Member-only story
I Found a GitHub Project That Gives AI Agents Actual Jobs
Eight roles, 60 recurring routines, and a useful reminder that agents need boundaries more than impressive demos.
I keep finding AI agent repositories that promise to run entire businesses. The demo looks good. Then I open the source code and discover that most of the work still begins with someone typing a prompt.
That is why Mark Fulton’s ai-employees repository caught my attention. The project starts with a different question: what happens when an agent has a defined job, a calendar, and a record of yesterday’s work?
The README describes eight business roles and 60 scheduled routines. The files are open source under the MIT license. The system runs through an agent environment such as Claude Code or Codex, rather than requiring a new custom application for every role.
I reviewed the repository documentation, installation guide, and guardrails. I have not run the employees in a production business. That difference matters, especially for a project that can operate a signed-in browser.
An agent with a job is different from a chatbot
