Building a progress tracker for AI coding agents

Running one AI coding agent is easy to follow. Running four at once is not. Every terminal says roughly the same thing, working, and the only way to find out which one has been stuck for twenty minutes is to go and look.

The tools that exist today tell you whether an agent is busy or idle. That's useful, but it isn't the question you're asking. You want to know how far along each one is, what it's doing right now, and whether it needs you.

What we're building

Agent Tracker is a small macOS app that lives in the menu bar. It works on three levels:

• A meter in the menu bar: how many sessions are running, their combined progress, and a colour for the most urgent state.
• A list of sessions, with stuck and waiting ones sorted to the top.
• A detail view for each session: its goal, its plan, which step it's on, and what it's doing now.

When an agent finishes, gets stuck or is waiting on you, you get a notification. You stop babysitting terminals.

Where the data comes from

Agents already write down what they're doing. Claude Code and Codex CLI keep session files and offer hooks on your Mac, so the tracker reads those locally. Your code stays where it is. For tools without that, a small local MCP server lets any MCP-capable agent report its plan and its steps.

What we're leaving out

The first version is a visual tracker and nothing else. We have ideas for spoken summaries and for replying to a stuck agent by voice, but they wait until the core is solid. That's the without mercy part.

It starts with Claude Code and Codex CLI. Everything else can wait its turn.