Mastering the Orchestrator -> Implementer -> Advisor pattern with AI Agents
Using 'sol-advisor' and 'fable-advisor'
A pattern to use when you work both Claude Code and Codex:
Orchestrator - Plans, reasons and specifies the change
Implementer - Carries out the typing
Advisor - Reviews the change and advises on improvements
On August 1st, I published sol-advisor. It’s my most popular open-source project. It already has 1,300+ GitHub stars in four days! Before sol-advisor, I created fable-advisor.
In this Agentic Engineering Field Note, I will show you how it works, some gotchas to avoid, and why I think this pattern is more powerful than it seems.
After reading, you will feel comfortable implementing this pattern and it will make your agentic engineering implementations more efficient.
How it works
Both plugins ship the same pattern, with the models from each provider that I believe fit the roles best. One session orchestrates and holds the thread. Bounded implementers do the typing. An advisor with a clean context decides whether the work is done.
