We have 16 partners funding 40 to 60 companies per year, usually amounting to more than 50 to 200 new SF residents per partner per year (depending on how many cofounders and how many already live here, but multiply by 16 and it’s a lot!)
Ankit Gupta: holy shit I am personally responsible for about as many people moving to SF this year than the entire city built housing units for
absolute embarrassment
Patrick Skinner - edu/acc
TLDR;
GBrain gives AI agents synthesis + gap analysis, a self-wiring typed knowledge graph, hybrid search, and a nightly dream cycle; all on top of your existing .md files.
Obsidian is great for human eyes, GBrain is built for agents.
Mark Marshall: @PSkinnerTech @garrytan @NessieLabs What’s the tldr on GBrain vs Obsidian?
kache
There's going to be a point where intelligence saturates meaningfully to be useful for humans, in which case open source will catch up to the closed frontier. Then, the battles will be fought elsewhere. Serving cost mostly probably?
MiniMax (official): Introducing MiniMax M3: The First Open-Weights Model to Combine Three Frontier Capabilities
- Coding & Agentic Frontier: 59.0% SWE-Bench Pro, 66.0% Terminal Bench 2.1, 34.8% SWE-fficiency, 28.8% KernelBench Hard, 74.2% MCP Atlas
- MiniMax Sparse Attention scales context to 1M
-
Ok is there any difference between Codex automations and Claude Code routines? Which one is better?
I want to consolidate all my cron jobs in one list.
Ankit Gupta
Iv been pondering this a bit recently. At some pt if the models are smarter than us are we going to be smart enough to tell the difference?
(assuming they remain sycophantic to people which is a big if)
kache: There's going to be a point where intelligence saturates meaningfully to be useful for humans, in which case open source will catch up to the closed frontier. Then, the battles will be fought elsewhere. Serving cost mostly probably?
Machina
Matt has been shipping some incredible products lately...
you should definitely equip your agents with:
- /last30days
- PrintingPress
- Agent Cookie
completely different experience with Hermes especially
Matt Van Horn: Introducing Agent Cookie. 🥷🏻🍪 For anyone running @OpenClaw or @NousResearch's Hermes on a Mac mini: I kept finding my agent logged out of everything, and it sucked. So I fixed it.
"Add this to my Amazon cart." Sorry, logged out again. "Order my usual on Instacart." Nope, not
Matthieu wyart
LLMs learn by predicting tokens. World models (JEPA, data2vec) learn by predicting their own abstractions. Which needs more data? For data with hidden hierarchy, we prove the gap is exponential. https://arxiv.org/pdf/2605.27734
Beautiful example of a full-stack agent on @vercel. Great learning material!
Pontus Abrahamsson — oss/acc: Introducing Caltext, open-source calorie tracking in iMessage.
Built with:
• Bun + Turborepo
• Hono on Nitro (Vercel)
• Chat SDK + Sendblue
• AI SDK + GPT-4.1 vision
• Upstash Redis
• Vercel Workflows
• USDA FoodData Central
Link ⬇️🧵
Nir Zicherman
Standardized testing is back?
What a fascinating timeline:
1. Long history of people pushing back on SATs and ACTs
2. Belief that GPA is a better way to assess a college applicant
3. COVID causes colleges to make testing optional
4. AI makes GPA significantly less valuable as an indicator
5. No one really knows what standard to use anymore for college admissions
6. College professors band together to push for more standardized testing
Rep. Mike Levin
Let me get this straight.
Howard Lutnick, Trump’s Commerce Secretary, got caught lying about his relationship with Jeffrey Epstein. Democrats grilled him about it and finally got him to agree to testify before the House Oversight Committee.
Four weeks later, Lutnick wrote a $5 million check to the super PAC that funds House Republicans, including members who sit on that committee. Then he showed up for a closed-door interview where, despite pushback from Democrats, no cameras or recordings were allowed.
This is the most corrupt administration in American history and it’s not even close.
https://www.nytimes.com/2026/05/22/us/politics/howard-lutnick-donation-house-republicans.html
Peter Yang
My top 6 takeaways from @Shpigford on how to build multiple products solo with AI agents:
1. Keep shipping even if the fear of embarrassment never goes away.
After 25 years as a solo builder, Josh still thinks "It's terrifying launching something. Every single time it's just like, what if zero people care?" But he ships early anyway: "The idea of spending months working on something before you put it out for other people to use, I think that's a real bad idea."
