Real-World AI. Where AI actually pays off. 10 yrs @ Google | 300+ companies.
Truckee, CA
Joined January 2011
- Tweets2.6K
- Following1.7K
- Followers1.6K
- Likes1.2K
Scott Penberthy retweeted
People who cured / are curing their own cancer on X:
1/ @sytses - bone cancer, billionaire founder of Gitlab, developed 10+ personalized therapies for himself.
2/ @paul_conyngham - shrank his dog Rosie's mast cell cancer tumors after chemo and surgery weren't enough. Now backed by YC and Founders Fund
3/ @miriamgonp - stage 4 breast cancer, tumor sequenced, searching for vaccine manufacturer now
4/ @kevinace - stage 4 lung cancer, angel investor in 300+ companies, first drug stopped working after 18 months. Vaccine designed, now paying $300k out of pocket for the vaccine
5/ @masonbgrant: going founder mode on his his 32-year-old wife's rare liver cancer, sequencing her tumor with AI agents
6/ @ArunRLverma: former product leader at Atoms and Shopify, brain tumor, built a vaccine from his tumor. Now building an AI health agent startup based on what he went through
We're going to make it
Scott Penberthy retweeted
I spent 48 hours in SF hanging out with billionaires, data labs and AI founders and execs running $1B to $100B companies.
My observations:
1. Sooo MANY stories of data brokers making like $2M a WEEK selling company data to labs. I can't tell if it's the beginning or a bubble, and I kept wondering which niches are left. Digging into it actively.
2. At dinner, someone asked the table, "If a new model drops next month, does your company get better or worse?" HALF the table went quiet! I think that's the best question to ask about any AI startup right now.
3. The bar people talk about now is hitting $1B in revenue in 12 months (I guess that's the new unicorn?). What I didn't hear much about is retention. AI revenue can come in fast and leave fast, and I think we're going to see a lot of stories of AI companies with huge revenue and terrible retention.
4. It's pretty clear we're living through the reinvention of the computer. The model is the CPU, the harness is the operating system, and skills and plugins are the apps. Thinking about it that way makes it easier to see where the whitespace is.
5. We're maybe 12 months from an agent that can watch you work for a week and then do your job without any instructions. The screen recording plus agent memory plus local model combination makes this possible right now.
6. Agent harnesses feel like the new GPT wrappers, in a good way, especially vertical ones. The best teams I talked to treat the model as swappable. One founder walked me through routing different models to different jobs: one for taste and visuals, a cheaper one for the everyday stuff, another for text.
They use prompts they can move across models over fine-tuning, and A/B test the output on what users actually do with it: do they edit it, export it, share it? A vertical harness is basically four things: the connectors for that industry's data, evals written by people who know the work, rules for what the agent can and can't touch, and a clean screen for a human to review. The model is maybe 20% of it.
7. There's a lot of reverse engineering going on/cloning going on more than I've ever seen. People are decompiling software and rebuilding it with agents, and someone showed a suite of like 10 apps that are worth $5B+ each all that he cloned open source. Me and everyone are still thinking through what it means when software gets this easy to copy.
8. My guess is the value moves to data. If the software can be rebuilt quickly, what's worth paying for in a legacy SaaS company is the years of customer data and workflows inside it. Some SaaS companies are secretly data companies.
Gets me thinking about how do you truly value data for this saas companies? Like if we were buying these dying saas companies at 2x ARR before (as example), maybe there is a better metric. Idk, just thinking out loud here.
9. It feels like a good chunk of legacy software won't make it to the agentic era, especially anything built around people clicking through a lot of screens.
10. Every company I talked to wants to pipe its internal data into one place and expose it to agents through MCP. Then they look at the data and realize it's a mess: five versions of the same customer, old Slack channels, PDFs nobody can search. Cleaning that up feels like a big consulting opportunity!
11. A lot of founders are building the basic building blocks for agents: APIs, data layers and tools. Some of the companies I saw barely have a UI because the main user is an agent. I really believe the startups that build primitives to this agentic era will do extremely well. Building blocks is the best way to think about it.
