Developer Cuts Agent Tools from 200 to 5, Reducing Errors and Token Use

On 2026-11-10, the AI section of DEV Community published a retrospective from a developer under the headline 'How I cut wrong answers and token usage by giving the model 5 tools instead of 200.' The work began with a ticket that was essentially one sentence: make the AI agent give fewer wrong answers. There was no spec and no acceptance criteria, just an agent that kept picking the wrong action. The developer's first assumption was that it was a prompt problem, and排查 ruled that out. Before changing any prompt, the developer logged what the agent was actually sending to the model on each request, and found every single call carried roughly 200 tool definitions. Too many tools made it hard for the model to choose among them, which is why the agent kept making the wrong pick. The fix was to reduce the tool set from around 200 to 5, and the result was a decline in both wrong answers and token usage. That is as far as the available reporting goes. One source is on record, and it provides only the comparison after the compression; no tool list, no quantified error or token figures, and no details on whether the change was kept, reverted, or applied elsewhere.