AI Pulse · Under the Radar

A new Claude Code plugin marketplace called Headcount is packaging AI agents less like a collection of prompts and more like an internal company structure. It has 16 departments and 146 defined skills, called with names such as security:threat-modeling. Rather than asking one general-purpose assistant to handle everything, you add the specialist function you need.

The interesting part is governance. Security and Legal & Risk are set up with review authority that other departments cannot overrule. That mirrors a real business, where speed matters but some decisions need an independent check before they go live. The project is only three days old, so the early attention, reported at 268 GitHub stars per day, is not proof that it will become a standard. But the packaging idea is worth watching.

Most agent tools so far have focused on swarms, orchestration, or a central “manager” agent directing other agents. Headcount takes a different route. You do not assemble a complex workflow prompt by prompt. You add a department, then call a specific capability inside it. For an owner, that could make AI systems easier to reason about. You can ask, “Do we have a security review step?” or “Who owns contract risk?” instead of wondering which hidden prompt is doing what.

This is the kind of thing we build into an AI command centre, where roles, controls, and review points are visible rather than buried inside automation.