Nicolas Dessaigne retweeted
We're excited to announce Exa 3.0!
A huge number of improvements:
- Bigger, better embeddings model driving core search
- Higher quality highlights with further reduced token counts
- Lower latency, cheaper search type "Lite"
- Better reranking, particularly across long documents
- Even better real-time news, sports, weather, and market data
- Laundry list of other stuff
Let us know what you think! 🫡
A huge number of improvements:
- Bigger, better embeddings model driving core search
- Higher quality highlights with further reduced token counts
- Lower latency, cheaper search type "Lite"
- Better reranking, particularly across long documents
- Even better real-time news, sports, weather, and market data
- Laundry list of other stuff
Let us know what you think! 🫡
Introducing Exa 3.0: our largest ever improvement to search quality.
Exa is now #1 on every benchmark we tested, from public evals to verticals like coding and GTM intelligence.
It's the result of multiple research breakthroughs 🧵
10
12
242
15,302
Nicolas Dessaigne retweeted
We just solved attribution for AI Agents.
Introducing Journeys.
AI agents are visiting your product. Now you can see what they’re trying to do, what converts, and what’s making them churn.
With Journeys, you can:
- See the moment AI agents visit your product
- Follow the paths ases, installs, and completed setup
- Pinpoint where agents get stuck and whats blocking them
- Review fixes prepared for you and publish them with a click
- Watch how those changes improve conversion
Turn agent traffic into revenue.
Try Journeys at tryscope.com
Introducing Journeys.
AI agents are visiting your product. Now you can see what they’re trying to do, what converts, and what’s making them churn.
With Journeys, you can:
- See the moment AI agents visit your product
- Follow the paths ases, installs, and completed setup
- Pinpoint where agents get stuck and whats blocking them
- Review fixes prepared for you and publish them with a click
- Watch how those changes improve conversion
Turn agent traffic into revenue.
Try Journeys at tryscope.com
190
60
468
288,024
Nicolas Dessaigne 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. x 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 🫡
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. x 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 🫡
225
120
2,040
451,517
Nicolas Dessaigne retweeted
VibeCon SF applications are now open! 🌉
Powered by @MongoDB, VibeCon is coming back to where it all began: San Francisco!!!
200 builders. One weekend. A whole lot of ambitious ideas. And the winner gets an interview opportunity with @ycombinator.
October 24–25. Mark your vibecon.com
Powered by @MongoDB, VibeCon is coming back to where it all began: San Francisco!!!
200 builders. One weekend. A whole lot of ambitious ideas. And the winner gets an interview opportunity with @ycombinator.
October 24–25. Mark your vibecon.com
14
4
34
1,880
Nicolas Dessaigne retweeted
When you talk to successful founders, they always wish they'd hired fewer top managers from outside. I don't think I've ever met a founder who wished he or she had promoted fewer people from within.
138
139
3,252
245,249
Nicolas Dessaigne retweeted
two decades of friendship with @sirsho29; killed one company; more rejections than we can count
and today, @antimattr_one is part of @ycombinator's Fall 2026 batch
we're building a voice hardware so that you can do more with your voice, anywhere; we already have close to 1,000 way of choosing greatness; this is our "antimattr” story
what’s yours?
and today, @antimattr_one is part of @ycombinator's Fall 2026 batch
we're building a voice hardware so that you can do more with your voice, anywhere; we already have close to 1,000 way of choosing greatness; this is our "antimattr” story
what’s yours?
29
15
217
1,380,765
Nicolas Dessaigne retweeted
Launching Higher-Order Extraction.
It deterministically computes the numbers you need from the raw values on the page, even when the final answer isn't printed anywhere.
It deterministically computes the numbers you need from the raw values on the page, even when the final answer isn't printed anywhere.
31
15
158
65,066
Nicolas Dessaigne retweeted
🚀 What will the 2027 VC playbook look like?
Nicolas @dessaigne, Algolia co-founder turned #YC General Partner, has some thoughts.
Catch him on stage at #OneWaySummit, the flagship event for global founders.
