Peter Yang
Google's free 5-day AI Agents course is back, and this time it's all about vibe coding with agents. The last one had 1.5M learners.
Here's what the course covers:
→ Day 1: Agents + vibe coding. Build agents with Google Agents CLI and your coding agent.
→ Day 2: Tools + interop. Connect agents to external APIs and each other.
→ Day 3: Memory + context. Give agents long-term memory and personalization.
→ Day 4: Quality + security. Add testing, guardrails, and evals to improve reliability.
→ Day 5: Prototype to production. Deploy and monitor your agents.
It's 1-2 hours a day, taught by Google engineers, and (best of all) free 🥹
📌 Register here: https://fandf.co/4nnYEhL
Thanks to the Google Cloud team for their collaboration on this post.
Google's free 5-day AI Agents course is back, and this time it's all about vibe coding with agents. The last one had 1.5M learners.
Here's what the course covers:
→ Day 1: Agents + vibe coding. Build agents with Google Agents CLI and your coding agent.
→ Day 2: Tools + interop. Connect agents to external APIs and each other.
→ Day 3: Memory + context. Give agents long-term memory and personalization.
→ Day 4: Quality + security. Add testing, guardrails, and evals to improve reliability.
→ Day 5: Prototype to production. Deploy and monitor your agents.
It's 1-2 hours a day, taught by Google engineers, and (best of all) free 🥹
📌 Register here: https://fandf.co/4nnYEhL
Thanks to the Google Cloud team for their collaboration on this post.
Dr. Catharine Young
Woah. A profound shift in American science is coming.
Every federal research grant could soon require sign-off from a political appointee.
Scientific progress depends on funding decisions grounded in merit and public health need, NOT ideological approval.
hear me out:
2016, but nobody pays anything because data
shift: Today, we're launching shift. We're starting by cleaning your apartment in New York City, for free.
Here's how it works. Book a shift cleaning. A vetted shift operator comes to your home wearing one of our devices. They clean. They leave. You pay nothing.
In exchange, we record
Sylvain Gariel
Took me a while to figure out what all the ESMFold2 rage was about. At first, the benchmarking data didn't look super remarkable to me but it turns there are many impressive aspects:
- Fully open source, open weights + massive ESM Atlas (1.1B structures vs 0.2B for AF3).
- SOTA performance despite no MSA use. MSA search and triangular attention were simply taken out of the base model.
- Direct consequence, super low latency inference: 1024-residue protein structure prediction in 9 secs, still outperforming prior models on antibody-antigen tasks.
- Best in class PPI and antibody-antigen results. 65% pass rate on antibody-antigen benchmarks after inference-time scaling, significant improvement over AF3.
- Tons of experimental data, in particular with lab-validated miniprotein binders plus single-chain antibodies across 5 targets in cancer and immunology. Binding affinities consistent with therapeutic activity.
- Inference-time scaling benefits PPI: Multiple seeds + selection by confidence show real gains on challenging antibody-antigen predictions, leading to comments/hypotheses that it has learned an energy-function-like behavior via the folding module.
- Base model works without MSAs, but providing them further boosts prediction quality on difficult protein-protein interaction cases.
One caveat: No true scoring for protein-protein interactions, making it harder to assess which specific residues or domains are reliably involved in binding.
Alex Rives: Today we're announcing ESMFold2, an open scientific engine to power prediction, design, and discovery across protein biology.
The new model delivers state of the art performance on protein interactions, especially antibodies, a critical modality for therapeutics.
We have
Ryan Mulligan
To my European Nix friends: my team at Replit is hiring for a Site Reliability Engineer with infra, devops, and Agentic AI experience. DM me if you're interested, we use Nix.
The real AI whitepill is: There will always be a market for translating what the 200 IQ people and agents make down to everyone else
Tyler Bosmeny
/goal make something people want
(has no one tried it?!)
David Lieb: Thought experiment: if every company suddenly had infinite free compute, what new products would emerge?
My take: with very few exceptions, not much would change. The bottleneck is figuring out what people want, and it’s not so easy to apply compute to solve that.
Simon Willison
I'm suspicious of that that whole story about Uber blowing their AI budget and being disappointed in the results - I dug into it and it appears to have been built on very shaky foundations
Noob YouTuber question:
How do you arrange your prompter (@elgato) and monitor so that you can easily look back and forth between the two during a recording? Pics welcome.
Also anyone in the Bay Area good at home studio setups?
