⚡ Powered by xgo.ing

Lot has changed in the last few months. Try out Devin if you haven't already - it's free to get star...

Scott Wu (@ScottWu46) — Sun, 22 Mar 2026 17:30:29 GMT

Lot has changed in the last few months. Try out Devin if you haven't already - it's free to get started!

Ryan Carson@ryancarson

Paid $500 for @DevinAI - liking it so far.

You can tell this team is much further than other agent labs when it comes to being truly remote-first.

Very mature, advanced tooling and it just works across all surfaces (iPhone, Slack, Browser, GitHub, Linear).

I was tired of trying to hand-connect everything with a custom setup of Open Inspect + Codex + Linear.

When you look at your hourly effective rate, it stops making sense trying to hand-build all this stuff.

Nvm the all the maintenance hours you need to put in.

I'll keep using Devin 100% for the next week and report back.

So far, my PR shipping velocity is higher than before - so that's good obv.

🔗 View Quoted Tweet

💬9🔄8❤️180👀23728📊26

⚡ Powered by xgo.ing

This week 70% of all Devins were started by humans (webapp, slack, linear) and 30% were started auto...

Scott Wu (@ScottWu46) — Fri, 20 Mar 2026 21:28:16 GMT

This week 70% of all Devins were started by humans (webapp, slack, linear) and 30% were started automatically (API, and now scheduled + managed Devins)

In a few months that probably flips to 30/70 the other way and within a year it'll be 10/90. What does it look like to run a truly agent-native dev team?

-Handing whole project specs to a top-level agent that manages its subagents to do the groundwork
-Agents that kick off automatically on Sentry/Datadog alerts and as first-line incident-response
-Agents continuously running integration test / QA workflows and auto-investigating UX changes or performance issues

And of course, all of the "eng scaffolding" needed to make this possible - comprehensive unit testing, good documentation, and a reproducible dev environment.

Cognition@cognition

Devin can now schedule itself.

Run any task once, like feature flag cleanup, release notes, or QA. Then tell Devin to make it recurring, so that one good session becomes an automated workflow.

Available now for all users.

Your browser does not support the video tag.

🔗 View on Twitter

🔗 View Quoted Tweet

💬12🔄20❤️236👀45188📊49

⚡ Powered by xgo.ing

Interesting stat - our enterprise customers have already done more Devin sessions (and more merged D...

Scott Wu (@ScottWu46) — Wed, 11 Mar 2026 22:41:02 GMT

Interesting stat - our enterprise customers have already done more Devin sessions (and more merged Devin PRs) in 2026 than in all of 2025. Not bad for 2-ish months into the year!

💬19🔄23❤️444👀267714📊81

⚡ Powered by xgo.ing

Early days but a big milestone for us! This model is still in preview and we expect to tune behavior...

Scott Wu (@ScottWu46) — Sun, 01 Mar 2026 22:34:00 GMT

Early days but a big milestone for us! This model is still in preview and we expect to tune behavior a lot over the coming weeks - but wanted to get folks a snapshot as soon as we could.

Cognition@cognition

We are sharing an early preview of our ongoing SWE-1.6 training run.

It significantly improves upon SWE-1.5 while being post-trained on the same pre-trained model - and it runs equally as fast at 950 tok/s. On SWE-Bench Pro it exceeds top open-source models.

The preview model still exhibits some undesirable behaviors like overthinking and excessive self-verification, which we aim to improve. We are rolling out early access to a small subset of users in Windsurf.

🔗 View Quoted Tweet

💬15🔄14❤️249👀29838📊40

⚡ Powered by xgo.ing

https://t.co/oBh3gfnhCw

Scott Wu (@ScottWu46) — Tue, 24 Feb 2026 17:38:10 GMT

x.com/i/article/2026…

💬49🔄101❤️1098👀640626📊337

⚡ Powered by xgo.ing

1-2 yrs ago the story was always "AI-for-code works pretty well so far but soon AI-for-every-other-v...

Scott Wu (@ScottWu46) — Mon, 23 Feb 2026 20:33:05 GMT

1-2 yrs ago the story was always "AI-for-code works pretty well so far but soon AI-for-every-other-vertical will catch up"

Instead, code just got another 100x bigger...

💬43🔄15❤️578👀137839📊99

⚡ Powered by xgo.ing

The breakthrough of "Attention is all you need" was translating English to German. Makes sense that ...

Scott Wu (@ScottWu46) — Tue, 17 Feb 2026 08:54:31 GMT

The breakthrough of "Attention is all you need" was translating English to German. Makes sense that the models would be good at porting C to Rust too!