2. Charge from day one, and kill products that can’t pay for themselves.
For any product with built-in costs (e.g., hosting, LLM), Josh ships a paid version from day one. If the product can't cover its own costs, he doesn't hesitate to shut it down and refund recent payments.
3. Build features in parallel with separate git worktrees
Each worktree is a separate working copy of your codebase on its own branch, so features won't interact with each other. Worktrees stop context rot, isolate mistakes, and force you to test every chunk before moving on. Josh uses @conductor_build to manage them all.
4. Have GPT review Claude's work and vice versa
Josh builds with one model and then runs a review pass with another “GPT invariably finds three to five bugs that Opus overlooked.” Different models spot different mistakes. In Conductor, he sets a default review model so it runs on every PR.
5. Build a skill to make AI better over time.
Josh built a /learnings skill that runs at the end of each phase so that every "no, that didn't work" moment can be distilled into new rules to help the agent avoid repeating the same mistakes.
6. AI lets anyone ship, but real experience still matters.
Josh credits 25 years of building before AI for how fast he ships now with agents. "I know the general shape of how I want things to work, so I can very quickly get to that point." His advice for newer builders is to "just fail a lot, because the only way that you'll ever figure out what not to do is by doing the thing incorrectly."
Josh walked through his entire development workflow in our episode and shared more skills.
📌 Watch the full episode here: https://youtu.be/GdxLaeyu33c
Peter Yang: "It's terrifying launching something. I've launched 100s of products, and every time it's like...what if zero people care?"
Here's my new episode with @Shpigford, a solo builder who sold Baremetrics for $4M and is now building 5 products at the same time with AI agents.
We
http://x.com/i/article/2061176923531984896
Back in the Armadillo Aerospace days, our liquid oxygen loadouts were 50 to 100 gallons, and the tanks only weighed about a tenth of the propellant, so cooldown effects didn’t dominate. I have an effort going on now dealing with several-gram quantities of multiple cryogenic fluids, and it is an enormous pain when your valves and plumbing mass tens of times the fluid you are trying to deal with. I wish there was a viable cryogenic peristaltic pump!
Ankit Gupta
i suspect we've been in the mainframe era of AI computing and we're about to enter the PC era of it.
data centers are obviously still critical but oh man so much personal hardware and software is about to come
NVIDIA GeForce: It all starts with the @NVIDIARTXSpark Superchip.
RTX Spark reinvents the personal computer for agents, creating and gaming.
Learn more → https://www.nvidia.com/en-us/products/rtx-spark/
Paul Graham
Re @headinthebox I disagree. You should work on whatever you find most interesting. For some people that's databases. And there's nothing wrong with a 16 year old making another instance of something that already exists. That's what all the bands are doing too. That's what you do at 16.
Zach Coelius
I love that we are in an era of completely rethinking everything.
Garry Tan: http://x.com/i/article/2061176923531984896
Barr Yaron
Another week, another opportunity to fill out this year’s AI Engineering Survey! Raffling off prizes galore https://ntn.so/ai-survey
@aiDotEngineer @AmplifyPartners @NotionHQ @vercel
Latent.Space
🆕Grok Imagine’s Video Agent Moment: Cosmos, xAI, World Models, Generative UI, & the Codex Phase for Video!
https://www.latent.space/p/video-agents
@EthanHe_42, former @xai world model lead and @nvidia Cosmos researcher, explains why AI video may follow the same path as coding agents, how Grok Imagine went from zero to one, why text-to-video is only the autocomplete phase, how world models become real-time and interactive, why language models may become the control layer for video, and why the future of AI video may look less like a prompt box and more like an agent with a camera, editor, timeline, and tool belt.
Garry's List
The SF Zoo spent $12M without required approval, paid $1.6M to alleged relatives of senior staff without competitive bids, and stonewalled auditors until the Board froze funding.
Three weeks later, the city approved an $8.5M loan.
Not one supervisor asked a question.
