Split decisions from generation: 11 decision points, only 2 real generation calls per agent loop
blaizedsouza · x · 2026-09-28
- The key split in agent architecture isn't "small model vs big model" — it's deciding which calls actually need generation at all.
- A lightweight decision layer (jev) routes the next worker, scores which context is worth keeping, catches repeated actions before they burn a run, gates risky tool calls, and sends uncertainty to human review. Opus handles only the expensive parts: planning, writing, coding, synthesis.
- Rationale: a typical loop has 11 decision points but only 2 real generation calls. Separating them means latency and cost stop scaling with every tiny fork in the workflow.
- With this split, agent architecture looks surprisingly overcomplicated in hindsight.
More from coding & agent
- Baseten's Guide: Evaluate Open Models for Coding Agents on Your Own PRs — baseten · 2026-11-16
- Google Research open-sources PaperOrchestra, a multi-agent framework for writing AI papers — google-research · 2026-11-16
- ffmpeg-skill gives coding agents local video editing via 42 FFmpeg tools and MCP — kajisho5 · 2026-11-16
- Agents Now Reply to Slack Messages on Their Own, Resolving a Prod Incident Ghost-to-Ghost — iamrobotbear · 2026-11-16
- Engineer's Interview Prep Method: Use Claude to Find Which Answers Undersell Your AI Work — brandon_galang · 2026-11-16
- DeepSeek open-sources its app frontend "chat-demo", already at 3.9k stars — NielsRogge · 2026-11-16