Pull last week from your calendar honestly. Two days of meetings. A morning of code review. An hour drafting a doc nobody will read. Lunch with a vendor. A standup, a planning session, a one-on-one. The actual hours where you sat down and wrote code? Maybe ten. If you’re a manager, four.
Every AI productivity pitch is about that ten hours. Cursor speeds up the typing. Copilot writes the boilerplate. Claude writes the test scaffolding. Wonderful. Now the ten hours is six. You got back four hours a week.
Meanwhile thirty hours sat untouched.
What’s actually slow about the week
Three things, in rough order of how much time they eat:
- Decisions that nobody made. Tickets sit in the backlog because no one owns the call. Threads die in the middle. Three engineers are blocked on a question that needs ten minutes of attention from one product person who’s been in interviews all day.
- Context loading. You return to a project after two days and spend forty-five minutes re-reading code you wrote, because nothing wrote down why.
- Reviews of reviews. The PR comes back, you address the comments, the reviewer doesn’t look for two days, you context-switch back, repeat.
These are not coding problems. They are organisational problems with code-shaped symptoms.
What helps
A meeting hygiene policy that defaults to async and demands a decision-maker on every meeting that survives. A standing rule that every PR description has to answer “why now” in one sentence. A bot that pings the reviewer at 24 hours and the manager at 48. None of these have AI in them. All of them save more time than the entire AI tooling stack.
The tools that make the ten hours faster are great. They are also a small lever on a small fraction of the week. The big lever is the org. It always was.