member of general partner staff @ycombinator
Cambridge, MA
Joined November 2015
- Tweets3K
- Following1.2K
- Followers21.3K
- Likes2.4K
Pinned Tweet
Fun update: I got tired of disliking every email client I’ve ever used and built my own. It’s called Exo (for exoskeleton). It’s Claude Code for my inbox. It manages my inbox for me, and it’s open source. Link to repo + some notable features in thread!
Ankit Gupta retweeted
We have raised a $550M Series E at an $8.8B valuation, led by Atreides.
Every major existing investor is back in, and we’re welcoming Baron Capital, Capital Group, Fidelity, Glade Brook and J.P. Morgan. We picked investors who think in decades. Several of them work with law firms and legal teams every day, so they can see where the work is heading.
Lawyers are moving from AI assistants to agents that do multi-step work. The money goes there. We’re growing our teams across North America, Europe, Latin America and Asia-Pacific, and going deeper on the Legora aOS.
Thank you to the 1,000+ people at Legora, and to the 140,000 legal professionals who use what we build.
I say this a lot because I believe it: it’s the best time in history to be a lawyer. We're thankful to be a part of it.
Full story: legora.com/blog/series-e
Ankit Gupta retweeted
@sequoia is leading our Series C at a $1B valuation.
Since January, we’ve gone from running 5M CI jobs a week to over 90M. Agents are writing an order of magnitude more code, and all of it still needs to be built and tested before it ships.
Ankit Gupta retweeted
GPT-6 Astra can now paint the Golden Gate bridge and type on a keyboard. But can it throw a dinner party for real guests?
@claireymao and I are on a mission to find out. We gave it a robot arm (named Podrick) and a kitchen. As a first task, Podrick, controlled by Astra, garnishes each cup of water with a slice of lemon! 🍋
Ankit Gupta retweeted
Discovering dexterity:
At home with my infant, and at Origami @origami_robotic ...
More tomorrow 👀
Ankit Gupta retweeted
You may think the frontier models structure their thinking somewhat similarly. Our research found that their generation process differ wildly, and you can see it in the output. Watch as we compare @OpenAI's GPT-6 Astra and @Anthropic's Claude Opus 5.5
Ankit Gupta retweeted
We are hiring at @autumn_ai_inc to build the people intelligence platform.
Roles:
- Search Lead (index + search over 2bn people)
- Harness Lead (co-train small model with our harness)
- Founding Growth (community, content; i.e. get the bag)
ycombinator.com/companies/au…
Ankit Gupta retweeted
1) Today we're announcing Atomwell, our state-of-the-art foundation model for predicting experimental outcomes, alongside scaling laws for molecular discovery.
We trained models from 800M to 9B parameters on ~80M experimental measurements, and found clean scaling laws in both model size and data.
Ankit Gupta retweeted
Greptile has raised a $50M Series B led by Menlo Ventures. The round values the company at $750M.
Greptile reviews and tests more than 1% of all PRs opened on GitHub every day. Thousands of engineering teams including Nvidia, Coinbase, Decagon, and Vanta use Greptile as their independent and central validation layer.
Fewer than 50 people, we are a small, talent-dense team of engineers and researchers in a quiet corner of San Francisco. If you’re interested in joining us, we’d love to hear from you.
RFS: work on problems hard enough that smarter models matter
OpenAI, Anthropic and other leading AI labs face a growing challenge: Better models aren't necessarily attracting more usage. A 10-point intelligence advantage now earns just 20% more developer traffic, down from nearly 69% in late 2025.
Full story: thein.fo/4hGJ9jY
Ankit Gupta retweeted
We clustered models purely by how their websites look (cosine similarity on the style vectors, then hierarchical clustering). Models whose websites look alike end up on nearby branches.
The labs sorted themselves out:
• every GPT-5.6 and GPT-6 model on one branch
• Gemini 3.5 → 3.8 Flash in a tight chain
• all three Muse Spark models together
Ankit Gupta retweeted
We found that open-weight models often resemble @AnthropicAI model outputs' style vectors
This is not true for @Zai_org and later @Kimi_Moonshot models
We clustered models purely by how their websites look (cosine similarity on the style vectors, then hierarchical clustering). Models whose websites look alike end up on nearby branches.