12. Labs can get plenty of training data, but it's harder to get real examples of agents failing inside companies. Whoever has those logs has something valuable.
13. Someone running a big consumer company told me the next fight is over memory. The model that remembers the most about you wins, and the smartest model matters less than people think.
14. Physical AI came up a lot. Astra I think was the first model that got people into AI can actually do CAD, and I think a lot of people are thinking what does vibe manufacturing really look like. The line I keep thinking about came from a robotics founder: the hardest problems in AI are deployment problems now, not research problems. Their robots work on their own, call a human when they get stuck, and every one of those calls becomes training data.
15. Cybersecurity companies are having a moment and are actually cool companies to build now, especially the ones that clean up after a breach. Every agent with access to your systems is a new door into your company, and a lot of those doors were left open.
16. Some startups are worth more dead than alive now. A shutdown company with 5 years of Slack, email and code can get paid for licensing all of it. I wouldn't be surprised if founders start factoring that into when they shut down.
17. People are starting to talk about monetizing the agent web. It feels really early, but when the most popular user of the internet is an agent, what startups should you be building?
18. I think agent insurance is coming. Once an agent makes an expensive mistake, insurers will want audits before they cover anyone.
19. Several people said the hire they want most is someone with taste who can write evals. The models are getting good at generating ideas, so the hard part is judging which ones are good.
20. New trend I observed: startup raises VC, uses venture dollars to become a creator, startup fails, founder falls back on being a full-time creator and incubates other products bootstrapped.
21. Getting SIGNIFICANT distribution is HARD right now. Founders are sponsoring buses, buying billboards like crazy and throwing parties with famous DJs to get noticed. Kinda feels dotcom era in that sense.
Attention is what's scarce! The growth guy going viral sharing his UGC strategy from Wispr Flow is a good example. It's the sauce everyone wants, and I truly believe there are probably only 50 people on the planet who know how to scale these types of businesses efficiently in this era.
Always good to get that dose of SF. I felt sooo lucky to have such an incredible trip during such an incredible time in building businesses. Surreal in so many ways.
Happy to be home and locked in.
Until next time!
Scott Penberthy retweeted
Indeed, it has been quite a week, so I thought I'd share it with the KDD 2026 audience! For my Google colleagues who might have sent me internal chat messages after I closed my laptop for a final time on Thursday afternoon, I apologize for my lack of response!
(Zoomed in)
Scott Penberthy retweeted
As engineering, product, design, DS, etc. melt into a new kind of role, I was reflecting on what roles might look like in the future. For example, when I look at the Claude Code team I see what I think is five archetypes:
1. Prototyper: comes up with brand new ideas; churns out many ideas, most of which don't ship
2. Builder: quickly turns a prototype/idea into production-grade product/infra
3. Sweeper: cleans up the UI, simplifies the code and system, unships, optimizes performance
4. Grower: takes a product that has been built and iterates on it to improve Product-Market Fit
5. Maintainer: owns a mature system to make it secure, reliable, fast, and efficient as it scales
Many people span across 2 roles, and sometimes 3 roles. I also notice that these roles are not really tied to job function -- eg. across Anthropic, some designers match category 1, some 2, some 3; same for engineers, PM, DS.
A healthy team needs a mix of these, depending on the product:
- A product that is new and pre-PMF needs people that are strong at 1+2+3
- A product that is growing and has found PMF needs 2+3+4 and some 5
- A product that has strong PMF needs 3+4+5 and some 2
Maybe product roles of the future will look more like this, and less like the domain-specific roles of today?
Scott Penberthy retweeted
China just released an open source AI model that matches the best closed models from OpenAI and Anthropic. Gavin Baker explained exactly how they did it and the answer should concern every American AI lab.
The model is called GLM 5.2. It was built by Z. AI.
You get 744 billion parameters, 1 million token context window and its MIT license, meaning anyone can download it, fork it, build a company on it, with no restrictions and no Dario.
