AI Pulse · Under the Radar

A new Claude Code skill pack called headcount is taking a different approach to AI assistance. Instead of handing one model a long prompt and hoping it remembers every rule, it structures the work like a virtual company. It includes 16 departments, covering areas such as executive work, security, legal, marketing and finance, with 143 named skills assigned across them.

The important detail is governance. Security and Legal are set up as reviewer-class departments, with authority to block output from other departments. That mirrors a real business process. Marketing might draft a campaign, but legal still gets to say no. The project’s pitch is simple, “add a department, not a prompt.”

For an owner, the useful idea isn’t that AI can pretend to be your whole company. It’s that repeatable AI work needs roles, boundaries and checks. A finance task should not operate by the same rules as a marketing task. A customer-facing draft may need review before it leaves the business. Putting those controls into the structure, rather than rewriting instructions every time, could make AI use more consistent across a team.

The GitHub project picked up 629 stars and 91 forks in its first two days, with zero open issues reported. That shows early interest, not proof that it works reliably in a real operating environment. Still, this is the kind of thing we build into an AI command centre, where different business workflows have clear responsibilities and approval points.