Update: I started a new company! As AI makes it easy to remix any content, a new type of role is emerging: the media engineer. Their job is to build systems that continuously turn a company's work into a steady stream of media.
Personal update: today we’re launching Mission Street Capital.
@SurbhiTodi and I met as housemates at Mission Control, the oldest hacker house in SF, where we lived, breathed, and cooked with the best founders well before anyone knew them.
As hacker houses rise in popularity, they’re becoming talent clusters for exceptional hackers to meet and work together.
Our fund’s thesis is in identifying brilliant oddball builders emerging out of these talent clusters. Builders recognize other builders.
Our fund is backed by hacker house alumni, technical leaders from OpenAI and Anthropic, and GPs, partners, and emerging managers at the most iconic funds (a16z, CRV, Emergence, Uncork).
Thank you to our LPs including @AriX@thedanigrant@KevinUnkrich@tanzealous@ashvinair@mgrczyk@nikhilthota@SoheilK@jamdac@caitlinbolnick1@anamitra@DOitChoco@v_maini@aaronligon@crapshoot101@PachexMix
every AI tool i use has a kill switch someone else controls. but not my coding agent.
introducing https://t.co/JzNpr2iev1: open source coding agent/harness that runs on your own computer.
sign up and i cover 50M deepseek v4.1 flash tokens a week so you can try it out for free.
afterwards you can bring your own keys or subscribe to a paid plan.
run it, break it, tell me. share it with a friend and u both get 10M free tokens.
I lead product for Claude subscriptions. Today we're renaming our Max plans, and I want to explain why.
When we launched our top plan in April 2025, it had a single limit: 5 hours. It gave you 20x Pro's limit, so we called it Max 20x. When we added weekly limits to Max 20x, we set them to 10x Pro's (the 5-hour limit stayed at 20x).
I get that some new users assumed 20x applied to both limits. We could have done better here, and I'm sorry.
The Ultra plan keeps the weekly limit (10x Pro), but we're removing the 5-hour limit so you can keep building without getting blocked mid-session. This starts today and rolls out in stages. To help prevent fraud and abuse, a temporary limit applies until we've verified your account. You'll get an email when your limit is removed.
We've got more cooking and can't wait to share it. Thanks for building with Claude 🙏
The 512GB memory variant of Mac Studio Ultra that will come out sometime this week might be decent. I estimate it can handle 700B models 4-bit quantized at ~20tok/s.
My opinion: This is much slower than what you get on Nvidia hardware, an order of magnitude more expensive than paying per token in the cloud if you manage to run it 24/7 at max capacity, you end up quantizing, and you can’t even fit the flagship >1T parameter open weight models on it
To be honest, most people I know who use open models have moved to the cloud because they don’t care about on-prem enough to spend >$1M in infra for it
I no longer get frustrated that LLMs feel dumb. Until Fable and Opus 5.5, some tasks were beyond LLMs no matter how hard I pushed. Not anymore. These two models feel smart enough for everything I want them to do.
What slows me down now:
- Not knowing my context and preferences
- Permission prompts and refusals
- Passwords and 2FA in the browser
- Inexperience (no RL data for some domains)
- Claude speak and UI slop
All but the last seem solvable with good engineering, no further AI breakthroughs needed.
I no longer get frustrated that LLMs feel dumb. Until Fable and Opus 5.5, some tasks were beyond LLMs no matter how hard I pushed. Not anymore. These two models feel smart enough for everything I want them to do.
What slows me down now:
- Not knowing my context and preferences
- Permission prompts and refusals
- Passwords and 2FA in the browser
- Inexperience (no RL data for some domains)
- Claude speak and UI slop
All but the last seem solvable with good engineering, no further AI breakthroughs needed.
Can't say it loud enough...
CodeRabbit Secure is running GPT CYBER MODELS + an ensemble of models to help secure your codebases.
And the first $1,000,000 of scans are FREE.
Secure your applications and services right now. 🙏 Couple clicks and you will thank yourself (and maybe freak out a bit...it's scary good)
Great question. My rough framing is that AI has given engineering a 100x boost, product 10x, and design 5x. That imbalance creates real strain: engineers can merge lots of individual changes while the overall UX gets less coherent, and designers burn out trying to keep up.
Design is a deep craft. It takes thought and a view of the whole experience. We haven’t given designers tooling that matches how quickly we can now produce code.
That’s why we’re building Inspect for designers. The ambition is to give them the 200x boost. Then engineers can be the ones trying to keep up :)
With the current wave of always-on personal agents (Muse, Instinct, Grok Bot, Dots), the question is how many of them you’ll want. I think there’s room for only one proactive agent: