AI Builders
Digest

OpenAI VP Science Kevin Weil pointed out a key advancement in AI capabilities. Not only is AI solving more open problems, but its proofs are getting more elegant as the models improve. This reflects how rapidly AI is enhancing both its problem-solving power and the sophistication of its reasoning.

Box CEO Aaron Levie shared a contrarian take on the hype around AI agents. He believes the ultimate rate limiter on productivity gains from agents will be on critical stuff like security, compliance, governance, the ability to review the work of the agent, and ensuring compatibility with regulations. Enterprises have been living in a little bit of la-la land around how much software they'll want to vibe code themselves, and recent events show why you won't take on every risk. There's no free lunch with AI productivity—companies will need 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.

Swyx, AI builder affiliated with Cognition, Temporal, and Latent Space, wondered what happened with no Codex growth in March. He shared a chart highlighting the unexpected plateau in usage of this AI coding tool.

The Takeaway: One person will soon be able to build and run a billion-dollar company powered by AI agents that autonomously handle product creation, sales, support, and operations.

Andrew, CEO of the General Intelligence Company of New York—his second venture after becoming CEO at 18 and dropping out of college—is dedicated to making self-running AI organizations a reality. His approach focuses on the middle management layer of agents that coordinate specialized lower-level agents for tasks like coding, customer support, finance, and marketing, while leveraging other tools for specific functions.

Their initial product, Co-founder, launched on September 8, acts as a cloud-based AI assistant with advanced memory that mimics your style for drafting emails, automating sales outreach, sourcing LinkedIn candidates, and more. It has been used by over 8,000 startups to automate operations, providing valuable insights into building robust agents.

To scale beyond an assistant to a full company, they defined the core loops of a product business: create the product, sell it, support it, and operate it. Today marks the alpha launch of Co-founder CTO, described as the first fully autonomous engineering department running entirely in the cloud. It uses manager agents to delegate to Claude-powered coding agents, browser agents for testing, and Vercel for deployments and infrastructure monitoring. A feature request triggers the full lifecycle—coding, PR creation, automated testing, merging, and self-healing for issues—all without local development.

Internally, the team has become 400% faster, averaging 25 PRs per engineer per day, with token costs sometimes surpassing salaries. 'We've started spending more on tokens than on salaries depending on the day,' Andrew shared, signaling a shift toward intelligence capital. Upcoming additions include customer support agents and a complete CRO revenue and sales stack by summer. Early access is available via a sign-up form for those lucky enough to join the limited alpha.

This vision turns software building into a video game-like experience inspired by StarCraft and Civilization, where one founder commands AI teams to scale massively.

General Partner at South Park Commons Aditya Agarwal, ex early engineer at Facebook and CTO at Dropbox, looked back at early Facebook's philosophy of 'fixing bugs in production' that sounded crazy to most people. He suggests the modern version of that might be just letting your source code be public. With software being so dynamic today, this approach could make sense, though he notes it might be totally batshit crazy.