Should AI coding agents be managed like members of an engineerin

Imagine assigning an AI coding agent a bug that needs to be investigated every morning.

The agent can inspect the repository and prepare a report. But who checks whether the report is correct? Who decides whether a fix should be implemented? And where does the team follow the work?

This is why I think AI coding agents need to be managed within a clear project workflow.

A useful process could look like this:

1. A developer creates a task with acceptance criteria.

2. An agent receives the task.

3. The agent executes it in a configured development environment.

4. Progress and blockers remain visible.

5. A developer reviews the result.

6. The task is accepted or sent back for changes.

Scheduling can add another layer: recurring maintenance, code checks, or reports can be triggered automatically.

Sharkly.ai combines a task-management environment with Agents and Automations. It supports scheduled and event-triggered work while keeping people involved in review.

Would your team allow coding agents to execute tasks automatically, or would you require human approval before every significant change?