helping founders @felicis | previously product @openai, @instagram, @uber, @facebook, @airtable

Joined March 2007
Peter Deng retweeted
Excited to announce that Jeff Pinner has joined Exa as CTO! Jeff was formerly CTO of Robinhood, a distinguished engineer at Cruise, CTO of Lyft, and way back a systems engineer at NASA. We have humongous engineering and research to do at larger-than-google-scale and we need to do it quickly. Pinner has scaled massive ML systems before, super excited to work together on this one.
41
16
7
751
406,682
What makes Ion special? He's deeply in the details. In the office every day, these days. Writing PRDs like an IC PM. Ion loves the game. Nothing/no-one is below him. Nothing is taken for granted. Here he is next to me, working on Arena, literally right now. A true 🐐
How do you turn research into a billion-dollar company? We asked Ion Stoica, who did it with Databricks (>$100B), Anyscale and Arena AI, while his Berkeley lab built Ray, vLLM, and SGLang. His signal that a project is ready: users start asking "who maintains this after the students graduate?" New episode of The Information Bottleneck. You can find it on our website, Youtube or the apps
2
10
61
7,367
Peter Deng retweeted
What began as a UC Berkeley research project has become one of the fastest-growing and most trusted AI infrastructure companies in the world. This week, the vision set forth by Arena founders @infwinston and @ml_angelopoulos reached another remarkable milestone: @Arena announced it has surpassed $100M in annualized revenue run rate just eight months after launching its enterprise offering, while growing to a community of more than 10 million users helping shape how frontier AI models are evaluated. In the @Felicis profile on the company, Wei-Lin and Anastasios shared how they approached building a platform rooted in scientific rigor, neutrality, and community trust; principles that have become even more important as AI capabilities accelerate. We've been true believers in Arena since Seed. felicis.link/lDzsSiy
2
15
1
49
7,912
Peter Deng retweeted
A new internet economy, built for agents: Agents break the old economics of the internet - when everyone has an AI assistant consuming and taking actions, what happens to ads? Exa Connect is a step toward a new internet economy built on data, not ads. It's a marketplace where data providers choose what to charge for their data, and developers choose which data providers their agents use. We feel this is the right economic model -- market-based, transparent, value-aligned, and as agent requests scale >1000x over the coming years, will be a more valuable market for providers than the ad market is today. This is also a good path forward for the internet. It's one that monetizes transformer attention, not human attention. We don't know exactly what that great intercontinental information exchange called the internet will become, but it's going to change, and we think very much for the better.
Introducing Exa Connect: connecting agents to data beyond the public web. Available today with ZoomInfo, Crunchbase, Similarweb, and many other leading data providers. exa.ai/connect
24
17
5
265
69,463
Peter Deng retweeted
Two in five home health patients get turned away, not because care isn't available, but because the admin overhead costs more than the agency gets paid. $40B in referrals are rejected every year. That's the problem @AlexRWendland and Ryan Tolsma are rebuilding from scratch with @joinAdaptive. AI-native operations. 100K+ visits delivered. 4.9% rehospitalization rate vs. 12.9% state average. @Felicis investors @Pxd and @EricFlaningam sat down with Alex and Ryan to discuss the future they're building with Adaptive.
3
19
1,275
Peter Deng retweeted
Two in every five patients referred to home health care in America never get the care they need. Not because clinicians aren't available. Not because the patient doesn't qualify. They get turned away because the paperwork costs more than the visit pays. Roughly $40B in referrals get rejected this way every year. So the agency says no, and the patient ends up in a nursing facility or gets sicker at home until the ER takes over. We led @joinadaptive's Series A because they stopped trying to sell better software to broken agencies and became the provider instead. AI runs the intake, scheduling, prior authorization, documentation, and billing that traditional agencies burn $.60 to $.90 of every clinical-labor dollar on. When those costs collapse, the patients who would have been rejected instead get the care that they deserve. So far in Adaptive's first market: Patients served by Adaptive have a 4.9% rehospitalization rate compared to the 12.9% Texas state average and 10.2% national average. More than 95% of patients see improvement in ambulation, meaning they are back on their feet, no longer homebound, compared to an 89.5% national average. Patients are set up with a home care provider extremely quickly, with 99% of patients admitted within 48 hours of referral. They have a 4.5-star CMS quality rating against an industry average of around 3. Hear from co-founders @AlexRWendland and @ryanatolsma on the future they're building with Adaptive. 👇 cc: @pxd @EricFlaningam ordnl.link/8R805uM
2
3
29
1,317
Peter Deng retweeted
1/ Today we're announcing Gigascale Capital's $250M first institutional fund to back early-stage founders rebuilding the physical economy.
67
62
31
822
188,954
Peter Deng retweeted
Industrial-scale science. Coming to a lab near you this summer.
