SwirlAI: nine places to plug decision models into agentic RAG, with runnable code
blaizedsouza · x · 2026-11-20
Aurimas Griciūnas (SwirlAI Newsletter) published a deep dive on where Decision Models fit in agentic RAG systems, with code and runnable notebooks.
Decision models, launched as a product category in September 2026, take state (text/JSON/images) plus code-declared answer options and return a probability distribution — no explanations, just low latency, low cost, high reliability. Two API shapes exist: TypeSafe AI System One and OpenAI Decisions API.
The nine use cases: model routing (cheap vs expensive tiers), per-hazard guardrails, retrieval sufficiency (answer now vs search again), passage filtering/reranking by probability, citation checking (supports/contradicts/doesn't address), tool-call gating (allow/escalate/deny), loop termination, context compaction (what to summarise vs drop), and rubric-based evals on traces.
More from coding & agent
- Open-source repo offers 500+ AI-generated low-poly 3D landmarks for maps under CC0 — tomjohndesign · 2026-11-20
- Agent safety rule: human-approve any irreversible, externally visible action — and show the exact action, not 'proceed?' — Beneficial_Use2116 · 2026-11-20
- Viral 'Claude Code Harness' article argues the harness, not the model, makes the agent — blaizedsouza · 2026-11-20
- Prefill beats decode: SGLang wins agent runs on DGX Spark despite slower tok/s — AIFlow_ML · 2026-11-20
- Running ComfyUI fully via Claude in a container — no manual UI anymore — xWranglerWranglerx · 2026-11-20
- AI writes all code to make a real robotic arm play chess, open-sourced — FinanceYF5 · 2026-11-20