How should teams control AI coding agents before they start coding?

AI coding agents are becoming very capable, but I think the bigger challenge for engineering teams is control rather than raw coding ability.

A developer can give an agent a task and let it start coding immediately, but in a real engineering process there are usually several steps first:

Requirement → Planning → Scope Review → Human Approval → Implementation → Pull Request

This is one of the main ideas behind AI Pipeline.

Instead of treating AI as an autonomous developer that immediately starts changing code, AI Pipeline puts the agent inside a controlled workflow where people can review the plan and scope before implementation starts.

I’m curious how other teams are handling this today:

Do you let coding agents start implementation directly, or do you require planning/review first?