The work nobody logged

Employees feed the tools context, check what comes out, correct the errors and clear up afterwards.

In Glean’s recent Work AI Index, 6.4 of the 11 hours the technology supposedly saves each week go into exactly that. The source is a vendor, and the numbers should be read accordingly — but the pattern recurs elsewhere. In BCG’s latest survey of working life, 47 per cent say they spend more time steering and supervising artificial intelligence than doing the work itself.

Harder data than either

Anders Humlum and Emilie Vestergaard have linked surveys to Danish register data — 25,000 workers in eleven occupations regarded as particularly exposed. The study is best known for what it did not find: the effects on earnings are so close to zero that average effects larger than two per cent can be ruled out.

The less-cited finding is the more interesting one for anyone running an organisation. Around 8 per cent of users have taken on entirely new tasks as a result of the tools — and 17 per cent where the employer is actively investing in them. The work did not disappear, then — it was redistributed, and something new was added. And 4 per cent of those who do not use the tools at all have also acquired new work, because their colleagues do.

The work spreads to people who never pressed the button.

A word Norwegian is missing

The vendor side has started calling it botsitting, which is apt and a marketing term at the same time. The Norwegian ettersyn does the job better: the routine inspection of something that runs by itself right up until it does not.

The jobbslaps debate last winter was about the sender — the half-finished deliverables that push work onto the recipient. The person clearing up afterwards was barely mentioned. It is a striking omission, because it is the one clearing up who makes the productivity numbers look the way they do.

Where the work is recorded

That work is invisible says nothing about how difficult it is. It says where it is recorded. Invisible work does not enter the staffing plan, because nobody logged it. It is distributed unevenly, and most often to whoever is most thorough rather than whoever has time. And it does not go away because nobody counts it — it is paid in overtime, in quality, or in someone ceasing to care.

The pattern is familiar from research on organisational change. In our own research on the psychosocial work environment during organisational change we found that the strain rarely lies in the change itself, but in the ambiguity around it: who is to do what, to what standard, and in what time. Role ambiguity is among the best-documented stressors in working life. Introducing artificial intelligence produces it in greater quantities than most reorganisations, because the tasks keep changing without anyone writing it down.

Three questions can be put at an ordinary leadership meeting. Who has taken on new tasks in the past year without anyone writing them down? How much time is set aside for checking AI-generated work, and in whose calendar does it sit? And when the gains were totted up, was the oversight subtracted?

Most organisations can answer the first with a conversation. Fewer can answer the last.