Skip to content
← All notes

The productivity trap: AI saves time, but what are we doing with it?

Faster output is not the same as better outcomes. Most teams spent the time AI saved on producing more of what they already produced, which is the least valuable thing they could have done with it.

Leadership7 min read

Ask a team what AI changed and you will usually hear a production number. More documents, more tickets, more analyses, more code. That is real and it is measurable and it is the wrong thing to be pleased about.

The question that matters is what the extra capacity went into, and for most teams the honest answer is more of the same, slightly faster.

Time saved gets absorbed by whatever is easiest to measure

When time is freed up, it does not sit idle waiting to be allocated wisely. It flows toward whatever the organisation already counts, because that is what people are rewarded on.

If the team is measured on delivery volume, the saved time becomes more delivery. If it is measured on documents, it becomes more documents. Nothing consciously decided this. It is just what happens when capacity increases and the incentives stay where they were.

The result is a team producing considerably more output with the same hit rate, which multiplies the work without multiplying the value.

Time freed by automation flowing straight into whatever the organisation already counts. Nobody decided this; it is what happens when capacity rises and the incentives stay where they were.
Saved time does not wait to be allocated wisely.

The specific trap: producing more of the unvalidated thing

The cost of being wrong did not fall. Only the cost of production did. So a team that was building the wrong thing slowly is now building the wrong thing quickly, and will find out at the same point in the process as before, having spent more.

Worse, higher volume makes the underlying problem harder to see. When everything ships faster, there is less time between decisions to notice that the last several did not work.

This is why speed can genuinely make an organisation worse. It shortens the gap in which somebody might have said the quiet thing out loud.

What the time is actually worth spending on

The high-value uses of freed capacity are all things that were previously squeezed out for lack of time, and they share a property: none of them produce anything this week.

  • Validating before building. The cheapest work is the work you did not do, and this is the only activity that reliably prevents it.
  • Talking to the people who use the thing. Still the highest information-per-hour activity available, and still the first casualty of a busy quarter.
  • Reducing the work nobody should be doing. Faster execution of an unnecessary process is not an improvement to the process.
  • Developing people, which is now urgent for the reasons the junior-work problem describes.
  • Deliberate slack. Judgement needs time that is not committed, and a fully booked team has no capacity to notice anything.
The work that gets squeezed out first and is worth the recovered time: validating before building, talking to the people who use the thing, removing unnecessary process, developing people, and keeping genuine slack.
All of it shares one property: it produces nothing this week.

The question to put on the agenda

If your team is measurably faster than a year ago, ask what became possible that was not possible before. Not what got produced. What changed.

If the honest answer is more of what you were already doing, the productivity gain has been converted into volume rather than value, and that conversion happened by default rather than by decision. It can be decided differently, but only if somebody asks.

Common questions

Does AI actually make teams more productive?
It reliably makes production faster. Whether that is productivity depends on whether the extra output was worth producing, and that is a separate question most measurement does not ask. A team that doubles output with the same proportion of wasted work has doubled its waste as well.
What should teams do with time saved by AI?
The things that were always squeezed out: validating ideas before building them, talking to users, removing unnecessary process, developing people, and keeping some genuine slack. They share the property of producing nothing this week, which is exactly why they lose to more output unless somebody protects them.
Why does faster output not lead to better results?
Because the cost of being wrong did not fall along with the cost of production. Building the wrong thing faster reaches the same disappointment sooner and more expensively, and the higher tempo leaves less room to notice the pattern.
How should you measure the impact of AI on a team?
By outcomes rather than volume. What proportion of what you shipped achieved what it was supposed to, how quickly you abandon bad ideas, and whether decisions are getting better. Volume metrics will improve regardless of whether anything got better, which makes them the easiest and least informative thing to report.

PDP Quest exists because of the problem underneath all of these: when output stops indicating capability, you need another way to know who can actually do the work.

See how verification works →