Every KAgents worker starts in copilot mode. The worker prepares the work, and a person on your team approves it before anything is sent, paid or changed. That holds for every worker and every task, including the tasks that will run on autopilot later.
It is not a trial period or a formality. It is how we think new work should start, and it is the same principle KCorp applies in every company it builds. KDoctor, our sister company, is an AI copilot for physicians. It drafts the visit note, and the physician reviews and signs every one. The AI does the work, and a person keeps the final say.
Here is what copilot gives your team at the start, and how to tell when a task is ready to move.
Your rules are never complete on day one
Every team has written rules, and every team has rules nobody wrote down. Which customers get a softer tone. Which supplier always invoices late. Which hiring manager wants a call before an offer goes out. People carry these in their heads, and a new colleague learns them one correction at a time.
A digital worker is no different. On day one it knows what you told it: your policies, your help center, your approval limits, your criteria. Copilot is where the rest shows up. Each edit a reviewer makes points to a rule that wasn't written down yet. Capture it, add it to the worker's rules, and the next draft is closer to what your team would have written.
Review is how your team learns the worker
Trust in a new colleague comes from seeing their work, not from a description of it. The same goes for a digital worker. In copilot, reviewers see how it reads a ticket, which sources it used, what it found and where it stopped to ask.
The audit trail keeps the record behind each draft: what the worker read, which rule applied and who signed off. After enough reviews, the people who own the work know how the worker behaves from their own experience, not from our description.
What a good review looks like
Reviewers can edit a draft, approve it, or send it back with a note. A few habits make that time count:
- Name a reviewer for each task. One person, or one role with a clear owner. Drafts that sit in a shared queue with no owner tend to sit there.
- Read the reasons, not only the output. A correct answer built on the wrong source is a problem waiting for a harder case.
- Send it back with a note when it's wrong. Keep the note short and specific: what was wrong, and what the rule should be.
- Turn repeated notes into rules. If you write the same note twice, it belongs in the worker's rules.
- Watch for rubber-stamping. If approvals become automatic, either the task is ready to move or the review has become too shallow to protect anyone. Both are worth knowing.
Signs a task is ready for autopilot
There is no fixed number of days. A task is ready when its review history says so. Look for these signs:
- Edits are rare, and the ones you make are small.
- The cases reaching reviewers look alike, and the worker handles them the same way each time.
- Your rules cover the exceptions you have actually seen.
- When a case is unclear, the worker stops and hands it to a person instead of guessing.
- The person who owns the task agrees it can run alone.
When those hold, move that one task and leave the rest where they are. Then follow its first days on autopilot through the audit trail, as you would with anything new.
What changes after the switch, and what doesn't
In autopilot, the worker finishes the task end to end in your systems. It stays within the limits you set, such as amounts, recipients, channels and hours. Anything outside them goes to a person with a summary of what happened so far.
Some things don't change. Every action is still logged and attributable. Roles still decide who can assign work, approve actions and change rules. And you can move the task back to copilot anytime, for example when a policy changes or a new product launches.
Some tasks never make the switch. Payment runs and refunds above your limit always need a named approver, and hiring decisions always stay with a person. Copilot is where every task starts, and for a few tasks it is where they stay.
Want to see what copilot would look like on your own work? Bring one job to a demo. We'll map its tasks with you and show which drafts would come to your team for review first.