AI Saves Time. Why Is There More Work?

A few years ago, at a management meeting, someone floated a hypothesis: why don't people see how beneficial AI is, when it will let us spend more time with our families and on what we actually care about. Anna Volchek disagreed at the time — with AI we will work even longer. This episode returns to that answer and tests it.

The mechanism is simple. A unit of output gets cheaper: a letter, a deck, a document or a piece of code can be produced tens of times faster. But we almost never stop at the previous volume — we start producing more. The technology shortens one activity, and the system around it quickly fills the freed time with a new task.

Researchers at the University of California, Berkeley spent about a year observing a technology company: forty interviews, more than two hundred participants. Employees worked faster, yet took on ever larger workloads, ran several processes in parallel and stayed late — without any manager asking them to.

AI creates a layer of work with no natural boundary: you can always do better, go deeper, add more, check again. At the same time the boundaries of the profession itself widen — people move into adjacent specialisations, and the work does not shrink even though each individual task runs faster.

The second line of the conversation is productivity itself. That AI raises output is not in doubt. The question is different: does the whole job get better, and how is that productivity measured? The speed of a first draft and the productivity of a process are not the same thing.

An example from practice: analytical documents from a team of fifteen are consolidated in a few minutes. Then another person spends hours working out what was lost in that consolidation and why the facts do not line up. We count the minutes of generation, but rarely the hours spent framing the task, checking it, getting it approved and running the cycle again.

Researchers from Stanford, New York University and MIT have described a new layer of labour created by generative AI: collective checking, re-checking, experimentation, coordination, implementation. That layer usually stays invisible, and one person's saved time turns into work for others.

AI never lands on empty ground — it layers onto existing processes. Leave the processes unchanged and acceleration simply multiplies the old errors and habits that stopped being useful long ago. Before adding AI, it is worth asking whether the process is needed at all.

Fear gets in the way: fear of falling behind, of not automating, of not showing enough AI. In many companies AI is declared a strategic priority and deployed wherever possible, without asking where the freed time is converted or whether new work is being created for someone else.

A recent McKinsey study of seven hundred and fifty employees and managers shows the gap: more than seventy per cent feel personally ready to work with AI, but only about twenty-seven per cent of managers consider their organisation and processes ready. Thirty-two per cent report real business value where workflows were reviewed and rebuilt — against six per cent where nothing changed.

Three cases show what a deliberate stop looks like. After nine months Starbucks wound down a large computer-vision project for shelf inventory: the system confused similar items and staff had to rescan. Commonwealth Bank of Australia launched a voice bot, announced forty-five contact-centre redundancies, then reversed the decision and apologised — the load turned out to be unmanageable. Over three years Ford brought back and hired more than three hundred and fifty experienced engineers: handing their tasks to automated systems had not produced the result the company needed.

None of these cases is about abandoning AI. All three are about abandoning the belief that using AI is itself productivity and that human expertise is no longer needed. Stopping a scenario usually means not defeat but a tested hypothesis and a way forward.

The episode closes with five questions worth asking before launch: which concrete outcome for the business or the customer should improve; what old work will disappear; did the full time shrink rather than only the time to a first draft; who understands the result, owns it and decides when to stop; and where the freed time will go.

Speed does not entitle anyone to call the work more productive. As long as the old work does not disappear, acceleration only multiplies it — along with the errors it already contained.