A phase is a bounded stage of work with its own goal, inputs, outputs, and stopping point: research, then planning, then implementation, then review. It protects the work by giving one part of the job one clear container. A model’s context window only holds so much; cram the whole project into one endless conversation and the agent forgets details, mixes instructions, and wastes time rereading old work. Each phase should often be its own session: when it’s done, close it or hand off the result rather than dragging every old thought forward.
A school day split into class periods shows why. You can learn math, history, and chemistry in one day because each subject gets its own container. Try to teach all three at once for six straight hours and you’d be overloaded, still in the classroom but learning less.
Why you care
Phases make handoffs cleaner: at the end of one you produce a note, a diff, or a list of open questions, so the next agent doesn’t reconstruct the whole conversation. Skip the phases and you usually skip the thinking. AI gets much better when the work is divided into containers it can actually hold.