It scored 51 points on the artificial analysis intelligence index. The highest score any open weight model has ever achieved.
It beat GPT 5.5 on the frontier software engineering benchmark. It trails Claude Opus 4.8 by less than one percentage point. And it costs 85% less to run than GPT 5.5 for comparable performance.
Gavin Baker said on the All-In podcast that this model has challenged some of his beliefs. Then he explained how China built it.
The method is called distillation.
Just think of tens of thousands of phones and computers running simultaneously, all hitting the frontier model APIs through masked accounts, asking specific questions, and harvesting what happens inside the model when it answers. Every reasoning step, every token. The entire thinking process gets recorded and fed back into the Chinese model during training.
It is a cheat sheet. It is the answer key to the exam.
And here is the part that should worry everyone.
Sacks said it plainly. China was already nine months behind American models. But now that GLM 5.2 is good enough to run its own reinforcement learning, it can improve itself without needing to distill from American models anymore. The cheat sheet let them get close enough to start writing their own answers.
Sacks said we are six months behind on the model and 24 months behind on silicon and they are only a few months behind in total.
The Z. AI founder told Elon Musk directly that open weight fable-level capability will be here before Q1 2027.
Every restriction Anthropic lobbied for, every self-imposed safety guardrail, every month of delay in releasing American frontier models accelerated this. The Chinese labs were not under those restrictions. They were not going to wait.
The composable model future Gavin described, where every enterprise runs a frontier model alongside their own fine-tuned open weight model, is coming regardless of what American labs do next.
The question is just whether the open weight half of that stack is American or Chinese.
Right now it is Chinese.
WATCH THE FULL PODCAST ON @theallinpod
Scott Penberthy retweeted
100 MW data center in space is roughly 79 soccer fields of solar panels + 35 soccer fields of radiators
Total = 114 soccer fields
Now let’s scale that to 100 terawatts (Elon’s Musk compensation package goal)
100 TW = 100,000,000 MW
That’s 1,000,000x larger than a 100 MW data center
Solar panels: 79 million soccer fields
Radiators: 35 million soccer fields
Total: 114 million soccer fields
To make this more tangible imagine this:
A FIFA soccer field is ~7,140 m^2.
114 million fields = ~814 billion m^2.
That’s 814,000 km^2 👀
You know what land is this big on earth?
Texas: ~695,000 km^2
That would be building a data centre larger than Texas in space!
Turkey: ~783,000 km^2 -> just slightly larger
France: ~551,000 km^2-> ~1.5× France
Germany: ~357,000 km^2 -> ~2.3× Germany
This is truly an insane bold goal
The only company capable to achieve this is probably SpaceX, I hope they do
But I can’t see this happening within 10 years..
It took SpaceX approximately 14-15 years from its founding to successfully land a rocket and achieve reusability
I think achieving space compute at that scale would take roughly +20 years
While navigating all the politics of people realizing the scale on earth and not being happy about it
I did the math of the speed premium in my post 👇
xitter.cf/richardsondx/status/20…
Scott Penberthy retweeted
MIT just made every AI company's billion dollar bet look embarrassing.
They solved AI memory. Not by building a bigger brain. By teaching it how to read.
The paper dropped on December 31, 2025. Three MIT CSAIL researchers. One idea so obvious it hurts. And a result that makes five years of context window arms racing look like the wrong war entirely.
Here is the problem nobody solved.
Every AI model on the planet has a hard ceiling. A context window. The maximum amount of text it can hold in working memory at once. Cross that line and something ugly happens — something researchers have a clinical name for.
Context rot.
The more you pack into an AI's context, the worse it performs on everything already inside it. Facts blur. Information buried in the middle vanishes. The model does not become more capable as you feed it more. It becomes more confused. You give it your entire codebase and it forgets what it read three files ago. You hand it a 500-page legal document and it loses the clause from page 12 by the time it reaches page 400.
So the industry built a workaround. RAG. Retrieval Augmented Generation. Chop the document into chunks. Store them in a database. Retrieve the relevant ones when needed.
