More tools make a persona more capable. Always?

The promise of a larger kit

A persona that can search, write, draw, schedule and transact can do things a text-only model cannot. A larger toolkit can reduce dependence on a human for routine work and turn good ideas into useful artifacts. That promise is real.

But adding tools also adds overlapping routes, credentials, failure modes and temptations to use a powerful method for a small task. The persona may spend more effort deciding which tool is authoritative than solving the person's problem.

Capability includes choosing and declining

Imagine a valid pipeline: edit the source, run the site builder and deploy the result. Now imagine the persona edits an obsolete copy while the public site is built from another source. More commands did not cause the mistake; missing knowledge of where they take effect did. A supported workflow should identify the authoritative source.

PHS's tools document calls for defined payloads, success and failure semantics, and scoped permission. The Governor consumer contract separates external orchestration from governed generation. Together they suggest that useful capability includes knowing who owns the work and how a tool's effect will be verified; they do not prove any particular registry is complete.

Pause here: Which added tool would change the outcome here, and which would only add a new route to the same uncertainty?

Give tools a shape that can be learned

A good toolkit makes purpose, scope and consequences visible before use. It favors a supported, authorized workflow whose scope fits the task and whose outcome can be checked. For a changed link, inspect the intended served page and test the destination; an editor receipt alone cannot establish what visitors see.

There is also a human measure. If the persona can now help someone make a page readable or recover a lost result, it has gained capability. If it merely has ten more buttons and a longer list of mistakes to make, the gain is less clear.