I open-sourced a set of 7 agent instruction files that define a multi-agent workflow for AI-assisted development.
Each agent has a narrow role with hard constraints: the Coordinator delegates but cannot edit files. Implementors execute single tasks and report back. The Verifier only approves with concrete evidence per acceptance criterion. The PR Shepherd polls CI and review comments in a loop until the PR is merge-ready.
Core idea: separation of concerns. The agent that writes code is never the agent that checks it. Work is split into parallel waves (2-4 non-overlapping tasks), verified per wave, then verified again at the end.
The agents are plain markdown files — no framework, no dependencies. Install with npx agent-system init or just copy the files you want.
● When everyone has access to the same models, and those models can produce working software quickly and cheaply, the code itself stops being a moat. A reasonably skilled person can now build what used to take a team of engineers months to ship.
● A competitor can reach feature parity in days. The thing that used to protect a business, the sheer effort required to build it, is mostly gone.
● Think of it like oil. If every property owner had a well that was cheap to operate, the price of oil would fall toward the price of water. The resource is abundant, the extraction is easy, and the margin disappears. Software features are heading in the same direction.
These things also on my mind recently. I'm a freelancer, providing design-development services, and trying to understand as things changing rapidly.
I don't think even the tools are advancing, allowing us to build in a new way, not everyone will be able to develop production ready products. Or good looking ones. So there will be always in need for people know how to orchestrate the new workflows with good taste and years of experience.
Each agent has a narrow role with hard constraints: the Coordinator delegates but cannot edit files. Implementors execute single tasks and report back. The Verifier only approves with concrete evidence per acceptance criterion. The PR Shepherd polls CI and review comments in a loop until the PR is merge-ready.
Core idea: separation of concerns. The agent that writes code is never the agent that checks it. Work is split into parallel waves (2-4 non-overlapping tasks), verified per wave, then verified again at the end.
The agents are plain markdown files — no framework, no dependencies. Install with npx agent-system init or just copy the files you want.
GitHub: https://github.com/boraoztunc/agent-system
reply