The schedule used to do your prioritizing for you

Gergely Orosz reports that Asana needed to get off the Enzyme testing framework, which meant a massive rewrite of test cases, and that with coding agents the project finished in two weeks. Then he adds the line that actually matters. Without agents, that work would surely have been kicked down the road. He says Airbnb and Uber tell similar stories.

The same week, Simon Willison was talking about conceptual integrity, the idea from The Mythical Man-Month that well designed software has no surprises in it and everything fits together. His worry is that agents make it very easy to keep adding rooms. His host reached for the Winchester Mystery House and he ran with it. Forty years of construction, because the price of one more room never forced anybody to decide.

His description of the old discipline is the sharpest thing I read this week. You would get an idea for a crazy feature, think about the week it would cost, and let it go. That estimate was doing your prioritizing for you, quietly, for free, for your whole career.

Which makes the Enzyme migration and the Winchester house the same event. Work that used to be filtered out by its price is now cheap enough to do. One of those was worth doing and one was not, and nothing about the cost falling tells you which is which.

The vault’s coding notes insist that an agent task carry an outcome, constraints, a definition of done, and an expected receipt. I used to read that as paperwork for robots. Against this week it reads more like a replacement part. Somebody now has to say out loud that a piece of work should exist, because the calendar has stopped saying it for them.