What counts as work now: meetings, offices, output versus outcome, and where the time AI saves actually goes.
Everyone reaches for Strava when they describe a tracker for knowledge work. The analogy breaks in exactly the place that matters: Strava measures something you chose to do and want credit for. Your work day is something that happened to you and that you cannot self-report. That is a sleep tracker, not a running app.
Vendors sell surveillance to solve it. The legal definition is narrow, the real cost is usually elsewhere, and monitoring reliably produces the behaviour it was meant to prevent.
Output is what you produced. Outcome is what changed because of it. Almost every productivity metric measures the first and claims to measure the second.
A calendar showing 12 hours of meetings is not costing you 12 hours. It is costing you most of the week, and the difference is why "we only have 12 hours of meetings" sounds reasonable.
Both sides quote productivity research and neither has the data. What the evidence supports, what it does not, and the measurement that would settle it.
The average US commute is about 27 minutes each way. The cost is not the 27 minutes, and the research on what commuting does to wellbeing is unusually consistent.
The most important productivity finding of the last two years is not about AI. It is about how badly we estimate our own time, and what it takes to find out the truth.
Reports of hours recovered from AI are everywhere. Reports of what happened to those hours are nowhere, and that absence is the whole story.
Lines of code broke. Pull requests broke. Story points were always broken. So teams turned to DORA, SPACE and DX Core 4, and hours quietly survived every one of these transitions. Here is why.