The agent

What it is for

Measurement tells you that you are missing from the best tools for X answers. The agent does something about it: it researches the sources those answers use, writes the article or the outreach email, and puts the result where your team works.

You give it a task in plain language. It plans the task as a list of steps, works through them, and shows you what it did at each one.

The To-do table

A session keeps a table of proposed actions. Nothing in it happens on its own — each row is a thing the agent believes should be done, and you decide which ones run.

What it can reach — and what it cannot

The agent reads the same data you can ask for over MCP: your metrics, your sources, your audit, your crawled pages.

It runs read-only, and that is a decision rather than a gap. Tools that change something return a plan and wait for a signed approval. If the agent could call them, it would receive that approval token in its own tool result and hand it straight back — approving its own change. A safeguard the caller can walk through on its own is not a safeguard.

So the agent writes drafts, and moving anything out of the session is your move: export it as PDF or Word, and what happens to it next is your call.

Cost

Agent work consumes credits, measurement does not. That separation is deliberate: a session you did not run is a session you run tomorrow, but a day of measurement you did not take is a hole in the series forever. Running out of credits never stops tracking.