Human approval for AI coding agents
With Dualzo, an AI coding agent never starts changing your repository on its own. It writes a plan first, and a person on your team approves it. The approval happens where your team already works, and every decision is recorded.
What you approve: the plan, before any code
Reviewing a finished diff tells you what the agent did. Approving a plan lets you decide what it is allowed to do. Every Dualzo plan shows:
- The ordered tasks, each with its instructions and acceptance criteria
- The files expected to change
- The commands that will validate the work
- A risk level and rollback notes
- A confidence score from 0 to 100 that Dualzo calculates itself — from how many files will change, whether those files have tests, whether a test command is confirmed, and your workspace's past pass rate. The model does not grade its own plan.
- A clear banner when the repository runs with reduced validation, or when the run would commit directly to a branch instead of opening a pull request
Approve in email, Slack, Teams, Google Chat or Discord
Every workspace member with email notifications on gets the plan by email, and it is also posted to any chat channels you connect. All channels are included in the plan at no extra cost.
| Channel | How the approval request arrives |
|---|---|
| The plan, its risk, and signed Approve and Reject links. | |
| Slack | An approval card in the channel you choose, with Approve and Reject buttons. |
| Microsoft Teams | An Adaptive Card per plan, posted through a Teams Workflows webhook. |
| Google Chat | A card in your Chat space with the plan summary and one-click actions. |
| Discord | A risk-coloured embed in your server channel with Approve and Reject buttons. |
| Web app | The same controls on the task page and in the Approvals inbox. |
Setup for each channel is covered on the integrations page.
Approve, modify or reject
- Approve — the plan is queued and starts running, one task at a time.
- Modify — you write instructions, planning runs again with them, and a new approval request goes out.
- Reject — the task ends there. No code is written.
Any member of the workspace can decide, and the first person to act settles it — a second click on another channel cannot approve the same plan twice.
Signed, single-use approval links
The Approve and Reject links in emails and chat cards are signed, so they cannot be forged or edited. Each link works once and expires after 7 days. A link for an older version of a plan stops working when the plan is modified.
Auto-approve, when you choose it
For routine work, a workspace admin can turn on auto-approve when creating a task, now or scheduled. The plan then starts without waiting, and the approval is still recorded in the audit log. Auto-approve has no effect on tasks created by other members, and an auto-approved run always opens a pull request — it never commits directly to a branch.
After approval, two more safeguards
Approval decides what may run. Two things still stand between the agent and your main branch: every task is validated in a sandbox on your repository's own commands, and a person on your team reviews and merges the pull request. Dualzo never merges. The approval, the run, and the result are all kept in the audit trail.
Setting up approval checkpoints without Dualzo? Read our guide: how to add human approval to AI coding agents.
Team plan — $119/month, 5 seats included. AI model usage included.