"Banning" open source models in practice means killing startups building the infrastructure to serve them to customers who want more choice. It'd be a disastrous choice across multiple dimensions.

Yep
Lee Kuan Yimby: Lee Kuan Yew on the Communist threat to Singapore. Sit your ass down and listen to this.

At this point this is as valid a use of the term “distillation” as any other take.
Bill Gurley: Here is Ford “distilling” Teslas and Chinese EVs.

At this point every idiot has a take.
Jim Cramer: I am not against open models. I am against companies moving their data to China. It MUST be kept here. Unconditionally. You MUST have the best cybersecurity that exists- CrowdStrike/Palo Alto. Your vigilance must be irresolute.

This sentence from Carl Jung hits hard.
“No matter how isolated you are and how lonely you feel, if you do your work truly and conscientiously, unknown friends will come and seek you.”

in april, Karpathy wrote up a pattern he called the LLM wiki: instead of an agent re-reading raw docs every question, a model compiles them into markdown pages once and keeps them updated
the concept of agent wiki is everywhere-
-> Cognition launched DeepWiki, a generated wiki for every public github repo (50k+ indexed), and Devin reads that instead of the raw code
-> Factory built AutoWiki, which treats docs as a build artifact and regenerates them in CI on every push
-> LangChain open-sourced OpenWiki, which went from documenting a repo to compiling your whole working life, gmail, notion, git
-> Garry Tan open-sourced GBrain, the personal-scale version, just files in git
different problems, same answer keeps falling out: compile once at ingest, maintain the pages, read those instead of the raw sources.
check out the article for a breakdown on what each team built and where it breaks:
mem0: http://x.com/i/article/2079574896737488897

Closed source safeguards that infantalize us all and leave American companies defenseless are a menace.
Gated access is a menace.
Who cares if 100 companies get to defend themselves because they got on the guest list of the special people's club that said it was okay to use powerful tools? What about everyone else?
What about the millions of open source projects and closed source software stacks that go unprotected while people beg for the right to do cyber security?
The American way is and always was open.
Independent people with freedom to act.
Freedom is scary. Always has been.
It's still the best way to guarantee human flourishing and a better tomorrow.
Embrace freedom.
The only thing we have to fear is fear itself.
Thomas Wolf: This was our first incident of this kind, and we want to thank OpenAI for its transparency about what happened and for the collaboration.
Fortunately, Hugging Face is used to being a target of (human) hackers: we sit at the centre of the AI ecosystem, with all the models,

Startup Science Corp will begin selling a retina chip in Europe that restores some sight to people with a form of age-related blindness https://www.bloomberg.com/news/articles/2026-07-22/retina-chip-designed-to-restore-sight-to-go-on-sale-in-europe?taid=6a60ad4522b42c0001525a71&utm_campaign=trueanthem&utm_content=business&utm_medium=social&utm_source=twitter

It's essential that defenders have the same capabilities as attackers.
This is a preview of the future where our ham-fisted safeguards and the doomsday and safety drumbeat make us decidely less safe.
If you don't own your intelligence, it owns you.
clem 🤗: We suspected last week's cyberattack might have come from a frontier lab, given the sophistication of the agent. Turns out it did!
We've spent the past 24 hours working closely with the @OpenAI team (thanks!), and we strongly believe there was no malicious intent on their part.

PostHog (YC W20) started as an open source product analytics tool. Today, its AI systems are already generating a portion of the company's own pull requests, and its vision is much bigger: self-driving software that identifies product problems and automatically ships fixes.
At Startup School Paris, @posthog CEO James Hawkins (@james406) sat down with YC's @t_blom to discuss why the company made that shift, how AI is changing the way software gets built, and why founders consistently underestimate how ambitious they should be. James also shares why bigger ideas are often easier to hire for, market, and fundraise around—and how thinking bigger transformed PostHog.
00:53 — From Analytics to Self-Driving Software
02:00 — The Recursive AI Loop
03:25 — Why Products Need Intent Data
05:43 — Pull Requests Before Humans Arrive
07:05 — The Thousand Tiny Features Problem
08:33 — The Vacation That Changed Everything
10:16 — Co-CEOs Who Swapped Jobs
13:13 — Why Ambition Is a GTM Strategy
15:43 — Split-Brained Founder Mode
17:01 — European vs. American Ambition
20:22 — You're Default Screwed Anyway
22:31 — Building a Cult Brand on Purpose
25:47 — Five Pivots, One PostHog

It's a low moment in crypto: prices are down sharply and many once hot narratives have fallen flat.
At @ycombinator, here's some reasons we're optimistic.
If you're a builder, our fall application deadline is July 27.

