Subagents fail because summaries are lossy — checkpoint agents fork context instead
narphorium · x · 2026-11-06
Subagents are a top cause of AI agent failures: the main agent summarizes what it knows, the subagent works from that lossy summary plus assumptions, and returns another summary — the same flaw as compaction, causing unwanted emails or purchases made just to check a price. The proposed fix is checkpoint agents: fork the main session, which already has full (and cached) context, so the fork behaves identically to the main agent. The companion essay 'Structured Prompting' invokes Dijkstra's structured programming to argue for new abstractions for programming with natural language now that context windows exceed available context.
More from coding & agent
- First look at Google Antigravity: an agent IDE that opens with a one-line prompt — AI_Andrew · 2026-11-06
- Code-Graph-RAG: Queryable Knowledge Graphs for Codebases, 5.3k Stars on GitHub — JeremyCMorgan · 2026-11-06
- T3 Code x Grok Bot Combo Is 'Breaking Brains' With Its Speed, Says theo — zealcaiden · 2026-11-06
- Give Your Agent a Dedicated Import API: Fixing Repeated three.js/gltf Load Failures — msfeldstein · 2026-11-06
- Sentry CEO seeks testers for Junior, a multiplayer coding agent harness he calls market-leading — zeeg · 2026-11-06
- Mercury launches Agent Controls: AI agents handle finances via MCP, humans approve — dunkhippo33 · 2026-11-06