69
73
31
1,296
252,979
Peter Deng retweeted
Naveen Gavini
@ngaviniApr 21
Gmail organizes your emails. @extradotemail organizes your life. Live today. extra.email
111
63
46
1,381
4,956,011
GPT-Image-2 literally broke the @arena chart. This is the largest gap we have ever seen in the history of arena IIRC
Exciting news - GPT-Image-2 by @OpenAI has claimed the #1 spot across all Image Arena leaderboards! A clean sweep with a record-breaking +242 point lead in Text-to-Image - the largest gap we’ve seen to date. - #1 Text-to-Image (1512), +242 over #2 (Nano-banana-2 with web-search aka gemini-3.1-flash-image) - #1 Single-Image Edit (1513), +125 over #2 (Nano-banana-pro aka gemini-3-pro-image) - #1 Multi-Image Edit (1464), +90 over #2 (Nano-banana-2) No model has dominated Image Arena with margins this wide. Huge congratulations to @OpenAI on this major breakthrough in image generation! More performance breakdowns by category in the thread below.
2
1
76
5,609
Peter Deng retweeted
EXCLUSIVE: Extra wants to manage your 'life's inbox' for you. The AI-powered consumer app launches publicly today with $9.5M. But can it challenge Gmail? I tested out Extra myself, spoke to users, and interviewed its ex-Pinterest founders led by CEO Naveen Gavini to find out.
6
10
2
33
7,334
Peter Deng retweeted
We hosted Prof. Alyosha Efros (UC Berkeley) at @SkildAI! He didn't believe that robots could actually cook eggs reliably. :) Tested back-to-back 5times without fail! One batch of scrambled eggs every ~2.5mins nonstop. The same model assembles a GPU on a server rack too.
42
156
27
1,362
182,985
Peter Deng retweeted
I don’t think we’ve fully internalized this. Yesterday a founder casually mentioned a single customer could be $250M/yr at full value. Services is already larger for them than their software even though it’s 4 yrs old vs 10 for software. TAM expansion is real.
2
2
13
2,596
Peter Deng retweeted
You can't trust AI you can't measure. @Arena is building the infrastructure to fix that - and we've believed since seed. New profile on Arena founders @ml_angelopoulos and @infwinston → ordnl.link/ywOwWNI
1
7
1
34
9,484
Founding a company is an act of courage. Today, we launch a fresh new look for @Felicis built on the core belief that has guided us from the beginning: Fortune favors the brave. 🔗 felicis.com
6
7
2
48
6,486
Peter Deng retweeted
Naveen Gavini
@ngaviniMar 17
1
11
3
34
2,032
Peter Deng retweeted
Excited to share that @deeptraceai has raised $5M to identify and fix every alert in production. This round is co-led by @felicis and @matrix, with participation from @ycombinator. As AI generates more of the world’s code, the real leverage shifts to systems that help engineers understand and run it in production. Deeptrace is that reliability layer. Today, teams like @opendoor, @mintlify, and @phantom use Deeptrace to automatically investigate every alert in production and proactively catch issues before they turn into incidents. If you’re looking to automate on-call and use agents to monitor prod, please reach out!
29
22
11
206
108,558
This release is fucking huge. It's one of the biggest updates to LMArena this year! Code Arena is our next generation of coding evaluations, beginning with web development tasks. Here you can use models to build interactive websites and share them with your friends. The links are persistent, so you can e.g. build a game and play it whenever you want. Here watch two models -- @claudeai Haiku and @grok-Code-Fast -- compete to build a galaxy. In this case, I liked the "star-wars" effect of Grok!
🚀Introducing Code Arena: the next generation of live coding evals for frontier AI models. Built to test how models plan, scaffold, debug, and build real web apps step-by-step. Try Claude, GPT-5, GLM-4.6 and Gemini in Code Arena today!
9
20
155
38,051
Peter Deng retweeted
At Salesforce, we built an early warning system to track customer health and predict churn. I’ve believed in that vision for 15+ years, and with AI, it’s finally real. Thrilled @felicis is backing @eliast @Agency as they announce their $20M Series A and launch Kai, an AI co-worker who knows every customer and never drops the ball. fortune.com/2025/11/12/elias…
1
4
16
2,072
Peter Deng retweeted
🚀 Introducing Arena Expert: a new LMArena evaluation framework to identify the toughest, most expert-level prompts from real users, powering a new Expert leaderboard. We also introduce Occupational Categories that underlie eight new leaderboards: 💻 Software & IT Services ✍️ Writing, Literature, & Language 🔬 Life, Physical, & Social Science 🎭 Entertainment, Sports, & Media 📈 Business, Management, & Financial Ops 🧮 Mathematical ⚖️ Legal & Government 🩺 Medicine & Healthcare Explore how models perform across fields in thread 🧵 👇
16
49
22
392
161,663