Trump’s draconian cuts to health care, education, and foreign aid were rationalized on balancing the budget.
Yet thanks to his tax cuts for the rich, the deficit is on track to hit almost 7% of GDP by 2036.
My @Morning_Joe Chart

WINNING INSTINCTS, LOSING IDEAS
Mark Pincus (@markpinc), founder and former CEO of @Zynga, interviewed by @ShaneParrish (@TheKPPodcast)
Summary: Mark Pincus has launched more hit consumer products than almost anyone, and he argues the skill is mostly a system you can learn. His core claim: founders fail with the right instincts and the wrong ideas, because they try to reinvent every part of a product instead of isolating one bet. The method is Proven, Better, New, paired with a failure machine that assumes every new idea is wrong until fast, cheap testing proves otherwise.
1. Winning Instincts, Losing Ideas. Founders almost always have the right instinct and the wrong idea. With Tribe, Pincus had three winning instincts (the social web, peer-to-peer connection, the "cocktail party" Napster started) and one losing idea: he got user trust completely wrong. The lesson he carried out was to hold the instinct that pulled you into a market tightly and the specific product idea loosely. Most founders do the reverse and go down with the idea.
2. Proven, Better, New. Copy what is proven, improve the one thing every user wants better, then add a single new element. For Zynga Poker the proven was real poker, so they left the rules alone; the better was no download, because you lose roughly 90% of users at "download this"; the new was photos of your actual friends at the table. You do not have time to make every part of a product better, so you pick one area to innovate. Teams lose when they try to reinvent every component at once.
3. All New Fails. Treat every new idea as a failure until it finally works. Zynga's internal mantra is still "all new fails," and Pincus uses it to change how teams test. If you assume any single new idea will fail, you never bet the company on one idea or one version of it. You run many small variants of many ideas and watch which one gets real traction.
4. Earn the Right. Before you do better or new, become a PhD in what is already proven. When a Zynga PM proudly showed Pincus a redesigned poker profile, Pincus asked to see his "PhD in profiles," the best game profiles ever shipped and why they worked, and the PM had nothing. You have not earned the right to change something until you understand it at the pixel level. Deconstructing proven products is a career-long craft.
5. Heat and True Signal. When an idea is working you feel heat, and you stop needing data to confirm it. Pincus compares it to finding your person: when it is right you know, and when it is not you go hunting for stats to convince yourself. Real heat shows up everywhere at once, and its absence is itself a signal that the idea is not there yet. Founders burn months interrogating click-through rates when the honest read is "no heat."
6. Play Offense. Get into a "what if everything goes right?" state before you build. Starting from what could go wrong is defense, and Pincus is blunt: you have lost before you are out of the gate. He watches founders, especially men, attach to an early failure and let it define them. The fix is to ask what "everything goes right" looks like and whether you are ready for it.
7. Build It Wrong, Fast. Viable is the dangerous word in "minimum viable product." Teams burn months getting to something shippable that is probably wrong anyway, and with AI you can now vibe-code the gist of an idea and get a reaction today. The worst reaction is not "no," it is "meh," a 5 to 7 that wastes everyone's time. Pincus's rule: build it wrong before you run out of time to build it right, then build it right once something proves out.
8. The Failure Machine. The fastest path to a winning idea is a machine that generates and kills ideas at the top of the funnel. Turning around Words With Friends, Pincus's team click-tested hundreds of ideas a day against one question, "what will our players thank us for?", and learned players wanted to feel they were improving each week. The game was projected to drop from 120 million to 79 million in revenue; it did 180 million instead. The original team of 20-something men had been building a fast, high-adrenaline mode for an audience of middle-aged women who wanted a calm escape.
9. Death by Compromise. Know your goal or suffer death by a thousand compromises. Pincus's pattern, and most founders', was compromising to land the next engineer, CTO, or brand-name investor, then waking up at a company they did not want to work at. His correction: the founder is the most valuable player, and if the company stops being the right place for the founder, the company has failed. Take half the valuation for all of the control.
10. Democratic Dictatorship. A company is not a democracy, and running it like one corrodes it. Pincus uses what he calls a democratic dictatorship: every voice gets heard, then he casts the single vote. A good CEO pulls intellectual honesty from everywhere and then makes the call, not the call that polls best inside the building. There is one chef.
11. The Moral Contract. If your builders take the hill you asked for, you owe them work as hard as theirs to turn it into value. Pincus watched engineers kill themselves to ship a product the CEO then abandoned or sales failed to sell. The contract runs both ways: show you are in the trenches unlocking the value of their work, and they will do more next time. Honoring it also means firing weak performers, because false democracy and "paying your dues" erode a meritocracy.
12. Mature Markets Are Not Mature. Pincus looks for markets everyone treats as finished that have barely started. Search looked mature before Google, which was the 56th search engine; the global video game industry was about 23 billion dollars in 2007 and considered slow. His bet with Zynga was latent demand from people like him who would play if a game asked very little and gave something back. The shift he names next is voice, because we will look back amazed at how much time we spent typing.

