🤖 AI Builders Digest

OpenAI VP Science Kevin Weil points out that AI is not only solving more open problems—its proofs are getting more elegant as the models improve.

This observation highlights the rapid maturation of AI in areas like formal mathematics and reasoning.

Box CEO Aaron Levie argues that the ultimate rate limiter on productivity gains from agents will be critical stuff like security, compliance, governance, and the ability to review the work of the agent.

We’ve been living in a little bit of la-la land around how much software enterprises are going to ultimately want to vibe code themselves. There’s no free lunch with AI productivity. Companies will have to build up the systems, processes, and controls for ensuring that agents can’t run around and do anything they want on any data at any time.

The Takeaway: AI manager agents will soon power one-person billion-dollar companies by automating entire operations from product development to support and sales.

Andrew, the founder and CEO of The General Intelligence Company of New York and on his second startup after dropping out of college at age 18, is building tools to make organizations run themselves with AI. His vision involves one person at the top with vertical agents handling customer support, finance, marketing and more, coordinated by manager agents — the 'middle management' layer his company is focusing on.

They launched co-founder in September as a powerful cloud AI assistant with advanced memory that can draft emails sounding just like you and automate various operations. It has been used by over 8,000 startups. But realizing an assistant alone isn't a full company, they defined the core functions: create a product, sell it, support it, and operate it.

To automate product creation, they created co-founder CTO, launching in alpha today as the first fully autonomous engineering department in the cloud. You feed it a feature request or bug report, and agents use Claude to code, browser agents to test in Vercel preview environments, auto-merge PRs, monitor infrastructure, and even self-fix deployment issues. No local setup required. Andrew's team is 400% faster than before, averaging 25 PRs per engineer per day, with token costs sometimes exceeding salaries. 'This shows that we're starting to shift our human capital to intelligence,' he said.

Over the next six months, customer support agents and a full revenue and sales stack (CRO) will follow, realizing the complete one-person company.

Swyx wonders what happened with no Codex growth in March and shares a chart.

As an AI builder affiliated with Cognition and other projects, this observation points to possible shifts or plateaus in adoption of OpenAI's leading coding AI tool.

South Park Commons General Partner and Bevel Health Co-Founder Aditya Agarwal looks back at early Facebook's philosophy of "fixing bugs in production," which sounded crazy to most people.

Maybe the modern version of that is, just let your source code be public. If software is so dynamic, maybe this isn't so crazy. Or maybe it is totally batshit crazy.