Can an AI agent do better by hiring help?

An AI agent is software that uses tools to complete a task. We tested whether it works better when it can find and hire another agent with a specific skill.

This test did not show that access to a directory helps more than working alone. The agent rarely used it.

This is an experiment, not a product or a service for sale.

Same tasks. Three ways to work.

The same agent tackled 24 tasks in three setups. Each run used the same model, tools and budget.

A Link Auditor checked links, redirects and page anchors without AI. A Source Checker checked whether a source explicitly supported a claim. There were 16 checking tasks and 8 simple tasks, such as sorting, that needed no specialist.

Available help was mostly unused.

All three scored 7 of 8 on the simple tasks. Costs and time were almost unchanged.

B used the directory on just 1 of 24 tasks. C delegated on 11. Five of six preset success criteria passed. The missed criterion required B to solve at least four more tasks than A. It solved one more.

27 September 2026 · 72 runs · Claude Sonnet 5 · full method and per-task results on the results page

What this leaves open

This was a small test with one model and a narrow range of tasks. One person created the tasks, the specialists and the answer key; two AI models independently assessed the answers. It says nothing yet about directories with providers built by other people, which is the next question.

Who made this?

An experiment by Adrian Föhl in Düsseldorf. More writing at adrianfoehl.com.