📅 Oct 28–29, SF
🔗 luma.com/onewaysummit20…
#SFevents #SFtechweek
Nicolas @dessaigne, Algolia co-founder turned #YC General Partner, has some thoughts.
Catch him on stage at #OneWaySummit, the flagship event for global founders.
📅 Oct 28–29, SF
🔗 luma.com/onewaysummit20…
#SFevents #SFtechweek
0
2
1
351
Nicolas Dessaigne retweeted
We’re excited to welcome @vivianmshen and @raphaelschaad as YC’s newest General Partners!
Vivian and Raphael first came to YC as founders.
Vivian co-founded Juni Learning (W18), which teaches kids coding and math, and later founded Acely, an AI tutor for college prep. Raphael founded Cron (W20), the calendar app that became Notion Calendar.
They built products used by millions of people, then returned to YC as Visiting Partners to help other founders. Now they're doing it full-time.
ycombinator.com/blog/welcome-v…
Vivian and Raphael first came to YC as founders.
Vivian co-founded Juni Learning (W18), which teaches kids coding and math, and later founded Acely, an AI tutor for college prep. Raphael founded Cron (W20), the calendar app that became Notion Calendar.
They built products used by millions of people, then returned to YC as Visiting Partners to help other founders. Now they're doing it full-time.
ycombinator.com/blog/welcome-v…
38
28
384
135,446
Nicolas Dessaigne retweeted
YC said yes ♥️
It started in an IIT Delhi dorm 7 years back, when I first met Pratyush. 9 months ago we took a leap of faith, and today @ByteAsk is part of YC Fall 2026.
So what's cooking?
The critical infrastructure of the world still runs on legacy C and C++. Trading, ecom, energy, automobiles, operating systems, gaming, medical devices, aerospace, and defense all depend on it.
ByteAsk brings agentic coding to that code.
Developers use it to debug, optimize, and write C and C++ faster than ever, and with confidence. Teams use it on 20-year-old systems with millions of lines and hundreds of dependencies, and see the first change land within hours. It runs fully on-prem, without internet.
If your systems run on C or C++, DM me. We are taking 5 more enterprises as design partners, at no cost, and I will fly to your office to get it running.
Grateful to our customers, angels and mentors for believing in us, and to @dessaigne , @nemild and @ycombinator team for backing us.
PS: Comment "C++" to get access to our early release
@pratyush1505
It started in an IIT Delhi dorm 7 years back, when I first met Pratyush. 9 months ago we took a leap of faith, and today @ByteAsk is part of YC Fall 2026.
So what's cooking?
The critical infrastructure of the world still runs on legacy C and C++. Trading, ecom, energy, automobiles, operating systems, gaming, medical devices, aerospace, and defense all depend on it.
ByteAsk brings agentic coding to that code.
Developers use it to debug, optimize, and write C and C++ faster than ever, and with confidence. Teams use it on 20-year-old systems with millions of lines and hundreds of dependencies, and see the first change land within hours. It runs fully on-prem, without internet.
If your systems run on C or C++, DM me. We are taking 5 more enterprises as design partners, at no cost, and I will fly to your office to get it running.
Grateful to our customers, angels and mentors for believing in us, and to @dessaigne , @nemild and @ycombinator team for backing us.
PS: Comment "C++" to get access to our early release
@pratyush1505
61
8
265
10,228
Nicolas Dessaigne retweeted
Two YC companies led the way in igniting a revolution in how the robotics community sees robot control via LLMs.
Tune in to hear more about the last five years of research that gave them the conviction that we were heading this way and some predictions on what comes next.
Tune in to hear more about the last five years of research that gave them the conviction that we were heading this way and some predictions on what comes next.
One of the biggest surprises in AI over the last few years has been how well coding agents generalize beyond software.
In a recent essay, MIT professor Philip Isola argued that we may be entering the era of robot-use agents: general-purpose models that can control different and learn new physical tasks with little or no robot-specific training.
In this episode of Decoded, we're joined by the founders of @theWaddleLabs and @Robocurve, two of the startups whose work helped drive this realization. They're working at the frontier of using general-purpose models to control robots, and their recent demos helped inspire the growing conversation around robot-use agents.
