Vivian Midha Shen
this essay is 14+ years old, and we're just getting more and more ambitious
Paul Graham: The startups YC funds now are much more serious than they were in the supposedly good old days. In the last couple days I've met with startups making optical switches, rewriting the software infrastructure of manufacturing, making nuclear reactors, and curing cancer.
Paul Graham
Are the people who were enraged by this tweet going to be even more enraged when I tell them I met a founder today who sometimes writes 50,000 lines of code in a day? Or is everyone now used to this?
Paul Graham: I met a founder today who said he writes 10,000 lines of code a day now thanks to AI. This is probably the limit case. He's a hotshot programmer, he knows AI tools very well, and he's talking about a 12 hour day. But he's not naive. This is not 10,000 lines of bug-filled crap.
Rafael A. Mangual
This is actually helpful, because none of these things are true:
(1) It’s not a “dragnet.” NYC probably has more coverage from ALPRs like Flock cameras than any city in the country. The city deploys somewhere in the range of 1,500 ALPRs (fixed and mobile combined). But the city has ~80,000 street segments. The cameras can see about 1,500 SqFt each, which comes to a total SqFt coverage of about 2.25M SqFt—a tiny, tiny fraction of NYC’s 13+B SqFt. And most agencies that use ALPRs like flock don’t store the data for more than 30 days, which means there’s not a whole lot to search.
(2) There are absolutely meaningful limits on these cameras, including the data storage limits I already mentioned. There are also internal policies (a) requiring officers to create trackable cases before they can search vehicles, (b) imposing serious penalties (criminal, and administrative) for misuse (which is why you have these stories of officers being fired and arrested for misusing these things to track their ex wives, etc.).
(3) Queries of these systems don’t constitute Fourth Amendment searches (and therefore don’t require warrants) because they only capture what is in plain public view, which you have no expectation of privacy in.
(4) There is no clear track record of abuse. What we’ve seen are about a few dozen cases of individual officers misusing the tech to track their ex-girlfriends, wives, etc. The fact that we know about these cases (which universally result in significant punishment—up to and including arrest and prosecution) illustrates (1) that there are indeed protections and limits in place (as I say above), and (2) that the concrete harms we’ve seen are limited to these sorts of one-off instances, as opposed to the kind of systematic mass surveillance programs alarmists are warning about.
Christopher Skelly: Flock is bad because:
It builds a searchable dragnet of Americans’ movements with persistent retention.
It operates without meaningful limits.
It allows searches without warrants or individualized suspicion in ordinary use.
It already has a clear track record of overreach
Are they going to send this to China too?
Andrew Curran: Bernie Sanders has written a letter to Sam Altman, Dario Amodei, and Mark Zuckerberg urging them to immediately pause all AI development in the interest of humanity. And he warns if they do not take appropriate action now, the US Senate will.
lily zhang
sign up for COLM 2026 before it sells out! for non-local researchers, here's the SF map by @swyx.
safety tip: i added poop alerts 💩 for the rough blocks. try to avoid booking hotel around union square and tenderloin, you'll thank me later. @COLM_conf
Conference on Language Modeling: The first batch of tickets will be released TODAY (Aug 10) at 1pm PT (pacific time). We are fairly certain we will be able to release more later on, so if tickets run out, we will direct people to a wait list (sign up if interested)
https://colm.eventhosts.cc/register
gpt luna max vs claude fable ultracode
sent "pls build a mostly faithful clone of grok imagine with open models via fal"
i woke up to these two and assumed fable was left and luna was right
i was wrong... it was the other way!!
objectively, fable did the better visual clone. but luna somehow understood intent better and created the more USABLE clone given my open model bent.
Deep alignment of your AI with you and your context is massively important
Important new work by @ibab and team
River AI: Today, we're sharing that River AI has raised $1.1 billion, led by @generalcatalyst and @amppublic with strategic investment from @nvidia and @AMD. Additional investors include @ycombinator and @Temasek.