Liz4SF
they got rid of the standardized tests to hide the skills gap and they got caught; from ucla med school to UC berkeley, SD, SB etc; UCLA med rationalized that more Black and Brown patients would die bc only doctors of same race could give equal care; turns out not being able to do middle school math means all patients die of overdosing; equity achieved!!
Steve McGuire: A new report shows that 30% of Berkeley Calculus students are severely underprepared.
These students have “non-passing rates as high as 46%.”
From 2021-2023, “over 800 students (or 24% of the total population) did not pass their Calculus I classes.” Of the 500 among them who
Exactly right. The bottleneck has never been compute or capital. Its taste and judgment about what humans actually want. Infinite compute just makes the great founders faster and the confused ones more confused. https://x.com/dflieb/status/2060196206572368067
David Lieb: Thought experiment: if every company suddenly had infinite free compute, what new products would emerge?
My take: with very few exceptions, not much would change. The bottleneck is figuring out what people want, and it’s not so easy to apply compute to solve that.
Many such cases
Philip Johnston: YC S24 is shaping up to be a prolific batch... three unicorns already, 1 year and 8 months out from demo day!🦄
@ycombinator
There is always new music to be made
I’m going to remember that
LeverBiz: @garrytan Yea that’s like saying in 2010 that all music had been made and it was impossible to make new sounding songs anymore because there were already so many millions.
How did GBrain become SOTA on LongMemEval?
I use my 300k markdown file knowledge wiki repo with GBrain every day in OpenClaw to do my work and thinking
If it forgets, it becomes a test case failure and I root cause the error and PR a fix until it works
Brad Gessler
Claude/Codex + Elixir + Phoenix + LiveView is such an insane superpower.
The entire stack, from low-level distributed systems modeling all the way up to the UI, is all built in one language & runtime that's easy to reason through.
Cointelegraph
🚨 LATEST: Visa has invested in AI coding platform Replit to power agentic payments, exploring how to enable developers and AI agents to accept payments directly within the platform.
Alessandro Favero
AI needs vastly more data than we do. One idea might close the gap: don't predict raw signals (tokens), predict your own abstract latent representation (JEPA, data2vec).
With @DanKorchinski @MatthieuWyart, on a toy model, we prove how much that helps: the gap is exponential.
🧵
new 5.5 instant model in chatgpt:
Michelle Pokrass: we shipped a new version of gpt-5.5 instant today. the previous model was too bullet pilled. the new one improves on some other important dimensions: sycophancy, factuality, and multilingual performance. hope you'll like it! always interested in feedback
There are two loops in every founder's head.
The autism loop: run your own model to the floor, ignore consensus, hold a thesis when everyone says you're wrong. That makes conviction.
The empathy loop: feel what the user feels, sense what the market wants before it has words. That makes traction.
Most people crank one and starve the other. Pure conviction builds something brilliant nobody wants. Pure empathy builds consensus mush.
PG put the whole job in four words: make something people want. The autism loop makes the something. The empathy loop knows it's wanted. The founder is the bridge.
Most great founders show up dominant in the first loop. That's why they're contrarian enough to try at all. The work is grafting on the second.
There is no place in the world that helps founders make the two loops work together to make great startups than Y Combinator. It is the most gratifying part of our work.
Everyone's bottleneck in voice AI is the same: retrieval. The agent thinks, network round-trips to a vector DB, and the magic dies.
Moss runs search at sub-10ms (no hop). Open source. This is the layer voice agents were missing. Build on it June 6-7 at the YC office.
Pete Koomen: Come build agents that can finally hold a fluid conversation at the 24-Hour Conversational AI Hackathon, hosted by @usemoss at the YC Office, June 6-7. First place wins an interview with a YC partner: https://events.ycombinator.com/conversational-ai-hackathon-2026
⬆️ tecrinex
Garry Tan: There are two loops in every founder's head.
The autism loop: run your own model to the floor, ignore consensus, hold a thesis when everyone says you're wrong. That makes conviction.
The empathy loop: feel what the user feels, sense what the market wants before it has words.
Noni (Bamboo)
My first Educational App I made completely on @Replit and got onto the App Store in April
With almost 2k downloads (monetised!) in just a few weeks x
@raymmar @MannyBernabe @amasad @Franciscocrz
https://apps.apple.com/gb/app/bamboo-brain-sats/id6758530606
Edward Lando
I changed my mind and started working on a startup last night.
Garry Tan: Oh man, this is going to be a very tired argument we are going to have for the next 5 years
You can still tell human writing because it's got that wabi sabi
Kosta Derpanis (sabbatical in Zurich)
.@ylecun’s definition of what is a world model.
