The New Company
Session Overview
The new company keeps humans thinking and millions of agents working. We’ll trace how specs evolved from aligning humans, to aligning models, to generating code, and now to redefining how companies operate.
Humans still do three things - define, communicate, verify - but even these are amplified by AI to help us think, express, and check crisply at scale. We’ll show the infrastructure: how specs become executable programs, evidence‑first verification, tracing and trials, SLOs and blameless post-mortems.
You’ll leave with a vision for how agents reshape roles and org design, plus templates to iteratively apply in order to scale yourself and your company to 100,000-1,000,000 agents, often with single‑digit humans.
- Understand how to amplify human judgment with AI without turning people into liability sinks.
- Import systems thinking into the org level to run 100,000-1,000,000 agents with observability, tracing, trials, SLOs, and blameless postmortems.
- Enforce observability to measure mean time to autonomy, cost per verified outcome, and trust SLOs.
Speaker
Sean Grove builds systems that turn human intent into reliable software. At OpenAI he spent years in post‑training and alignment reasoning, helping develop model‑spec infrastructure and deliberative‑alignment techniques. He founded OneGraph (acquired by Netlify), was Principal Architect at Netlify, and now builds Linzumi.
His recent talks “The New Code” showed how specs turn intent into executable systems. “Riding the Token Wave” showed the need reshape both infrastructure and problems to harness tokens as a power source, and now this talk tackles the next rung: operating the million‑agent company with single‑digit humans.