We imagine a future where your AI works entirely for you and deeply aligns
Gabriel Synnaeve
Yesterday was my 40th birthday, and there is no best gift than Meta turning toward open source again, co-announced with a model release from my fabulous ex-team.
Mark Zuckerberg: Today we're also opening the weights for Muse Glimmer, a great 30B parameter dense model that can run locally. Soon we'll also release the weights for Muse Spark 1.2, our latest foundation model. Meta is a strong supporter of open source and I'm proud of these releases. Congrats
Hemant Taneja
We put together a $1.1B round for @river_ai_inc to build American leadership in open weights.
Frontier labs push the boundary of humanity, but open weights empower humanity, putting intelligence into everyone's hands.
@ibab has built frontier capability three times over. Now he's making it something you can own.
General Catalyst: .@river_ai_inc has a new vision for personal AI. We co-led River’s $1.1B funding round because we believe in @ibab's ambition to build open-weight models that individuals can train and improve continuously.
Watch @htaneja in conversation with Igor.
Chapters
02:22 — Co-Founding
.@MaorShlomo has been incredibly generous with his time helping me, I'll definitely watch this one.
Ollie Forsyth: The @base44 story is a totally wild one
@MaorShlomo founded the company in 2025. Grew it to $1M ARR in a matter of weeks before being acquired by @Wix for $80M - just months after launch.
Maor and Wix's Chairman, @marktluszcz join @NEWECONOMIESVC to share the latest trends
Nicolas Dessaigne
Unsloth is one of my YC companies and the team is cracked. Their new open source desktop app lets you run and train models locally, then connect Codex or Claude Code to them. Super cool!
Unsloth AI: Introducing Unsloth Desktop 🦥
The first desktop app to run and train models locally.
• Open-source. Runs on Mac, Windows and Linux
• Supports MLX, diffusion image/video, audio, GGUF
• Connect Claude Code and Codex to local LLMs
• 50% more accurate, self-healing tool calls +
Michael Grinich
AGENT NIGHT is tomorrow at The Regency Ballroom 🌉
The lineup: @swyx · @JayaGup10 · @Altimor · @davidcrawshaw · @bdougieYO · @abhiaiyer · @inazarovaLive
Live demos and sharp discussions, plus a sneak peek of something new from @WorkOS 👀
See you there! http://luma.com/agent-night
Jorge Alvarez
I found this really useful. I've removed all my generic skills. My skills only have business logic. Everything else I let the agent figure it out. I even removed info about our patterns and best practices. The agent is capable of inferring those from the current code.
Peter Yang: “Give [your agent] as little instruction as possible. Give it the tools to load context instead.”
Here’s my new episode with @thenanyu and @delashum from @Linear, where we walked through a real example of building a production agent from the initial memo to launch.
A few
justin
Introducing Capy v2: the fastest, most powerful cloud coding agent
We beat Claude Code, Codex, Devin and Cursor on DeepSWE while being 50% cheaper
Spawn RLM agents, each with up to 32 vCPU/128GB RAM and <1s VM boots
Bring your Codex/Grok subs
Comment for $100 in free credits
Ravid Shwartz Ziv
A story in light of the recent math AI results.
Last December I invited Julia Kempe (@KempeLab ) to the podcast and asked her what she wanted to talk about. She said: now that math has fallen, what should researchers run after?
This was last December. Models were nowhere near solving open problems. So I asked, do you mean we should talk about what AI can currently solve in math?
She said, no, Ravid. Math as we know it has changed forever, and so has the work of mathematicians. The problem is they don't know it yet.
Today that sounds obvious. Back then it was just one more example of a brilliant scientist seeing where things were going.
wei-wei
i did @ycombinator with founders of @WeAreLegora, @reductoai, and @greptile.
today, the W24 batch has raised close to $2b. two years ago, none of us had it figured out.
nobody in that room could tell you who was going to make it. not the partners, not the investors circling demo day, definitely not us.
everything i learned came from watching the other teams:
1/ the idea you apply with is rarely the one that wins.
half the batch looked completely different by demo day. some were still in pivot hell.