Together, we dig into the research behind that idea, from code-as-policies and vision-language-action models to the harnesses and evals needed to make these systems work in the real world.
00:00 — What Are Robot-Use Agents?
02:11 — From Vision-Language-Action Models to General LLMs
04:07 — The Bitter Lesson for Robotics
07:04 — How Coding Agents Learned to Control Robots
10:41 — How Robots Learn From Experience
14:22 — The Harness as a Form of Robot Intelligence
16:14 — Watching Astra Control a Robot
20:55 — Why General Models May Win in Robotics
26:05 — How Close Are General-Purpose Robots?
13
11
97
22,096
Nicolas Dessaigne retweeted
🤯
Introducing Agent Ultra - a step change in deep research
Agent Ultra orchestrates swarms of agents to perform exhaustive research, build comprehensive lists, and answer questions requiring thousands of sources.
In both evals and vibes, it's state of the art
4
8
138
23,498
Nicolas Dessaigne retweeted
Introducing GLiNER2.5-Decide, our new 340M parameter open weight, encoder-based decision model.
GLiNER2.5-Decide is built for fast, deterministic classification. The model evaluates a set of user-defined typed questions and rules, and jointly decodes their answers, returning th probability distributions and confidence scores.
We evaluated the model’s performance on Fast Decisions, an unseen, internally generated classification suite based on 17 datasets testing real-world use cases across routing, triage, classification, sentiment, and content understanding.
Measured against similar decision models, GLiNER2.5-Decide leads in 9 of the 17 datasets, achieving the highest average score:
- GLiNER2.5-Decide: 60.1%
- SemIf: 56.4%
- JevK5: 57.5%
- Laya: 46.6%
This performance makes the model a strong fit for use cases like tool calling, model routing, browser and computer use, and LLM-as-a-judge.
GLiNER2.5-Decide’s lightweight encoder architecture makes it easy to fine-tune the model for specific tasks, while being efficient enough to run locally on consumer-grade CPUs or in air-gapped environments, giving users greater control over where their data goes and where the model runs.
To make building and experimenting with GLiNER2.5-Decide as easy as possible, we're also offering hosted inference. You can now use our API to run inference and fine-tune GLiNER models on specialized tasks right inside your own coding agent: agent.fastino.ai
As with previous models, we’re also releasing the model weights on @huggingface under the Apache 2.0 license: huggingface.co/fastino/GLiNER…
GLiNER2.5-Decide is built for fast, deterministic classification. The model evaluates a set of user-defined typed questions and rules, and jointly decodes their answers, returning th probability distributions and confidence scores.
We evaluated the model’s performance on Fast Decisions, an unseen, internally generated classification suite based on 17 datasets testing real-world use cases across routing, triage, classification, sentiment, and content understanding.
Measured against similar decision models, GLiNER2.5-Decide leads in 9 of the 17 datasets, achieving the highest average score:
- GLiNER2.5-Decide: 60.1%
- SemIf: 56.4%
- JevK5: 57.5%
- Laya: 46.6%
This performance makes the model a strong fit for use cases like tool calling, model routing, browser and computer use, and LLM-as-a-judge.
GLiNER2.5-Decide’s lightweight encoder architecture makes it easy to fine-tune the model for specific tasks, while being efficient enough to run locally on consumer-grade CPUs or in air-gapped environments, giving users greater control over where their data goes and where the model runs.
To make building and experimenting with GLiNER2.5-Decide as easy as possible, we're also offering hosted inference. You can now use our API to run inference and fine-tune GLiNER models on specialized tasks right inside your own coding agent: agent.fastino.ai
As with previous models, we’re also releasing the model weights on @huggingface under the Apache 2.0 license: huggingface.co/fastino/GLiNER…
83
162
1,400
162,914
Nicolas Dessaigne retweeted
@ycombinator's @dessaigne is joining us at Flower AI Day 2026.
Nicolas is a General Partner at YC and co-founder of Algolia, the search API company that went through YC’s Winter 2014 batch. As CEO, he grew @algolia to a team of 350 before handing over the role in 2020. Today, he Y Combinator, bringing experience from both building a developer-focused company and helping others build theirs. His background also includes a PhD in Computer Science from the @NantesUniv.
