Anthropic researcher: 99% of engineers now run swarms of 300+ self-improving agents
AlishaOutridge · x · 2026-09-11
A widely shared account from an Anthropic researcher: a team member spent eight months babysitting a single chat window (prompt, read, fix, repeat), then stopped supervising and built loops that check their own work — out-shipping his own team six weeks later.
Key points:
- 99% of their engineers now run swarms of 300+ self-improving agents; the other 1% still paste error logs into a chat box by hand.
- The gap isn't prompting, it's looping: give the model a way to verify its own output — "it takes one evening to close."
- The playbook: Claude → Loops → Plan Mode → Dynamic Workflows → Self-Improving Swarms, i.e. Plan → Run → Verify → Improve.
Includes a 20-minute talk on building a self-improving model workflow from scratch, billed as worth more than most $1000 AI engineering courses. The quoted article covers "The 3 A.M. Problem": five ways unsupervised trading loops quietly destroy accounts, and an architecture designed to survive them.
More from coding & agent
- MathModelAgent gains traction: auto-solves math modeling and writes a submission-ready paper — jihe520 · 2026-09-11
- alphaXiv open-sources OpenResearch to run parallel research agents with any model — alphaXiv · 2026-09-11
- DeskcommCRM: open-source AI sales CRM with native agents and WhatsApp hits 1k stars — melgarafael · 2026-09-11
- hyperresearch: agent-driven knowledge base that turns web research into a searchable wiki — jordan-gibbs · 2026-09-11
- Forter's 13 lessons from its agent sprint: skip custom RAG, lean on mature enterprise search — bibryam · 2026-09-11
- Two real 'company brains' opened up live: Gorgias' in-house Cortex vs Slite — femke_plantinga · 2026-09-11