Developers became agent managers—without management discipline

With an agent, a programmer increasingly does not produce code but assigns work to an executor, supervises it, and accepts the result. This is delegation, not merely the use of a tool.

If AI is treated as a tool, constant hints, micromanagement, and manual finishing seem natural. If it is treated as an executor, the same actions point to weak delegation: the task was poorly framed, the executor lacks autonomy, and the pipeline demands too much attention.

Real delegation ends with reviewing and accepting the result. A manager who finishes the executor’s work every time has not built a functioning system. Yet developers rarely count rejections, interventions, and failed one-shots; they do not measure their own attention or develop the skills of briefing, supervision, and acceptance.

Management cannot assess these competencies either: Jira does not show who built an autonomous process and who spent the day micromanaging an agent. New management work has appeared, but organizationally it is treated as if it does not exist.

Note

The delegation metaphor is useful here, but its limits matter. An agent is a probabilistic system, not an employee with stable memory and accountability. A large number of hints, manual finishing, or a failed one-shot therefore does not automatically mean poor management.

For a one-off task, a ten-minute manual fix may be more rational than spending two days on a universal skill, a regression dataset, and a separate harness change. The production system should be developed where the task class recurs, the cost of error is high, or the improvement can be reused.

Management practices help frame tasks and structure acceptance more precisely, but they do not provide a ready-made protocol for communicating with a model. Strong engineering practice is not maximum autonomy at any cost; it is a level of systematization that pays for itself.

New work · old evaluation system

Delegation cycleBriefing → execution → supervision → acceptance

Constant manual involvementMicromanagementHints, rejections, and finishing remain hidden inside a local session.

Verifiable briefing and acceptanceAutonomous processThe result is checked, and reasons for rejection improve the next run.

briefingrejectionsinterventionsacceptance

New management work has appeared, but organizationally it is treated as if it does not exist.