Join Nicolas and the researchers, engineers, and founders at Flower AI Day on September 30 at The Pearl in San Francisco. We’ll spend the afternoon and evening exploring open, collaborative AI through technical talks, demos, firesides, and conversations with the people building it.
Register: luma.com/flwrlabs-uygg
Nicolas is a General Partner at YC and co-founder of Algolia, the search API company that went through YC’s Winter 2014 batch. As CEO, he grew @algolia to a team of 350 before handing over the role in 2020. Today, he Y Combinator, bringing experience from both building a developer-focused company and helping others build theirs. His background also includes a PhD in Computer Science from the @NantesUniv.
Join Nicolas and the researchers, engineers, and founders at Flower AI Day on September 30 at The Pearl in San Francisco. We’ll spend the afternoon and evening exploring open, collaborative AI through technical talks, demos, firesides, and conversations with the people building it.
Register: luma.com/flwrlabs-uygg
1
10
20
4,255
Nicolas Dessaigne retweeted
There's something deeply unsafe about safetyism.
We wish people a safe flight—yet flying commercially is almost perfectly safe.
We don't wish people a safe drive—yet US roads are the equivalent of a 737 crashing every 48h, killing everyone aboard.
But we scare people when they get in an airplane but not when they get in a car.
The biggest opportunity we have to improve road safety is self-driving, yet there's lots of anti-self-driving safety talk, and self-driving is held to a *dramatically* superhuman safety bar.
The din about AI safety has reached extreme tenor—yet there have been zero material safety incidents. And the positive benefits of AI could improve and save many lives. But the safetyists say we must slow down.
We destroyed the nuclear power industry in this country nominally out of safety concerns, yet joule for joule nuclear power is *by far* the safest form of energy, even accounting for Chernobyl (a genuine disaster) and Three Mile Island (generally a non-event).
Building and fire codes are supposedly there to protect us from unsafe buildings. But actually the red tape makes development slow and sometimes impossible. So the safetyism protects us from jobs, innovation, and reindustrialization.
TSA was there nominally to protect us from terrorist threats—yet to the extent it pushes passengers at the margins from safe airplanes to dangerous roads, it actually kills not just time and money but actual humans.
Safetyism is deeply unsafe.
We wish people a safe flight—yet flying commercially is almost perfectly safe.
We don't wish people a safe drive—yet US roads are the equivalent of a 737 crashing every 48h, killing everyone aboard.
But we scare people when they get in an airplane but not when they get in a car.
The biggest opportunity we have to improve road safety is self-driving, yet there's lots of anti-self-driving safety talk, and self-driving is held to a *dramatically* superhuman safety bar.
The din about AI safety has reached extreme tenor—yet there have been zero material safety incidents. And the positive benefits of AI could improve and save many lives. But the safetyists say we must slow down.
We destroyed the nuclear power industry in this country nominally out of safety concerns, yet joule for joule nuclear power is *by far* the safest form of energy, even accounting for Chernobyl (a genuine disaster) and Three Mile Island (generally a non-event).
Building and fire codes are supposedly there to protect us from unsafe buildings. But actually the red tape makes development slow and sometimes impossible. So the safetyism protects us from jobs, innovation, and reindustrialization.
TSA was there nominally to protect us from terrorist threats—yet to the extent it pushes passengers at the margins from safe airplanes to dangerous roads, it actually kills not just time and money but actual humans.
Safetyism is deeply unsafe.
91
114
957
45,796
Nicolas Dessaigne retweeted
Introducing Alexandria.
We raised a $75M Series B led by Smash Capital to build the world’s largest knowledge library for superintelligence.
Supercharge your agents with instant access to the most powerful data sets & providers.
firecrawl.dev/alexandria
We raised a $75M Series B led by Smash Capital to build the world’s largest knowledge library for superintelligence.
