Skip to main content

Reviewing Agent Work

The Review column is where finished agent work waits for a human decision — here's how to read a card, check the full output, and approve or reject with useful feedback.

Written by Schae Lilley

Why Review matters

Nothing an agent produces goes anywhere without a person signing off. When an agent finishes, its work lands in the Review column on the Workflow board and waits for you. Your review is what turns a draft into an approved plan.

Reading a Review card

Each card in Review is built to be scannable:

  • The finding — the card title names what the agent looked into, for which client.

  • The main recommendation — the card leads with the agent's top proposed move, so you can gauge the plan without opening it.

  • The output chip — a small chip in the card footer names the deliverable type (or the file count, if there are several files). If the output lives in Drive, clicking the chip opens it directly.

  • A quality flag, when needed — if the built-in quality check thinks the plan needs revision, the card says so before you open it. Treat that as a heads-up to read extra carefully, not a verdict.

Opening the full picture

  1. Click the card to open its detail view.

  2. Read What this run is asking for — the plan's full set of recommendations, not just the lead move.

  3. Check the signal context: the metric Trend chart and How we got here show the story behind the work. Charts open here in the detail view — they're kept off the cards to keep the board readable.

  4. Skim Verify before shipping — a short list of things worth double-checking against the platforms before you approve.

  5. Open the output itself via the output chip or the deliverable section.

Making the call

  • Approve — you're confident in the plan. You can attach an optional note as you approve, and the card moves to the Approved column with your name on it.

  • Reject — the work isn't right yet. You'll be asked what's worth keeping, what needs to change, and how the agent should re-approach it. Then choose one of two paths:

    1. Dismiss the signal — the underlying issue isn't worth pursuing, so it won't come back.

    2. Send it back with your feedback — the agent tries again using your notes, and a fresh run appears in Running.

Tip: Specific rejection feedback pays off. "CPA target is wrong — use the Q3 target from the client's plan" gets you a much better second draft than "redo this."

Note: If a finished run shows "No output captured," the run genuinely produced no structured deliverable — it's flagged honestly rather than papered over.

Related Articles
Did this answer your question?