AI Pulse · Under the Radar

Garry Tan has published a GitHub project that lays out an AI-assisted software team as a set of 23 tools and roles. The setup includes personas for CEO, designer, engineering manager, release manager, documentation engineer and QA. It is built around Claude Code, with the aim of giving people a repeatable structure for getting work done through specialised AI agents.

The repository is attracting attention, listed at 132,857 stars and gaining around 715 stars per day when this item was reported. You can see the project and its role definitions in the GitHub repository. The interesting part is not that an AI can suddenly run a company alone. It is that someone has turned the vague idea of “an agent team” into a concrete operating template that others can inspect, copy and adapt.

For a business owner, the practical question is where defined roles could remove bottlenecks in your own operation. Think first drafts of product requirements, release checklists, test plans, customer-facing documentation, or handovers between teams. Each task still needs an accountable human, especially where judgement, customer relationships and commercial risk are involved. But a clear set of agent roles can make routine work more consistent and reduce the blank-page problem.

This is the kind of thing we build into an AI command centre, where agents support real workflows rather than sitting as disconnected chat tools.