The 100-Tool Agent Is a Trap - Sohail Shaikh & Ankush Rastogi, Prosodica
Giving AI agents every tool at once degrades speed, cost, and accuracy as the catalog grows.
“It fails because the design does not fail because one tool is badly written.”
Prosodica engineers argue that loading every tool definition into an agent's prompt (the 'fat agent') breaks down beyond ~10-30 tools, causing wrong tool calls, hallucinated tool names, higher cost, and latency. They propose semantic routing with just-in-time context as the fix. This is a practical engineering insight for teams scaling agentic systems into production.