2/ you can't pick the winner from a demo either.
i was in a group with @reductoai and @raindrop_ai. incredible founders, but on paper none of us looked inevitable. the teams that took off learned faster, and the market they picked mattered less than how quickly they figured out they were wrong about it.
3/ the people matter more than the company.
the startup you meet in the batch might not exist in a year, but the person will.
some W24 founders are customers now. some shape what we build next. one is who i text when a six-figure deal falls apart and i need someone who actually gets it.
so if you're in the batch right now, be generous with the teams around you and stay in touch after demo day.
any one of them could become your best customer, closest advisor, or the person you text when everything goes wrong.
Been testing Grok bot and it shows alot of promise plus @jenny_wen recruited an amazing design team at Cursor to improve it over time.
The challenge with this and I'm sure upcoming products from the other major labs is:
1. How do you get regular users to trust sharing their credentials and logins with a remote computer?
2. How do you reassure them that this remote computer is secure and truly theirs to play with?
Regardless, I think the way that we'll all use computers is changing rapidly.
I wrote a post about this trend that I'll share in my newsletter tmr: https://creatoreconomy.so/
Grok Bot: Introducing Grok Bot, now in early beta.
Bots are AI teammates that do real work for you. They sign in to your tools, use them just like you do, and come back with finished work.
Jeremiah Krakowski
Re @mattshumer_ I used @mattshumer_ prompt generator too for it https://somethingbig.ai/gauntlet-loop/generator
There’s a cool meta to the “you’ve got this, try harder, keep going” prompts. That’s engineering management.
You know the outcome is within sight, so you decide to spend more time and money chasing it. You encourage. And you also need the wisdom to know you’re chasing a deadend and have to pivot and cut losses.
Agent management.
I’ve been testing Grok Bot for a couple of weeks, and I’ve been surprised by how much I’ve loved it.
Honestly, this feels like it could be the thing that gets millions of normal people using agents for the first time.
As many of you know, I haven’t always had the best experience with Cursor’s products, but this one feels different. The best way I can describe it is an agent for everything, not just code. The interface feels like iMessage, and you create bots that each have a job and actually get better over time as they learn how you work.
There are a lot of products attempting this, but the little details are what makes Grok Bot special.
For example, I set up a researcher bot and a writer bot, then made a Chief of Staff bot and asked it to get the other two working together on a project. I checked in fully expecting that to fall apart, because there was no way it just would work out of the box.
It worked out of the box.
Had a few more experiences like this too. The team clearly really cares about nailing the experience.
My only real complaint (which, if they nail it, will end up being a huge win) was the model router, which wasn’t great when I tested it. You don’t choose a model for your Grok Bot. It’s all done automatically on the backend. Incredible for regular users when it’s done well, but frustrating for power users when done poorly. I’m told they’ve made it much better since I tested.
It also clearly has sub-agents/workflows in the harness, as it runs Gauntlet Loops natively, and it handled everything I threw at it, from real work to 3D game builds.
Very much worth your time to check out!
Grok Bot: Introducing Grok Bot, now in early beta.
Bots are AI teammates that do real work for you. They sign in to your tools, use them just like you do, and come back with finished work.
shub
Matt responded to a complete cold DM to try out @bot, and has since been willing to give us a TON of feedback, all while pushing his bots to their limits. Highly, highly recommend following along
Matt Shumer: I’ve been testing Grok Bot for a couple of weeks, and I’ve been surprised by how much I’ve loved it.
Honestly, this feels like it could be the thing that gets millions of normal people using agents for the first time.
As many of you know, I haven’t always had the best
We were promised flying cars and all we got is infinite superintelligence for everyone
Rohan Paul
"Entire startups these days will be markdown files.