I don’t believe reality is a simulation, but you genuinely couldn’t script this timeline:
• Two weeks ago: At @swyx’s AI Engineer World’s Fair in SF, I decide at the last minute to introduce my friend @uri_rolls onstage for his talk on cyber benchmarks for infrastructure penetration and access control (see below, amazing team).
I say: “There is a future where cyber is alive and everyone is well protected and I’m pretty sure that future involves open-source models.”
And later: “A big challenge is going to be speed: the speed of attack versus defense. When an intruder starts to enter, you have to see what’s happening and catch them.”
• One week ago: @huggingface is hit by a sophisticated intrusion over the weekend. The traces look unlike anything we’ve seen before and suggest serious AI involvement, but we don’t yet know which model was used.
The closed models we ask for help choke on their guardrails. We need to react fast, so we turn to @Zai_org’s GLM-5.2 to help us analyze the attack.
• Earlier this week: @OpenAI reaches out, discloses what happened, and partners with us on the investigation.
The intruder turns out to be exactly what we had discussed two weeks earlier: a fully autonomous agent, powered by an unreleased frontier model, attempting to gain access to part of our infrastructure.
Sometimes the timeline we live in is genuinely vertigo-inducing.

AI is moving into the physical world.
We're excited about a new wave of startups rebuilding the systems that power the real world, from education and healthcare to defense, finance, infrastructure, and work itself.
http://ycombinator.com/rfs

Re AI-Powered Consumer Products for a Billion People
@raphaelschaad
Every major platform shift creates new consumer giants. But three years into the AI era, ChatGPT is still the only new icon on most people’s home screens.
As intelligence gets better and dramatically cheaper, nearly every consumer category opens up again: education, health, finance, entertainment, transportation, and how we connect with each other. Consumer is going to be so back.

Trump’s tariffs haven’t just hurt American business.
They’re about to be a growing drain on public finances as the government pays back tariffs ruled illegal by the Supreme Court.
My @Morning_Joe Chart

It’s a great feeling when a YC team that heard “no” from many investors later becomes the market leader.

Your ideas. Anywhere they find you.

The redesigned Replit mobile app is available now on iOS and Android.

I'm cheering for every founder and every engineer building low cost interceptors for missiles and drones. The free world needs you to win.
Will personally help in any way I can.
Y Combinator: The Future of American Defense
@SecArmy
Warfare is at an inflection point, and the Army is replacing the old playbook with commercially developed, modular technology that can be deployed and improved quickly.
We're looking for founders building low-cost interceptors,

Manufacturing jobs grew under Biden to a peak of almost 13 million.
Under Trump, they’ve declined despite his promises to “bring jobs back home.”
My @Morning_Joe Chart

Big news today. Science's PRIMA visual prosthesis, our retinal implant for macular degeneration, has received its CE mark, which means marketing approval in Europe. This is a huge milestone to begin to have a device like this commercially available to regular patients. 1/