There's an infinite well of things to do that could work that are of this kind
Paul Graham: @charliermarsh Sometimes they're a good reason to work on something. When people say a market is "crowded," what that often means is that there's a real problem and none of the solutions are good enough yet.
Have to report: Opus 4.8 is fucking awesome with OpenClaw
It's much more clear about its fixes, what it's thinking, and how it works with you to fix problems.
Startup Archive
Jeff Bezos: “People who are right a lot change their mind a lot”
“Because of AI, new technologies, and all the dynamism in the world, so many things are changing — and they’re changing rapidly,” Jeff observes.
The best solution he’s found for dealing with this rapid change is “thinking long-term” because it forces you to ask yourself, “What are the points of stability?” and “What is not going to change?”
He continues:
“One of the things that changes very slowly is customer needs. So you can build a strategy around customer needs. That will have durability.”
When building Amazon, for example, Jeff built the company around the customer needs of fast delivery, low prices, and vast selection.
“The technologies will change. Your competitive set will change. Everything will change, except those customer needs,” Jeff argues.
And it’s this idea that is behind Jeff’s core idea of “Be stubborn on the vision, and flexible on the details.”
He explains:
“You have to be [flexible on the details] because the world is changing and so you change your mind. I’ve noticed that people who are right a lot change their mind a lot. People who are wrong a lot are very stubborn on the details.”
Source: @reuters (Oct 2025)
Tomasz Tunguz
I've been using state-of-the-art models to teach small models running on my computer how I work.
The result : a personal agent that runs my inbox, my deal pipeline, my blog, my calendar, & my research. 🧵
Paul Graham
Justin Kan back when what later became Twitch was just him walking around with a camera on his head 24/7.
Paul Graham
A rare 2009 photo of Kate Courteau, the architect who created YC's look.
Tariq Habash
It’s primary election season and people ask me all the time, who should I support in my area that opposes Israel's genocide?
So I pulled together this interactive voter guide with Sway: https://sway.co/voter-guide?j=tariq
Share with your friends and let’s vote to stop arming Israel.
I have difficulty arguing that most of the writing changes suggested in gmail now aren’t improvements, but it does tend to wipe out my particular authorial voice.
Ankit Gupta
lots of great startup lessons from @varunvummadi leading Giga. quickly becoming a generational company that is taking down some formidable competitors. Great having him at Startup School in India earlier this year!
Y Combinator: Varun Vummadi (@varunvummadi) is the co-founder of @GigaAI, which builds AI agents for customer support for some of the biggest companies in the world, including DoorDash, one of the largest crypto exchanges in the US, and a top-three global telecom provider.
At Startup School
I’m very excited by this new benchmark dataset for visual generation that is suitable for the modern era of large scale generative models!🤩
Keshigeyan Chandrasegaran: 1/ Introducing GPIC: a Giant Permissive Image Corpus and benchmark for visual generation!
🚀100M VLM-captioned image-text pairs for training
📊1M image-text pairs for benchmarking
🖼️~28 trillion pixels
🤗Centrally Hosted
✅Fully permissive for research + commercial use
Dataset,
Joel C. Sercel, PhD
The brilliant professors of the University of California have shown themselves once again to be the fools those with common sense know them to be. As a guy who hires fresh out graduates I can say that since this ban on standardized testing has gone into effect, recently grads have in general become even less impressive than they were. At least before this you knew that a UC grad was at least bright enough to read, write and do basic math. Link below the post.
defensive acceleration in biology with Rosalind:
OpenAI: We’re taking steps to accelerate defensive progress in biology:
- Launching Rosalind Biodefense to help trusted builders develop new biodefense and pandemic preparedness capabilities.
- Expanding trusted access to GPT-Rosalind for select U.S. government and allied partners
Replit ⠕
Here's everything you need to know about Replit in 60 seconds ⭐️
→ Plain English prompts turned into real working software
→ End-to-end workflow from UI to deployment
→ Real-time team collaboration with just a link
→ Parallel AI agents building different parts of your app at once
Sar Haribhakti
"...in the 2020 year of progressive pandemic madness, the University of California led the Ivory Tower movement to drop standardized tests as an admissions requirement in the name of equity. The experiment has been a failure.."
"Drop standards, and learning mastery declines."
Significant upgrades for Codex users on Windows:
OpenAI: Windows users, this one’s for you.
Computer use now works on Windows, so Codex can take action on your Windows computer.