Andrej Karpathy@karpathy

I think it must be a very interesting time to be in programming languages and formal methods because LLMs change the whole constraints landscape of software completely. Hints of this can already be seen, e.g. in the rising momentum behind porting C to Rust or the growing interest in upgrading legacy code bases in COBOL or etc. In particular, LLMs are *especially* good at translation compared to de-novo generation because 1) the original code base acts as a kind of highly detailed prompt, and 2) as a reference to write concrete tests with respect to. That said, even Rust is nowhere near optimal for LLMs as a target language. What kind of language is optimal? What concessions (if any) are still carved out for humans? Incredibly interesting new questions and opportunities. It feels likely that we'll end up re-writing large fractions of all software ever written many times over.

🔗 View Quoted Tweet

💬15🔄24❤️636👀88323📊94

⚡ Powered by xgo.ing

honestly once LLMs started beating us at the aime the rest was kind of obvious

Scott Wu (@ScottWu46) — Fri, 13 Feb 2026 19:36:20 GMT

honestly once LLMs started beating us at the aime the rest was kind of obvious

💬44🔄33❤️1064👀159766📊184

⚡ Powered by xgo.ing

Think of software engineering as a pipeline. Historically, the hardest part was writing the code its...

Scott Wu (@ScottWu46) — Wed, 11 Feb 2026 23:20:43 GMT

Think of software engineering as a pipeline. Historically, the hardest part was writing the code itself. Not the case anymore - honest question: how much of your code do you actually have to type at this point? For us it's probably <10%.

As agents get more capable, the bottleneck becomes less about writing code and more about two things: 1) making it easier for humans to understand, plan, and ask questions. 2) making it easier for agents to ingest all the real-world context surrounding your task.

Hence, DeepWiki & DeepWiki MCP. There's a "nihilist" view that as AI gets better, interfaces won't matter anymore. I think the opposite is true - AI will soon be so good that the way you interact & knowledge-transfer will be the only thing that matters.

Thanks @karpathy for the shoutout!

Andrej Karpathy@karpathy

On DeepWiki and increasing malleability of software.

This starts as partially a post on appreciation to DeepWiki, which I routinely find very useful and I think more people would find useful to know about. I went through a few iterations of use:

Their first feature was that it auto-builds wiki pages for github repos (e.g. nanochat here) with quick Q&A:
deepwiki.com/karpathy/nanoc…
Just swap "github" to "deepwiki" in the URL for any repo and you can instantly Q&A against it. For example, yesterday I was curious about "how does torchao implement fp8 training?". I find that in *many* cases, library docs can be spotty and outdated and bad, but directly asking questions to the code via DeepWiki works very well. The code is the source of truth and LLMs are increasingly able to understand it.

But then I realized that in many cases it's even a lot more powerful not being the direct (human) consumer of this information/functionality, but giving your agent access to DeepWiki via MCP. So e.g. yesterday I faced some annoyances with using torchao library for fp8 training and I had the suspicion that the whole thing really shouldn't be that complicated (wait shouldn't this be a Function like Linear except with a few extra casts and 3 calls to torch._scaled_mm?) so I tried:

"Use DeepWiki MCP and Github CLI to look at how torchao implements fp8 training. Is it possible to 'rip out' the functionality? Implement nanochat/fp8.py that has identical API but is fully self-contained"

Claude went off for 5 minutes and came back with 150 lines of clean code that worked out of the box, with tests proving equivalent results, which allowed me to delete torchao as repo dependency, and for some reason I still don't fully understand (I think it has to do with internals of torch compile) - this simple version runs 3% faster. The agent also found a lot of tiny implementation details that actually do matter, that I may have naively missed otherwise and that would have been very hard for maintainers to keep docs about. Tricks around numerics, dtypes, autocast, meta device, torch compile interactions so I learned a lot from the process too. So this is now the default fp8 training implementation for nanochat
github.com/karpathy/nanoc…

Anyway TLDR I find this combo of DeepWiki MCP + GitHub CLI is quite powerful to "rip out" any specific functionality from any github repo and target it for the very specific use case that you have in mind, and it actually kind of works now in some cases. Maybe you don't download, configure and take dependency on a giant monolithic library, maybe you point your agent at it and rip out the exact part you need. Maybe this informs how we write software more generally to actively encourage this workflow - e.g. building more "bacterial code", code that is less tangled, more self-contained, more dependency-free, more stateless, much easier to rip out from the repo (x.com/karpathy/statu…) 
There's obvious downsides and risks to this, but it is fundamentally a new option that was not possible or economical before (it would have cost too much time) but now with agents, it is. Software might become a lot more fluid and malleable. "Libraries are over, LLMs are the new compiler" :). And does your project really need its 100MB of dependencies?

🔗 View Quoted Tweet

💬100🔄105❤️1730👀408764📊449