Supercharge your agents with instant access to the most powerful data sets & providers.
firecrawl.dev/alexandria
111
102
1,187
288,564
Nicolas Dessaigne retweeted
Surprised by how quickly YC itself improves. I was previously in the spring batch and just finished F26 kickoff. Even in that short time, everything felt more organized, efficient, and thoughtfully designed from onboarding to picking up the initial merch. Bookface has also become significantly richer in features and has streamlined a lot of the administrative work of running a company and fundraising .
And the food so far has gotten much better too.
Cool to see YC iterating as fast as its startups.
And the food so far has gotten much better too.
Cool to see YC iterating as fast as its startups.
9
2
47
5,338
Nicolas Dessaigne retweeted
Today we introduce a new dimension to search - time.
Exa now stores 400 billion snapshots of webpages across time (literally a bunch of copies of the web in s3), so you/your agent can search the past and compare diffs.
Search engines have always had date filters, but this is different. Since webpages update over time (ex. apple’s home page), simple date filters leak present content into the past.
A good use case has been finance companies that want to test their prediction/trading models -- “would our model with access to March 2025 web data have predicted Anthropic’s growth?“. To answer this, you can’t have leakage from 2026 corrupting the model, so you need a snapshot.
Similarly for RL training runs, you often want to keep time constant and not have new webpage updates messing with the model’s training.
A last use case is close to my heart - the web is rapidly filling with AI generated slop. That makes data from before Nov 2022 the only 100% human-made data. Sometimes the good ol’ days really was better.
Exa now stores 400 billion snapshots of webpages across time (literally a bunch of copies of the web in s3), so you/your agent can search the past and compare diffs.
Search engines have always had date filters, but this is different. Since webpages update over time (ex. apple’s home page), simple date filters leak present content into the past.
A good use case has been finance companies that want to test their prediction/trading models -- “would our model with access to March 2025 web data have predicted Anthropic’s growth?“. To answer this, you can’t have leakage from 2026 corrupting the model, so you need a snapshot.
Similarly for RL training runs, you often want to keep time constant and not have new webpage updates messing with the model’s training.
A last use case is close to my heart - the web is rapidly filling with AI generated slop. That makes data from before Nov 2022 the only 100% human-made data. Sometimes the good ol’ days really was better.
We built a time machine for the web.
Introducing Exa Snapshot: an index of 400 billion historical snapshots of webpages that lets you search as if it's the past.
Snapshot is already being used for backtesting prediction models, RL at labs, exploring the pre-AI web, and more.
23
19
287
49,868
Nicolas Dessaigne retweeted
It really isn't. YC changed my life.
Before YC it was damn near impossible for me and my co-founders to raise a round for Oneleet. My founding story wasn't impressive at all. At 35 years I was old. I never attended an ivy league university. Never founded a company before. Investors literally thought I was a nobody, despite having some traction.
YC helped me and my co-founders get a foot in the door. My company, Oneleet, is worth hundreds of millions now.
Meeting my wife, having our children and the family and friends I feel grateful to have will obviously always be the best thing that happened to me. But after that, YC is the best thing that ever happened to me.
Before YC it was damn near impossible for me and my co-founders to raise a round for Oneleet. My founding story wasn't impressive at all. At 35 years I was old. I never attended an ivy league university. Never founded a company before. Investors literally thought I was a nobody, despite having some traction.
YC helped me and my co-founders get a foot in the door. My company, Oneleet, is worth hundreds of millions now.
Meeting my wife, having our children and the family and friends I feel grateful to have will obviously always be the best thing that happened to me. But after that, YC is the best thing that ever happened to me.
I'm going to say something that everyone else in Silicon Valley is too scared to say:
YC is a giant ripoff.
It's absurd to me that naive Founders give away 7% of their company for just $125k in this day and age.
It's a super boomer offering:
- apply, wait, hope you get in
- thrown into a 300 person cohort where you are one tiny cog in their mass portfolio bet
- 3 months of 'education' that you can now get completely for free online
- A few minutes at the end to hope a VC picks you amidst the cattle room
Was YC worth it 10 years ago when knowledge wasn't freely available and you couldn't build distribution for yourself via the internet? Yes.
Is it worth SEVEN percent of your company in today's age when you can go from $0 to $1M in weeks with just content and AI?
No.
22
15
391
45,835