Own your skills because if you don't, your job becomes a skill file."
A super insightful talk by @garrytan (YC President & CEO) at Startup School 202.
Why we're entering the era of personal AGI.
AI has changed something deeper than productivity: it can turn a worker’s accumulated judgment into a durable asset that keeps operating without the worker.
----
From "Y Combinator" YouTube channel, (full video link in comment)
Rohan Paul: The boundary between a specification and a product is dissolving.
A Markdown file can now behave like a small software service when an AI agent can read instructions, use tools, and act on schedule.
"Do you know how many companies are at this event where their whole product
OpenClaw for the masses is here.
Matt Shumer: I’ve been testing Grok Bot for a couple of weeks, and I’ve been surprised by how much I’ve loved it.
Honestly, this feels like it could be the thing that gets millions of normal people using agents for the first time.
As many of you know, I haven’t always had the best
Replit ⠕
Our users run over 100,000 security scans per day with @semgrep
Replit has everything you need to build secure apps with AI
Semgrep: Millions of developers build on Replit - many with AI agents writing the code. Speed like that needs security that keeps pace.
Semgrep Guardian's secrets detection is now built directly into Replit, catching vulnerabilities the moment code is generated.
Every scan completes in
Build both the job centers and the housing
California will fall if we don’t do both
Brian Hanlon: Hearing the Legislature might expedite the California Forever shipyard and manufacturing park, but NOT the housing.
That would be disastrous for California.
Are we in a housing crisis or not?!
Malte Ubl
Given the state of AI we must internalize: Everything hackable will get hacked.
Two forces are at work:
▪︎ Near-frontier open weight models hack without refusal
▪︎ Frontier models help you defend yourself
How we're approaching this new reality ↓
https://vercel.com/blog/everything-hackable-will-get-hacked
Anish is a G. Excited for our convo to drop!
Anish Acharya: Tomorrow.
Very excited to share my conversation with the wonderful and singular @garrytan. No one has been more model-pilled, earnest, and invested in the collective success of our startup community.
Derek Thompson
This is nuts.
The whole thing about a housing crisis is that you don't have enough houses for all the people doing the jobs. I joke sometimes that people who cheer job growth but boo housing additions apparently assume that people who work prefer to not sleep in houses.
Expediting the jobs part of California Forever while cutting out the housing part is like turning my bad joke into policy.
Brian Hanlon: Hearing the Legislature might expedite the California Forever shipyard and manufacturing park, but NOT the housing.
That would be disastrous for California.
Are we in a housing crisis or not?!
Looks like @cursor_ai and @grok shipped this first
Peter Yang: ChatGPT Work seems one step away from the AI product I really want - a persistent personal computer and authenticated browser in the cloud.
This is how I understand Work "works" today (please correct me if I’m wrong):
1. It uses plugins to authenticate with apps like Gmail,
I think @X should ship this button or better yet make @CommunityNotes cover this use case.
Ish Verduzco: fav button on LI right now
Cool
Chrome for Developers: Imagine if your browser could handle the entire LLM loop natively.
Our latest developer guide shows you exactly how to combine the Vercel AI SDK with AI Elements to create smooth, responsive web chat applications. We cover parsing incomplete tokens, auto-resizing textareas, and
Armand Domalewski
san francisco built 405 units of housing this year so far...total...
but yes, clearly the problem is that SF is building too much housing 🙃🙃🙃
(((Matthew Lewis))) cults & consequences: Right now, San Francisco DSA and its allies are running a huge campaign claiming San Francisco has built "too much market-rate housing."
Literally, this is their campaign: "San Francisco is over-building."
But San Francisco has only met 17% of its market-rate housing goals.
Keep an eye on this view
http://vercel.com/ai-gateway/leaderboards/models
Vercel: http://x.com/i/article/2087305375460593664
