Intel / video

The 100-Tool Agent Is a Trap - Sohail Shaikh & Ankush Rastogi, Prosodica

Source: youtube.com
Author: AI Engineer
Date: Published Jun 28, 2026

Context

An need not receive instructions for every tool on every request. Prosodica's tool-routing talk proposes indexing tool descriptions by meaning, searching that index using the user's request, and supplying the model with only the matching tools and their required arguments.

This reduces the amount of tool documentation the model must process and narrows its choice among similar functions. It also creates a new failure point: the search can leave out a necessary tool. The speakers propose widening the selection or searching again when the agent cannot complete the task. The design therefore depends on evaluating retrieval misses as well as the model's final tool choice.

Terms in this piece · Glossary

AI agent — An AI system that doesn't just answer once but works toward a goal in a loop — taking actions, reading the results, and deciding what to do next.
system prompt — The standing instructions a model receives before any user input — defining its role, rules, tools, and tone for the whole conversation.
context window — The maximum amount of text a model can consider at once — its working memory for the current conversation or task.

Key quotes

“the design does not fail because one tool is badly written. It fails because every request is forced to carry the entire catalog.”
Sohail Shaikh

“the idea is pretty simple. Semantic routing is basically a rack for tools.”
Sohail Shaikh

“The catalog can grow, but the model's working set should stay small.”
Ankush Rastogi

“the router does not only add the right tool, it also removes the wrong tool from the model's choice set”
Ankush Rastogi

“it doesn't automatically means your prompt are bad. It may means the architecture is asking the model to solve the wrong problem.”
Ankush Rastogi

Read the source www.youtube.com