And with Windows support for Codex in the ChatGPT mobile app, you can start, review, and steer tasks on the go while work continues on your Windows machine.
Bill Clerico
The two timeless traits in founders that @paulg and @jesslivingston look for in @ycombinator interviews: determination and adaptability
disasterproof: In 2008, @paulg and @jesslivingston took a chance on @billclerico's startup. Nearly two decades later, what's changed in how they pick founders? http://youtu.be/KBjtRJSblMU
guys Opus 4.8 is very very good at writing agent code
(zero dependencies, all llm 1P SDKs not just claude, sorry agent frameworks)
you should try it. think they trained @ErikSchluntz's and @barry_zyj's Building Effective Agents into this thing
The craziest thing every business can do in 2026 is just zap the rocks more and make more money
Nicolas Dessaigne: My advice to founders in 2026: spend tokens, not headcount.
Record everything. Make your company queryable. Build self-improving loops.
Before long, AI won’t just help you operate your company. It will make it self improving.
Don't think AI adoption, think AI transformation.
Docker inside ▲ Sandbox live 𝚗𝚘𝚠
Vercel Developers: Run Docker inside Vercel Sandbox.
▪︎ Build and run containers in full isolation
▪︎ Persist installs and images across sessions
▪︎ Run databases, test suites, or full apps
https://vercel.com/changelog/run-docker-containers-inside-vercel-sandbox
OpenAI for realtime translation — speak in any of 70+ input languages and translate into 13 output ones:
cayden 凯登: OpenAI just dropped a completely new kind of model
gpt-realtime-translate takes in speech audio from any language and outputs speech in your target language
LLMs are great, but you need specialized models for specialized use cases
We're running this on our smart glasses
The "we will upgrade the dependencies later" excuse just died. When AI makes library upgrades nearly free, staying current stops being a luxury and becomes the default. Tech debt was always a tooling problem and now it’s a thing of the past.
https://x.com/snowmaker/status/2060460947010158963
Jared Friedman: At YC we're now running on the latest versions of rails and react for the first time since we started the codebase. Now that AI made library upgrades easy, people should do them more.
Garry's List
Prop D claims to tax overpaid CEOs. But some of the wealthiest get exemptions. The real cost gets pushed downstream to immigrant families, workers, and small businesses. Vote No on D by next Tuesday June 2nd.
Watch Algún Día, our latest GL Studios production.
table of contents for your long-running chatgpt conversations:
ChatGPT: For every ChatGPT conversation that started as “one quick thing” and became a full on saga: table of contents is here.
Available now for chats with 5+ responses.
Justin Johnson
GPIC should be the new standard benchmark for generative modeling. Training 1 epoch on GPIC is the same cost as 100 epochs on ImageNet, but is a much better proxy for real-world problems. If you work in generative modeling, try GPIC for your next project!
Keshigeyan Chandrasegaran: 1/ Introducing GPIC: a Giant Permissive Image Corpus and benchmark for visual generation!
🚀100M VLM-captioned image-text pairs for training
📊1M image-text pairs for benchmarking
🖼️~28 trillion pixels
🤗Centrally Hosted
✅Fully permissive for research + commercial use
Dataset,
Neetu Arnold
UPDATE: the open letter to restore standardized tests in STEM admissions at the University of California has received 1000 signatures!
7/9 Math department chairs
44 STEM chairs
1 engineering dean
Neetu Arnold: University of California STEM professors want standardized tests back due to severe math deficiencies among students:
“We now observe preparation gaps so severe that instructors must reteach middle school mathematics”
“The current admissions metric, based primarily on GPA &
vikas sabbi
Dino Ep-2
made this on @Replit with Runway MCP.
https://apps.apple.com/us/app/dino-initiative/id6763940737
Amjad Masad: Use Runway in Replit
codex for managing the codex UI:
Guinness Chen: If you ever get tired of managing your Codex threads, just let Codex manage itself! Codex can now create threads, search them, organize them, pin the important ones, and spin up worktrees for parallel tasks.
The California Supreme Court just made it even harder to keep repeat offenders off the street before trial.
This is how public safety collapses: not all at once, but ruling by ruling, loophole by loophole, until normal people are the only ones following the law.
Brooke brought accountability back to San Francisco, now California needs to catch up.
The Successful Recall of Chesa Boudin: The CA Supreme Court just issued a major new ruling (In re Kowalczyk) that mandates that “attainable” bail be given to everyone not accused of a violent crime - including drug dealers, repeat offenders, and armed felons. It essentially guarantees their release back onto the
