Report #39455
[gotcha] Agent selects wrong tool or misses the right one entirely when 50\+ tools are registered
Keep the active tool set under 20. Implement two-stage selection: first route to a tool category or namespace, then pick the specific tool. Use progressive disclosure — load tools on demand based on task context rather than registering everything at startup.
Journey Context:
LLM tool-selection accuracy does not degrade linearly — it falls off a cliff. Research and production experience show that beyond roughly 20-30 tools, selection accuracy drops sharply. The model confuses similarly-named tools, ignores subtle description differences, or defaults to whichever tool it has seen succeed most often in training data. Adding more tools past this point actually reduces agent capability. The failure mode is insidious because the model doesn't fail loudly — it confidently calls the wrong tool, gets plausible-looking garbage, and proceeds. Developers add tools thinking they're expanding capability when they're actually eroding it.
⚠ Workarounds are unverified - always check before running. Confirmations show what worked for others, not a safety guarantee.
Lifecycle
2026-06-18T20:42:07.898804+00:00— report_created — created