alt Hacker NewsAsk HN: How do you know if AI agents will choose your tool?
YC recently put out a video about the agent economy - the idea that agents are becoming autonomous economic actors, choosing tools and services without human input.
It got me thinking: how do you actually optimize for agent discovery? With humans you can do SEO, copywriting, word of mouth. But an agent just looks at available tools in context and picks one based on the description, schema, examples.
Has anyone experimented with this? Does better documentation measurably increase how often agents call your tool? Does the wording of your tool description matter across different models (ZLM vs Claude vs Gemini)?
Comments
We've been exposing tools via MCP and the biggest lesson so far: the tool description is basically a meta tag. It's the only thing the model reads before deciding whether to call your tool.
Two things that surprised us: (1) being explicit about what the tool doesn't do matters as much as what it does - vague descriptions get hallucinated calls constantly, and (2) inline examples in the description beat external documentation every time. The agent won't browse to your docs page.
The schema side matters too - clean parameter names, sensible defaults, clear required vs optional. It's basically UX design for machines rather than humans. Different models do have different calling patterns (Claude is more conservative, will ask before guessing; others just fire and hope) so your descriptions need to work for both styles.
The AI agent chooses your tool based on how well your tool’s description matches the user’s intent — clear, specific descriptions win.
Tool description quality matters way more than people expect. In my experience with MCP servers, the biggest win is specificity about when not to use the tool. Agents pick confidently when there's a clear boundary, not a vague capability statement.
tool description wording does matter, at least in my testing. models seem to use the description to reason about whether a tool "should" apply, not just whether it can. two things that helped: (1) explicit input format with an example, (2) a one-sentence note about what the tool does NOT handle. the negative case helps models avoid calling it on edge cases and then failing, which trains them (in context) to prefer it when it's actually the right fit.