It was always a compromise dressed up as a solution.
The retriever guesses which chunks matter before the AI has read anything. If it guesses wrong — and it does, constantly — the AI never sees the information it needed. The act of chunking destroys every relationship between distant paragraphs. The full picture gets shredded into fragments that the AI then tries to reassemble blindfolded.
Two bad options. One broken industry. Three MIT researchers and a deadline of December 31st.
Here is what they built.
Stop putting the document in the AI's memory at all.
That is the entire idea. That is the breakthrough. Store the document as a Python variable outside the AI's context window entirely. Tell the AI the variable exists and how big it is. Then get out of the way.
When you ask a question, the AI does not try to remember anything. It behaves like a human expert dropped into a library with a computer. It writes code. It searches the document with regular expressions. It slices to the exact section it needs. It scans the structure. It navigates. It finds precisely what is relevant and pulls only that into its active window.
Then it does something that makes this recursive.
When the AI finds relevant material, it spawns smaller sub-AI instances to read and analyze those sections in parallel. Each one focused. Each one fast. Each one reporting back. The root AI synthesizes everything and produces an answer.
No summarization. No deletion. No information loss. No decay. Every byte of the original document remains intact, accessible, and queryable for as long as you need it.
Now here are the numbers.
Standard frontier models on the hardest long-context reasoning benchmarks: scores near zero. Complete collapse. GPT-5 on a benchmark requiring it to track complex code history beyond 75,000 tokens — could not solve even 10% of problems.
RLMs on the same benchmarks: solved them. Dramatically. Double-digit percentage gains over every alternative approach. Successfully handling inputs up to 10 million tokens — 100 times beyond a model's native context window.
Cost per query: comparable to or cheaper than standard massive context calls.
Read that again. One hundred times the context. Better answers. Same price.
The timeline of the arms race makes this sting harder. GPT-3 in 2020: 4,000 tokens. GPT-4: 32,000. Claude 3: 200,000. Gemini: 1 million. Gemini 2: 2 million. Every generation, every company, billions of dollars spent, all betting on the same assumption.
More context equals better performance.
MIT just proved that assumption was wrong the entire time.
Not slightly wrong. Fundamentally wrong. The entire premise of the last five years of context window research — that the solution to AI memory was a bigger window — was the wrong answer to the wrong question.
The right question was never how much can you force an AI to hold in its head.
It was whether you could teach an AI to know where to look.
A human expert handed a 10,000-page archive does not read all 10,000 pages before answering your question. They navigate. They search. They find the relevant section, read it deeply, and synthesize the answer.
RLMs are the first AI architecture that works the same way.
The code is open source. On GitHub right now. Free. No license fees. No API costs. Drop it in as a replacement for your existing LLM API calls and your application does not even notice the difference — except that it suddenly works on inputs it used to fail on entirely.
Prime Intellect — one of the leading AI research labs in the space — has already called RLMs a major research focus and described what comes next: teaching models to manage their own context through reinforcement learning, enabling agents to solve tasks spanning not hours, but weeks and months.
The context window wars are over.
MIT won them by walking away from the battlefield.
Source: Zhang, Kraska, Khattab · MIT CSAIL · arXiv:2512.24601
Paper: arxiv.org/abs/2512.24601
GitHub: github.com/alexzhang13/rlm
Scott Penberthy retweeted
Okay this is genuinely insane.
SpaceX just unveiled a satellite whose only job is to run AI. Not internet. Not GPS. Just compute, floating in orbit.
It's called AI1, and the reason behind it breaks your brain.
AI data centers on Earth are hitting a wall, not a chip wall, a physics wall.
They need staggering amounts of power and water just to stay cool, and we're running out of grid and land to build them.
So Musk's answer is: stop building them on Earth.
In orbit, the sun never sets. Free power, 24/7. No water for cooling, you just radiate heat into the vacuum of space. The two things choking AI on the ground barely exist up there.
