System 1 vs System 2 agent harnesses: bounded judgment vs open-ended planning, explained
blaizedsouza · x · 2026-09-28
A clear breakdown of the two ways to put AI inside an application:
- System 1 harness: the model makes bounded judgments (category, score, route, extracted field) from choices supplied by code; app code checks confidence and policy before executing or escalating. Best for intent classification, model routing, risk gates, and reranking — high-volume decisions where the output space is known upfront.
- System 2 harness: the LLM receives a goal and constraints and plans its own path, suited to work whose steps can't be specified in advance.
The core decision factor: whether the possible outputs are known ahead of time and who owns the workflow. System 1 buys speed and control; System 2 buys flexibility at the cost of determinism.
More from coding & agent
- ZergRouter launches: one dashboard to route models and track spend across multiple coding agents — idanbeck · 2026-09-29
- Paper-proposed sandbox-and-interception AI safety mitigation already implemented in verifiers v1 — xeophon · 2026-09-29
- WebMCP lands everywhere in one morning: Meta wearables, Shopify checkout, Cloudflare — jeff_weinstein · 2026-09-29
- A 9-feature roadmap for building AI agents with Claude Code, from CLAUDE.md to MCP — MaryamMiradi · 2026-09-29
- Zeeg: Paper won't work with WSL, seeking structured visual design agent workflows — zeeg · 2026-09-29
- OriginTrail ships DKG V10.0.19 on mainnet for faster AI agent context graphs — melnykowycz · 2026-09-29