The Zoo isn’t an exception. SF has outsourced essential services so deeply it can’t discipline failing nonprofits without disrupting the services it depends on.
https://garryslist.org/posts/san-francisco-s-nonprofit-trap
Ethan He
In @latentspacepod podcast, I shared my view on video generation, world models, LLMs, agents, continual learning and where the next frontier is.
1. Video models get most of their intelligence from language, not from video data.
2. Idea-to-code is fast now. The bottleneck is back to having enough compute to try every idea.
3. Iteration speed beats almost everything else in model development.
4. The next leap won't be a better video model. It'll be a video agent.
5. Diffusion will be the frontend of AGI, the LLM the backend. Generative UI will replace HTML/CSS: user intent straight to pixels.
6. Physical embodiment may become a tool a powerful AI picks up. Robotics may get solved by video-capable LLMs.
7. Continual learning may look like models that manage their own context, and even rewrite their own harness at test time.
Thanks @swyx and @vibhuuuus for having me 🙏
https://www.youtube.com/watch?v=jPtQlILfkhA
NYU Center for Data Science
Eero P. Simoncelli, CDS founding member and professor, was elected to the National Academy of Sciences, recognizing “distinguished and continuing achievements in original research.”
He studies how brains and machines represent visual information.
https://www.nasonline.org/news/2026-nas-election/
"Leadership is presence, not absence." —@bchesky
Nicolas Dessaigne: Everyone says: hire people better than you, then get out of the way.
One of my biggest mistakes at Algolia was taking that literally.
Yes, hire people better than you. But don’t confuse seniority with earned trust.
Stay close at first. Inspect the work. Pressure-test the
tigris
speaking ++ showing emotion alchemy at @Replit vibecon on June 17-18 in NYC! code! tech! art! and spike jonze!
ClaudeDevs
We've reset 5-hour and weekly rate limits for all users on Pro and Max plans.
We fixed an issue that caused some Claude Code sessions to spawn excessive parallel subagents, burning through usage faster than expected.
Kenneth Roth
Trump claims to have mastered the “Art of the Deal,” but he has just given us a master class in negotiating incompetence. I explain why the Iranians have the upper hand after his disastrous war of choice:
https://www.theguardian.com/commentisfree/2026/jun/01/trump-iran-war-plan?CMP=Share_iOSApp_Other
Zain Shah
Latest latentspace pod is excellent. @EthanHe_42 really gets it and lays out a lot of the thinking that led us to Flipbook and how to think about the future of generative UI
agents 🤝 video gen 🤝 users
swyx: This pod was an incredible gift to the community:
not only our first pod about @xAI, but Ethan really indulged on all our questions on how to train a SOTA Videogen world model, including specific areas (consistent extending/editing, voice) that Grok @Imagine is *still* SOTA,
Alpha Batcher
540,000 lines of Rails is a brutal way to learn the agent era
Garry's List shipped with:
> 262,000 lines of app code
> 276,000 lines of tests
> 127 background jobs
> 33 cron jobs
> a 1,778-line claim verifier
Then GStack hit 105k GitHub stars
Do not copy the Rails pattern
1. Finish the task with an agent
2. Say `skillify it`
3. Save the markdown skill
4. Keep TypeScript only for deterministic I/O
5. Add a unit test for the code
6. Add an LLM eval for the skill
7. Add an integration test across both
8. Add a resolver so the agent loads the skill next time
A Rails feature serves one app
A skill pack can run against the next hackathon, repo, customer inbox, research folder, or QA run
Audit your repo this week
Cut code that only exists to:
> re-check model output
> retry the same model call
> validate what the model already has context for
> schedule work because the agent has no reusable procedure
> preserve a one-off prompt in chat history
Keep the code that touches files, APIs, auth, billing, devices, databases
Move the judgment into markdown
Test the markdown
Reuse the skill
Garry Tan: http://x.com/i/article/2061176923531984896
Gene Kogan
Very excited to be giving a workshop at @Replit's Vibecon on June 17 in NYC.
I'll be sharing techniques for getting coding agents to make music and art.
https://vibecon.ai
This Week in Startups
How a small xAI team shipped a state-of-the-art video model in 3 months:
🦾 Strong talent with a shared goal
🤝 One sync per day
🏗️ All the rest of the time building
Less coordination, more compute, and more iterations.
