Tool X-ray
The same request, phrased five ways. Pick one and watch an AI agent assemble its tool call — and every argument your sentence didn't specify, it has to guess. Rewrite the instruction to be precise and watch the guesses vanish. Runs entirely in your browser (0 uploads, works offline).
How to use it
- Pick one of the phrasings, from hand-wavy to precise.
- Watch the agent's
book_meeting()tool call — guessed arguments are flagged in red. - Move to a clearer phrasing and see the guess count drop to zero.
What this clears up (the fundamentals)
- Every vague word is a guess — when your instruction leaves an argument unspecified, the agent invents a value to complete the call.
- Wrong actions are usually ambiguity, not stupidity — the model did exactly what an under-specified request forced it to.
- Precision beats a smarter model — naming the arguments (who, when, how long) fully determines the call, so there's nothing left to invent.
- Tool calls are structured — an agent must fill every parameter; silence in your prompt becomes a default it chose for you.
Where it's used
An AI-agent literacy tool — the fastest way to see why precise instructions matter more than model size for reliable tool use. It's a miatz build-lab concept playable; learn to build it yourself.
FAQ
Why does the agent guess arguments?
Because a tool call requires every parameter to have a value. If your instruction doesn't provide one, the agent must infer or invent it to make the call at all.
How do I stop it guessing?
State the arguments explicitly — the who, the when, the duration. A fully-specified request leaves nothing to infer, so the tool call is determined entirely by you.
Is this how real agents work?
Yes in shape: an LLM maps your instruction to a structured function call and fills any missing arguments from context or assumption. The specifics vary by framework; the ambiguity problem is universal.
Is anything uploaded?
No. The phrasings and tool-call analysis run entirely in your browser — nothing is transmitted or stored. Turn off your Wi-Fi and it still works.
Limits
A teaching model with fixed phrasings and precomputed ambiguity scores, not a live agent. The lesson (precise instructions remove guessing) transfers directly.
Related
Part of the miatz playgrounds. Explore the rest from the playground catalog.
Bookmark this page (Ctrl+D, or ⌘D on Mac) — it works offline the next time you need it.