Agents Don't Call Tools: The LLM Only Generates Text You Parse and Execute
ariG23498 · x · 2026-10-28
ariG23498 clears up a common misconception about agents: the LLM itself never calls tools — it only generates text.
- An LLM's interface is chat: you provide input (text, images, etc.) and it answers token by token.
- In an agent setup, the LLM is the brain and tools let it "do things" — but the model merely emits text indicating which tool it thinks should be called and with what arguments.
- The actual execution is done by the engineer's parsing layer: read the tool name and args, then invoke it, akin to spawning a process.
Takeaway: the better your parsing and calling layer, the smoother the whole agent system runs.
More from coding & agent
- DeepMind: AI agents spot bad advice 92% of the time—and follow it anyway — alex_verem · 2026-10-28
- Delta brings real-time multiplayer collaboration to Claude Code via a share link — xiaohu · 2026-10-28
- JetBrains Junie Local ships for RTX Spark and Windows, runs coding agent token-free — kimmonismus · 2026-10-28
- Metatron: a System 1 controller that lets coding agents command humanoid robots — ChongZzZhang · 2026-10-28
- Two days tuning one Tesla P40: 35B MoE with full 262k context at ~60 t/s — sargetun123 · 2026-10-28
- OpenClaw's web UI now works offline, thanks to smart websocket reconnect logic — steipete · 2026-10-28