Git is all you need. Always has been
shadcn: You can now turn any GitHub repo into a registry.
Drop in a registry.json. Define your items. Install with the CLI.
Distribute components, utilities, config, docs, rules, design systems, workflows, agents, skills and more.
Replit ⠕
Can you build a real business for free with a single prompt?
Starting today on Replit, the answer is yes.
From a single prompt, get a website, mobile app, slide deck, and launch video.
Plus unlock perks from @stripe @atlas, @QuickBooks, @mercury & @doolaHQ
Kathryn Wu
I think the biggest thing YC changed for me was reducing noise.
Before YC, startup advice on the internet felt very random.
Everyone sounded smart.
Everyone had opinions.
A lot of it was bluff.
YC made startups feel much more grounded:
ARR, retention, distribution, market size, growth.
There’s a much more standardized way to evaluate whether a business is actually working.
At the end of the day, the best founders I’ve met are usually just trying to build real businesses with real demand.
Not optimize for appearances.
Tyler Bosmeny
One of the craziest videos I've ever seen.
117 hidden Ukrainian drones spontaneously launch out of trucks and obliterate an entire airfield (41 aircraft)
Not even near the front lines - in Siberia!
Volodymyr Zelenskyy / Володимир Зеленський: Today, I presented state awards of Ukraine to the people without whom Operation “Spiderweb” would not have happened. We have no right to name them – everything remains absolutely classified and will remain so for a long time. This is already a historic operation by the Security
The OpenAI Foundation is doing a lot of wonderful things.
Helping society become resilient to AI is going to be incredibly important. Much more to come here!
The OpenAI Foundation: AI is advancing quickly. Society’s ability to manage its risks must advance just as fast.
Today we’re sharing our vision for AI Resilience, with more than $130M in initial grants underway across bio-resilience, cyber-resilience, AI model safety, and AI’s impact on young people:
whoah - Grace + Blackwell chips in a laptop.
@Microsoft + @NVIDIA teaming up to take on 6 years of total dominance of Apple Silicon
Want to try GStack /office-hours for your product idea as quickly as possible? It's one click here:
https://app.lightsprint.ai/office-hours
Aaron Epstein
Common fundraising deck mistake: using boring slide titles like "Team" and "The Problem".
Instead, state the conclusion that you want investors to take away from the slide, and use the body as supporting evidence.
If an investor only read the titles on your slides, would they understand your business and want to invest?
Prompt to business:
- website
- mobile app
- monetization
- Delaware corp
Start for free:
Replit ⠕: Can you build a real business for free with a single prompt?
Starting today on Replit, the answer is yes.
From a single prompt, get a website, mobile app, slide deck, and launch video.
Plus unlock perks from @stripe @atlas, @QuickBooks, @mercury & @doolaHQ
Elias Reyes
Built https://dialmoments.com/ (ios app and website) with replit.
This has changed the game.
Thanks to the earnings I have made in 2 months, me and my wife will be going to watch Spain in their opening game for the world cup and I paid off a credit card I had😁
Amjad Masad: Prompt to business:
- website
- mobile app
- monetization
- Delaware corp
Start for free:
my #1 most used skill lol
OpenAI is now available for enterprises on Amazon Bedrock:
OpenAI: OpenAI frontier models and Codex are now generally available on AWS, giving enterprises a new way to build on Amazon Bedrock with OpenAI through the security, compliance, and governance workflows they already use.
This is also the beginning of a broader expansion of OpenAI
MiniMax M3 is now the leading open model on the Next.js agent evaluations (http://nextjs.org/evals).
Right behind Opus & GPT5, but 10× cheaper (And 20× cheaper right now on ▲ AI Gateway!)
Vercel Developers: MiniMax M3 is available on AI Gateway.
MiniMax's first long-context model, with support for multimodal inputs.
50% off for the next week.
𝚖𝚘𝚍𝚎𝚕: '𝚖𝚒𝚗𝚒𝚖𝚊𝚡/𝚖𝚒𝚗𝚒𝚖𝚊𝚡-𝚖𝟹'
https://vercel.com/changelog/minimax-m3-on-ai-gateway
