Approving and rejecting work

The approval workflow is the core of Z2X's human-in-the-loop model. No agent output becomes "official" until you approve it.

The approval banner

Whenever one or more agent runs are waiting for your review, a yellow banner appears at the top of the screen:

🔔 2 agents are waiting for your approval before acting — Review now →

Click Review now to go directly to the pending items. The banner stays visible on every page until you've cleared the queue — this is intentional. Pending agent work is an open loop that needs your attention.

Reviewing pending approvals

The approval queue shows each pending artifact in order. For each one, you can:

Approve ✅

The artifact is accepted:

• It's added to your evidence ledger (if it contains evidence)
• It's made available to subsequent plan steps as context
• The plan step is marked Done
• If it's an outreach draft: it becomes a draft in your First 100 CRM, ready to send

Request revision ✏️

Send the output back to the agent with notes. The step returns to Running status, credits are charged again, and the agent re-runs with your feedback.

Reject ✗

The artifact is discarded. The step returns to Ready status. No evidence is added. You can:

• Run the agent again (with revised required inputs)
• Edit the step instructions and re-run
• Mark the step as Skipped and move on

Bulk review

If multiple agents ran during an Autopilot session, you may have several items to review at once. The approval queue supports bulk actions:

• Approve all — approve everything in the queue at once (use with care)
• Review individually — go through each one at your own pace

The summary view shows confidence scores and a one-line summary for each item, so you can quickly identify which outputs need careful reading vs. which are clearly fine.

Why this model

Some platforms let agents act autonomously — sending emails, publishing content, making decisions — without human review. Z2X doesn't do this for a specific reason:

Agents make mistakes. Not often, but enough that unsupervised execution would occasionally embarrass you or waste your effort. The approval model means that:

• Outreach emails are always reviewed before they're drafted for sending
• Competitor analyses are checked against your knowledge of the space
• Decision gate recommendations are considered, not blindly followed

The goal is to make approval fast (< 2 minutes per item) so it doesn't become a bottleneck. If reviewing agent output regularly takes you more than 5–10 minutes, the agents aren't producing good enough output — request revisions or update your brief.

Approvals in Autopilot mode

When Autopilot is enabled, agents run automatically and queue their outputs for your review. You don't need to be online for agents to work. When you return, you'll see the banner with everything that ran in your absence.

See Autopilot workflow → for full details.