New Dashboard Lets One Person Supervise a Whole Team of AI Coders

AI coding assistants are getting more independent. Instead of asking for one small piece of code at a time, programmers now set several AI helpers loose at once, each working on its own task. That sounds efficient — but it creates a new problem. Someone still has to keep track of who is doing what, notice when an AI goes off track, and decide when to jump in. Think of it like a chef suddenly supervising five cooks in five kitchens instead of one.

To understand that problem, researchers watched 14 developers work and identified five habits that good supervisors share: planning the work, keeping tasks separate, logging what happened, watching progress, and sorting out messes. They built a tool called ParallelPilot around those habits — a planning screen, an activity log, and a live dashboard that sits alongside the coding tools people already use.

Then they tested it with 16 programmers, each trying both their normal setup and ParallelPilot. The results were strong: people completed 63% more tasks on short jobs, kept about one more AI helper running at peak, and reported less effort spent tracking things and less switching between windows. Fourteen of the 16 said they preferred it. The tool also made plans, task dependencies (which job depends on another), and moments needing human help clearer.

There is an honest catch. Users did not report feeling significantly more in control, or better at redirecting the AI when it went wrong. Finishing more tasks is not the same as feeling confident about them. The researchers argue that future AI coding tools should give people a quick, cheap way to check the actual work — the code and the reasoning behind it — not just a tidy summary. That advice likely applies well beyond programming, to anyone supervising AI that acts on their behalf.