thoughts on a week in sf

hiring vibe coders vs engineers

May 06, 2025

I write a newsletter about startups and investing—for ai builders of all levels.

I record mini-tutorials, review tools I’m testing, share my insights and give you a peek behind the digital curtain from an exited founder turned investor.

Hey folks,

back in the uk after a week in SF, and not feeling jet-lagged (yet..)

i have so many thoughts and ideas from the week but many im going to percolate on a bit longer.

but here’s some of my quick thoughts:

US vs UK mindset. it’s full of optimism and thinking about the future. there’s obviously opportunists in SF but they’re easy to spot. if you meet the right people and have the right conversations there’s so much to be excited in and interested by. if everyone’s drinking the kool aid, i don’t think that’s a bad thing (in moderation). we could use more optimism and founders especially (from friends, family, investors, peers).
every dinner here is talking about companies, building, ai etc - yes that’s eye-roll for many but for people in the space, its interesting and energising. you come away with ideas, thoughts and more questions
no one has it figured out. everything is still so new and everyone’s building on the models that change a lot
most user issues of ai products are down to….you guessed it, model behaviour
companies customer profiles are very frothy. lots of different people are trying tools, making it harder for teams to spot their true, biggest personas to target.
overbuilding is happening. the smartest people know this and act accordingly (read: how to not build bad AI software).
the api business is phenomenal
you basically have to go around town with curiosity and figure out who you agree/disagree with, get different opinions and decide for yourself

and thats what i did. i went around asking questions, putting my thoughts out there, hearing others and figuring out my view on it all. i had the best time.

thanks to everyone i hung with!

Transamerica Pyramid, factory.ai offices, yc hq, cursor power-user meetup

🔎 News worth knowing

OpenAI is turning its profit branch into a Public Benefit Corporation (just like Anthropic and xAI). The non-profit board of OpenAI will still control this PBC and be a large (not major) shareholder. Also, OpenAI is likely buying WindSurf for $3B. (even though i thought it wouldn’t…)
Apple and Anthropic are in cahoots to make X Code (Apple’s own IDE) a “vibe coding” app. Also, Claude has web-search access for all paid plans and the web app can now connect to MCP servers via a new feature called “Integrations”
Grok can now create PDFs. I really like it for deep research answers because the PDF has an index that makes reading through it easier.
Visa and MasterCard are making it easier for agents to buy things for you. There’s a tweet showing Visa hiring vibe-coders but its actually hiring engineers that just know the latest tools and how to use them - big difference.
Lots of AI tools say they can build apps. Gadget’s actually can. It’s a full dev environment + a Shopify-savvy AI assistant that writes code, debugs, connects to Shopify, and ships. If you've tried building Shopify apps, you know that's impressive. No boilerplate. No context switching. Just build.*

*sponsored

want to partner with us? Click here

🌐 What I’m consuming

Gemini 2.5 Pro finished the game Pokémon Blue. The secret behind gemini’s performance is a good harness. Think of a harness as the app/system where the ai model is plugged in. In this case, the harness provides gemini raw data in addition to images from the game (valid, not cheating).
- o3 in chatgpt has an amazing harness with many tools—web search, image cropping, running code, etc.
- Manus is so good because it can read/write documents and get its thinking out of its head.
The harness is built by the developer based on the "game you want to play” (task to be completed) and makes the ai model significantly more powerful at that game.
Build a WordPress calculator plugin in 30 mins W/ Cursor, V0, and Google AI Studio, from one of our community 🙌
Andrej Karpathy’s review on his latest vibe-coded project
The rise of Cursor: The $300M ARR AI tool that engineers can’t stop using (now valued at $9B)
A public CEO’s internal email about being an AI-first company
MCPs enabling better support agents for Intercom
LLMs code by brute force, we shouldn’t be forcing them into structured code. Let them write whatever code they need to. I’m seeing this convo happen more and more.
ChatGPT rolled back its ‘overly yes-man’ personality, and reviewed what went wrong
o3 is really good at Harvard Business School cases, and at translating greek poetry. These might be good ideas to build a wrapper—Jenni AI does millions in ARR, helping students write research papers. (and no I’m not stopping using em dashes just coz chatgpt uses them)
Logan and a Gemini researcher talk about long context in AI models: RAG, making the whole of it useful, cost and what’s next. tldr for that: cheap long context is coming first, then 10M context, 100M needs more research.
three highlights about the demand for LLMs on openrouter - new models get adopted fast, they replace old models as well as expand the market, many apps use multiple models (from different labs too).
Things Theo loves/hates about every AI model API.
Tips for prompting to get good and accurate design from AI models.