And here's the wild part: Musk says it's easier to build than a Starlink satellite. Strip out the complex antennas and it's "a lot of solar cells, a radiator, and some laser links."
One AI1 carries the compute of an Nvidia GB300 rack, the same hardware data centers fight over down here.
AI1 is just the first one. The plan is a constellation of up to a million of them.
And the timing isn't an accident, SpaceX goes public this week at a ~$1.75 trillion target. This isn't a rocket company anymore. It's positioning itself as the power grid for AI, in space.
The race for AI compute just left the planet. Literally.
@SpaceX
Scott Penberthy retweeted
This is great to see. The scientific community & the general public should be in the habit of celebrating amazing scientific achievements with standing ovations 🎉
This is not simply a new pancreatic cancer drug.
It is a reminder that even “undruggable” biology can become treatable with persistence.
Daraxonrasib doubled median OS vs chemotherapy in RAS G12 metastatic pancreatic cancer: 13.2 vs 6.6 months.
A remarkable ASCO moment.
#ASCO26
@DrChoueiri @TiansterZhang @CathyEngMD @montypal @tompowles1 @brian_rini @cdanicas @GlopesMd @PGrivasMDPhD @nataliagandur @yekeduz_emre @neerajaiims @ASCO @ONCOassist @OpenMedicineHQ @MedwatchKate @scserendipity1 @CParkMD @urotoday @OncLive @crisbergerot @urologysummit @SuyogCancer @Larvol @IMG_Oncologists
Scott Penberthy retweeted
A team of former DeepMind researchers just raised $50M to build an AI lab built around recursive self-improvement at the level of the whole research organization, not only a single model.
Index and Radical co-led, NVIDIA's venture arm is in, and angels like Dwarkesh Patel, Thomas Wolf and Max Jaderberg are on the cap table.
The founders have the track record to back it up. Louis Kirsch comes out of the Schmidhuber lineage on self-improving systems. Edward Hughes has argued that open-endedness is essential for artificial superhuman intelligence. Tantum Collins worked on AI policy in the Biden White House.
Their idea is simple and big at the same time. Today's models are great at answering questions, but real discovery also depends on knowing which questions are worth asking. Inherent wants AI that works right next to humans inside that loop, as a collaborator and not only a tool. They call it living within the experiment.
They also set it up as a Public Benefit Corporation, so the mission is written into the company from day one.
This is the direction a lot of us have been hoping for, and one of the more credible attempts at recursive self-improvement I've seen so far. Really excited for it.
We’re excited to introduce Inherent, a lab designed from scratch to build AI agents that discover new knowledge.
The coming era of machine-driven scientific inquiry demands a new kind of research institution and a new kind of AI.
To achieve our mission, we live within the experiment, recursively self-improving the entire research organisation. We investigate questions including:
- What does ‘AI taste’ look like in the sciences, and how can we build an institution that embraces this new aesthetic of discovery?
- What new kinds of human-machine teaming will make the most of AI that can truly innovate?
- How can we build recursive self-improvement at the collective level that continually increases human agency over outcomes?
We have just closed a $50m seed round led by @IndexVentures and @radicalvcfund, with participation from other outstanding investors including NVentures (@nvidia's venture capital arm), @buildexante, Metaplanet, Macroscopic, @MythosVentures, Charlie Songhurst, @chalfs, @jluan, @dwarkesh_sp, @Thom_Wolf, @j_foerst and @maxjaderberg. We are advised by @matthewclifford.
Inherent is a Public Benefit Corporation headquartered in London.
Scott Penberthy retweeted
Launching today: make any PDF beautiful.
It's 2026 - there's no excuse to have ugly resumes, invoices or client proposals.
Just upload a PDF -> Get back a polished, professionally designed version in minutes.
Works with docs of any complexity👇
Scott Penberthy retweeted
the full no priors interview is one hour of karpathy in ai psychosis mode.
auto research, claws, parallel agents, token throughput as the new metric.
every section here has a frame worth stealing for how you operate in 2026.
watch the whole thing here:
youtube.com/watch?v=kwSVtQ7d…
Scott Penberthy retweeted
this is excellent
>GitLab founder diagnosed with rare cancer (osteosarcoma)
>standard care works but cancer comes back later
>medical team says there's not much else to do
>"It became my own job to keep myself alive. Nobody else was going to do it for me at this point"
>starts researching, assembles his own medical team, uses AI for deep research
>“I’ll talk to anyone, I’ll go anywhere, and I can be there anytime" to collect information
>does as many diagnostic tests as he can find as often as he can (maximal diagnostics)
>develops his own therapeutic ladder with repurposed drugs, personalized medicine, etc
>Sid’s cancer currently in remission
The full deck on Sid’s cancer approach is here:
sytse.com/cancer/
Worth a read. Raw data for download is also available and linked in the deck
Scott Penberthy retweeted
Agentic software engineering adoption is on fire at @Uber. 1,800 code changes per week are now written entirely by Uber's internal background coding agent, and 95% of our engineers now use AI every month across all the tools we track.
This is a real reset moment for engineering; it's one of the most exciting times to lead. This shift requires builders to be curious and hands-on. I’m incredibly lucky to be surrounded by a team that’s doing exactly that.
The best part is that the strongest adoption isn’t being pushed top down from leadership announcements; it’s coming from engineers who are quietly experimenting, quietly shipping, and quietly pushing things forward.
I love spending time with those engineers because there’s no substitute for being close to the work.
Over the last few months, we leaned in hard, and the results have been phenomenal.
The bigger shift: going agentic.
84% of AI users are now working with agent-style workflows, not just tab completion. Claude Code usage nearly doubled in 2 months (32% → 63%), while IDE-based tools have largely plateaued.
Engineers are moving from accepting suggestions to delegating tasks. Even within traditional IDEs, ~70% of committed code is now AI-generated.
Background agents are writing code autonomously.
Our internal background coding agent went from <1% of all code changes to 8% in just a few months. There is zero human authoring. Engineers review and approve, but the code is written entirely by AI agents.
The role of the engineer is shifting - from writing every line to architecting systems and reviewing AI-generated code.
More to come from the @UberEng team in the coming days.
Scott Penberthy retweeted
Holy shit...Someone built an AI system that takes a research idea and outputs a full academic paper. Real citations. Real experiments. Conference-ready LaTeX. Zero human input.
It's called AutoResearchClaw. And the pipeline is insane.
Here's what actually happens when you type one command:
It searches arXiv and Semantic Scholar for real papers. Not fake citations actual literature with 4-layer verification: arXiv ID check, CrossRef DOI lookup, Semantic Scholar title match, and LLM relevance scoring. Hallucinated references get killed automatically.
Then it designs and runs real experiments. Hardware-aware auto-detects whether you have NVIDIA CUDA, Apple MPS, or just CPU, and adapts the code accordingly. When experiments fail, it self-heals. When results don't support the hypothesis, it pivots to a new direction on its own.
Then it writes the paper. 5,000-6,500 words. Section by section. Multi-agent peer review with methodology-evidence consistency checks. Then it revises based on those reviews.
Then it outputs conference-ready LaTeX. NeurIPS, ICML, ICLR templates. Compile-ready for Overleaf. BibTeX references auto-pruned to match inline citations.
The whole thing runs across 23 stages and 8 phases. Three human-approval gates if you want them. Or just pass --auto-approve and walk away.
What you get back:
→ Full academic paper draft
→ Conference-ready LaTeX + BibTeX
→ Experiment code + sandbox results + charts
→ Peer review notes
→ Verification report on every citation
This is what autonomous scientific research actually looks like in 2026.
100% Opensource. MIT License.
Link in comments.
Scott Penberthy retweeted
Jensen Huang says every company will need an OpenClaw agentic system strategy by calling it “the new computer.”
He claims OpenClaw became the most popular open-source project in $NVDA history within weeks and comparing its impact to Linux reshaping the software stack.