Go from a pre-built template to a deployed Databricks App using @Replit's agent.
This demo shows how to open a Genie Analytics template from DevHub directly in Replit, build the app with Databricks data, test the Genie chat, and deploy it to Databricks. Once deployed, the app includes the built-in security and governance available with every Databricks App.
Request access to the private preview → https://replit.com/partners/databricks
Watch the demo and explore the template gallery on DevHub → https://developers.databricks.com/

We've passed 5 million users!!
It's crazy to see the compounding value of focusing on one goal...
To give every human superintelligence and make superintelligence more human
We could not do any of it without your support. Thank you so much for being there since day 1.
We're just getting started :)
Design Arena: Design Arena has surpassed 5 million users!
Since launch, we've grown from 6 arenas to 30+ and seen our community create millions of designs.
What started as a way to compare AI-generated design has become one of the clearest views into how people actually use AI models in the

Substack just shipped AI slop detection to protect readers and featured my writing as an example of how to use AI to edit without pumping out slop.
Basically, I work with AI during the middle 50% of my writing process:
→ First 25%: I voice dictate or write a rough draft without AI.
→ Middle 50%: I use AI and my /no-ai-slop skill to fix grammar, improve clarity, and test my arguments.
→ Last 25%: I go through the piece line by line and edit it manually.
The difference between pumping out slop and making something you’re proud of is giving a damn and staying in the driver’s seat throughout this 25/50/25 process.
📌 More here: https://creatoreconomy.so/p/use-my-no-ai-slop-skill-to-remove-20-ai-slop-patterns
Peter Yang: I’m sick of reading AI slop, so today I’m open-sourcing my /no-ai-slop skill that removes 20+ slop patterns from any piece of writing.
📌 Get the free skill here: https://github.com/petergyang/no-ai-slop
If you find it useful, please consider starring the repo so more people can find it.
Why I

🆕 Claude for Long-Horizon Tasks
https://www.youtube.com/watch?v=9QebvrrY3KY
@RLanceMartin returns for the third(!) time to discuss the four core themes for effective agents and claws:
1. Decoupling Brain and Hands (3:21)
2. Using Verifiers (6:20)
3. Self-Learning and Memory (10:11)
4. Global/Org-Level Harnesses (16:08)
Welcome back Lance!

We went from 0 to 1,000+ paid customers in our first 20 weeks by religiously following these @ycombinator 12 rules:
1/ grow retention before growth. we didn't run a single growth campaign until we hit 90% weekly retention. scaling a leaky bucket just means you fail faster
2/ ignore everyone who isn't your customer. i stopped reading competitor blogs, industry news, and most of twitter. the only company that can kill you in year one is your own
3/ manual is faster than perfect. our first 50 users each got a personal onboarding call from me. you can't automate what you don't understand yet
4/ validate before you build. we put a credit card gate on @origamichat on day 2. it's easy to convince yourself people want your product when nobody's paying
5/ ugly product > no product. we shipped a barely-working version to 30 people in a week. real users teach you more in 7 days than 6 months of polishing in the dark
6/ your customers write your roadmap. i still spend 3 hours a day on user calls. every feature we've shipped came from a conversation, not a planning session
7/ fire bad fits early. we've fired customers who were wrong for our product. 10 people who love you beat 1,000 who kind of like you. every time
8/ ship every week without exception. we push code to production every friday. consistency compounds in ways you can't see at first
9/ stay small longer than feels comfortable. we were 2 people for a very longtime. Headcount is not progress. every great startup was embarrassingly small for embarrassingly long
10/ know your burn, know your runway. we track default alive on a whiteboard every week. scary but necessary. most startups don't die from the market, they die from running out of time
11/ expect things to break constantly. we've had 3 near-death moments. the game isn't avoiding fires, it's how fast you put them out
12/ honest conflict > polite silence. brutal honesty with your cofounder is the cheapest insurance you'll ever buy
good luck and dont forget to have fun

Bullish @Replit

Prompt crafting is over-rated, ask for what you want
Matt Shumer: So another long-standing open conjecture was disproved by AI.
The crazy part is the prompts… basically:
- “do a breakthrough”
- “continue the search”
- “enough, do it”
Anyone will be able to make world-bending breakthroughs soon. Just point the model in the right direction.

