Anthropic research lead: prompting is over, 99% of engineers run 300+ self-improving agents
pstAsiatech · x · 2026-11-01
A talk by Anthropic's research lead is going viral. Her core claim: "prompting is basically over" — 99% of Anthropic engineers already run swarms of 300+ self-improving agents, and everyone is now shifting to graphs.
The evolution path she describes: single Agent → Loops → Plan Mode → Graphs → Self-Improving System. The key idea: the old workflow organized work around prompts; the new one centers on the graph around the model — orchestrating separate agents into a system that improves itself. The poster calls the 20-minute talk worth more than most $1,000 agent engineering courses.
More from coding & agent
- Dev adds chatting AI bots to multiplayer games, doubling as server load tests — chongdashu · 2026-11-01
- OCI MCP Server lets you manage Oracle Cloud via natural language in VS Code — modelcontextprotocol · 2026-11-01
- pipeworx-io ships a Salesforce MCP connector over Streamable HTTP — modelcontextprotocol · 2026-11-01
- Building an overlanding sim game entirely in Runway Agent + App Studio — c_valenzuelab · 2026-11-01
- Adobe's EditAgent picks the right tool like a Photoshop pro, at 10x the compute cost — The Decoder · 2026-11-01
- Dev builds playable voxel survival game Wildmere with Claude Opus 5.5 + Unity + Blender MCP — TamoghnaKabi · 2026-11-01