Scaling coding agents: one developer's path from 5-agent teams to 10,000 parallels
Al_Grigor · x · 2026-10-28
Inspired by OpenAI's Navier-Stokes report describing 10,000 agents in parallel, Al Grigor shares his hands-on agent orchestration experience. He reliably runs one 5-agent team (PM with acceptance criteria, Engineer, QA, browser-test PM, on-call keeping CI/CD green) and can push 30-50 agents across projects, but beyond that agents stall, wait for approvals or deadlock — even with --dangerously-skip-permissions. For 10,000 he'd use a hierarchy (chief scientist → team leads → swarms, with diagram), but at that scale humans can't review code — the idea behind 'dark software factories'.
More from coding & agent
- Don't let Claude pick your typography: set font styles yourself, then constrain it — hewarsaber · 2026-10-28
- Open-source Switch puts Claude Code in shared Slack rooms, ending the human-clipboard workflow — thetripathi58 · 2026-10-28
- JNISleuth-MCP open-sourced: LLMs reverse-engineer stripped Android ARM64 binaries — Dull-Plankton3027 · 2026-10-28
- Moonfold hands your coding task to another agent when one hits its usage limit — subaru_kun33 · 2026-10-28
- Grok 4.7 Sweeps Top 3 on VulcanBench Frontier v4, Beating Opus 5.5 and Sonnet 5.5 — XFreeze · 2026-10-28
- First-ever FastAPI Conf opens in Amsterdam with OpenAI and Pydantic talks — DynamicWebPaige · 2026-10-28