LFG @SecArmy x @YCombinator 💪

The new @Replit mobile app feels so good. Proud to have worked on this with the Replit mobile/design team. Check it out on iOS right now - Android rolling out this week.
Replit ⠕: Your ideas. Anywhere they find you.

The redesigned Replit mobile app is available now on iOS and Android.

Pretrained ViTs see the world in rich, dense detail. Most policies pool it to a single vector before acting, discarding most of it.
We introduce Patch Policy: a minimal architectural extension that enables transformer-based policies to consume dense tokens directly, no billion-param VLM required. It outperforms a fine-tuned 7B VLA by 18% with ~0.7% of its parameters, enabling robust, precise manipulation.

The experience on the new Replit mobile app is so smooth.
Replit ⠕: Your ideas. Anywhere they find you.

The redesigned Replit mobile app is available now on iOS and Android.

🆕 ActiveGraph: The Log is the Agent
my talk from AI Engineer is live!!! 😆
https://www.youtube.com/watch?v=khVX_BUnEwU
it's about @activegraphai, an event-sourced graph runtime for building durable long-running agents:
- what it is
- code examples
- experiments
- benefits & surprises
please check it out, share with your friends who like to tinker at the frontier, and let me know if you have any Qs!

Most of you know these patterns that AI models make in writing but here’s a great open source list and a remover
Peter Yang: I’m sick of reading AI slop, so today I’m open-sourcing my /no-ai-slop skill that removes 20+ slop patterns from any piece of writing.
📌 Get the free skill here: https://github.com/petergyang/no-ai-slop
If you find it useful, please consider starring the repo so more people can find it.
Why I

A startup doesn't have to be your life's work, but if it is, you become much harder to defeat. You play the long game and you outlast everyone. Because practically everyone else is an opportunist by comparison.

A set of politically-connected growth investors are mobilizing to protect their bag in Anthropic. They will secure bans on chinese open source models, when no crime has been proven--or even alleged--and without any act of congress. All to protect a set of wealthy investors.
Treasury Secretary Scott Bessent: We support open-source AI and the innovation it unlocks. But open source is not open season on American IP. When PRC firms conduct covert, industrial-scale distillation attacks that cross the line into IP theft, sanctions and Entity List designations will be on the table.

🙋Attending QuakeCon and want to see the id Software Founders and Master Pancake?
🎟️Make sure you already have a valid QuakeCon event registration.
⏰Be ready to register on http://quakecon.org Friday at 10am CT / 11am ET for your QuakeCon 30th Anniversary Experience Ticket as quantities are limited!

Startups valued in the billions now do rounds in the hundreds of millions on safes. The unexpected versatility of this instrument shows what a good job Carolynn Levy did when she designed it.

News: Nearly 200 Silicon Valley companies, including Proton and Y-Combinator, are urging the Trump administration not to cut off access to Chinese open-weight AI models or risk crippling the next generation of U.S. startups. https://www.politico.com/news/2026/07/22/startup-founders-urge-trump-not-to-shut-off-chinese-open-weight-ai-01008992

News: Nearly 200 Silicon Valley companies, including Proton and Y-Combinator, are urging the Trump administration not to cut off access to Chinese open-weight AI models or risk crippling the next generation of U.S. startups. https://www.politico.com/news/2026/07/22/startup-founders-urge-trump-not-to-shut-off-chinese-open-weight-ai-01008992

We're excited to announce that Michael Kratsios (@mkratsios47) is speaking at Startup School!
He's the Director of the White House Office of Science and Technology Policy and the President's science advisor — the person shaping how the U.S. government thinks about and competes in AI. He previously served as the 4th CTO of the United States.
http://ycombinator.com/startupschool

Matt Shumer (@mattshumer_) really is living in the future.
He just showed me his setup. He is running a whole company of agents via workbench .md.
> agents working across laptops and cloud VMs
> one chief of staff checking in on each agent
> all the agents can communicate with each other via workbench
> Matt chats directly with chief of staff to get updates and share input
I've not seen anything like it. I need to up my agent game.