The labs sorted themselves out:
• every GPT-5.6 and GPT-6 model on one branch
• Gemini 3.5 → 3.8 Flash in a tight chain
• all three Muse Spark models together
Ankit Gupta retweeted
Introducing DeepRSI, a new industry standard for measuring recursive self-improvement.
On preexisting benchmarks, tasks are contributor-based and short-horizon (~12 hrs). In DeepRSI, we push agents to work 48 hours on ML tasks actually derived from ongoing proprietary research.
Ankit Gupta retweeted
“The contrast is awfully clear. Austin has added a lot of housing, and Boston just hasn’t…
Big picture, Austin has made it easy to build housing, and it used to be difficult.” @AlexHrwtz
Cambridge, MA lab space owners should take this advice and make all those empty biotech buildings into robot startup space
Boston lab space is 28.9% empty. The Bay Area is at 30.7%. Houston is at 26.3%.
The XBI is up more than 20% this year. So why is no one moving back into these labs?
Honestly, it shouldn't surprise anyone. Back in 2021 and 2022, the whole game was cramming as many researchers and benches into a building as you could. A few things changed that pretty fast.
- AI-native biotechs lease about a third less space per person.
- Pharma is licensing molecules and sending manufacturing overseas. When we ran a talk at MBC Biolabs, we heard that close to 80% of the companies there outsource some or all of their manufacturing to Chinese firms.
- Later-stage companies want office, not lab. They're managing assets already in trials.
The lab of 2026 just looks different. Teams are smaller. The science is AI-native. And a lot of the actual work isn't happening in Boston or SF anymore. It's happening in Shanghai and Suzhou.
The R&D that does happen here is smaller, leaner, and set up to scale out somewhere else. Those teams don't need rows of benches. They need machines. If your building has 500 benches but not a single mass spec, liquid handler, or flow cytometer, nobody's going to want it. The equipment is the real bottleneck for early-stage work.
The few groups that still want big footprints are building them around robots and power. @LilaSciences grabbed about 244,000 sq ft at Alewife Park in Cambridge for autonomous labs. @medra_ai put hundreds of robots into a concrete industrial building in SF.
So here's the uncomfortable part. These buildings were built for a tenant that doesn't exist anymore. That tenant isn't coming back. Leaving the space empty and hoping 2021 returns isn't a plan.
You have to rebuild it for who actually needs space now.
- Deep tech and hard tech. I've been talking with groups here in SF, and a lot of robotics and hardware teams are looking for exactly this kind of space. Strong floors, real power, good ventilation. Retrofitting a lab is way faster than building new.
- Incubators built around machines, not benches. Rip out the bench rows and put in the instruments early-stage companies can't afford on their own.
MBC Biolabs is already doing this. Shared instruments, shared equipment, small teams that don't need 50,000 sq ft.
Landlords who follow that blueprint will fill their buildings. The ones waiting on 2021 are going to keep sitting empty.
Ankit Gupta retweeted
Exa just passed $100m in annualized revenue.
We've grown almost exactly 10x per year for 3 years. We now handle 2 billion requests a month from 200k+ teams. And it’s still so early:
We predict agents will search 1000x more than humans by 2030.
Agents don't search like humans. They write long, complex queries and want all the world's data, not just the web. We've spent 5 years building a search engine designed for them.
We now serve personal agents, AI labs, coding agents, GTM agents, really any agent. Exa gives these agents access to 100s of billions of web documents and tons of proprietary data.
Our customers are increasingly noticing that information is the bottleneck to agent performance. While the world seems close to perfect intelligence with models like Fable and Astra, we are very far from perfect information.
To our small but legendary team, thank you for getting us here, from engineering to research to GTM to ops.
To our customers, it is an honor to serve you. We have some exciting releases coming soon and can't wait to share them 🫡
im 3 days into a multi-state road trip entirely on @Tesla FSD.
could not recommend this more.
for long stretches, it pulls into supercharger spots to charge right around when i need a break to stretch or get a snack, and i arrive at destinations way more relaxed than otherwise bc im basically just a passenger.
most of my hotels (average hotels near national parks) have had EV charging stations to top up for the next day.
think i touched the steering wheel for a confusing parking spot once and that's it.
oh and it saved me from a semi truck veering into my lane that i didn't notice until the car had already reacted
The Tesla Model 3 with FSD is mind-blowing. Jessica and I drove to the gym without touching the controls once. It backed out of our driveway, drove there, and backed into a parking space. It's amazing that you can buy so much future for $40k.