#llmtoolcalling

I build first AI agents as narrow, controlled tool-use loops: the model proposes actions while my code enforces permissions, state, timeouts, logs, and