Four launches in 48 hours just watch coding agents. That's a missing primitive.

Port22 and AgentMicro yesterday. Termexo and TreeWork today. Four launches inside 48 hours whose core job is telling you what your coding agents are doing and which one is stuck.

When four teams independently build the same wrapper in two days, that isn't a market forming. It's a primitive the agent runtimes should have shipped and didn't. Every one of them is solving the same thing. An agent can be working, blocked, done or failed, and there is no standard way to ask it which of those it is right now. So everybody tails a PTY, or scrapes session metadata off disk, or watches the screen, and ships that as a product.

The version I'd pay for isn't a dashboard. It's a status contract. One endpoint per running agent that answers working, blocked on X, done, failed, plus tokens spent so far. Then the menu bar app, the phone app and the tree view are thin clients over the same thing instead of four separate reverse engineering projects.

The uncomfortable part if you're building one of these: the day Anthropic or OpenAI ships that endpoint, three quarters of your product becomes a UI preference. Worth deciding now whether the supervision layer is your product or your moat.