Expanso's Split Architecture: Deterministic Pipelines With an LLM Judgment Layer
aronchick · x · 2026-09-20
Josh Rosen argues many 'deterministic' software decisions exist only because inference used to be too expensive. Expanso's article shows a concrete split: shape/fingerprint/count/route stays in deterministic nodes, while judgment questions (actionability, severity, owner) go through one readable HTTP call to the LLM, returning typed answers — demonstrated across ten real pipelines.
More from coding & agent
- An inbox re-ranked live by importance instead of time, built with jev — xkonjin · 2026-09-20
- DAIR.AI launches auto-updating Awesome Jev collection, curated by Jev itself — omarsar0 · 2026-09-20
- Open-source Tampermonkey scripts add paste-to-upload for WeChat, Xiaohongshu and Douyin — vista8 · 2026-09-20
- User reports OpenAI's Astra burning through limits faster, 20x plan exhausted in 2 days — iamrobotbear · 2026-09-20
- Jeffrey Emanuel ships bend2-mega-skill: a Claude Code skill to write, prove and port Bend2 code — pbaylies · 2026-09-20
- typesafe-ai ships open-source agent skills pack bringing typed decision loops to Codex, Cursor and Claude Code — multiply_matrix · 2026-09-20