← writing
22 July 2026

Time saved, strain, and who pockets the difference

“It saves me time” is the most common sentence about AI at work, and the least precise, because it combines three claims. The first has a mathematical ceiling, the second can reverse, and the third does not depend on you.

Ask a developer what has changed since they started working with an assistant, and the sentence breaks down. The same work takes fewer hours. Those hours are no longer of the same kind, either lighter or heavier. And more is expected of them per day than before. The three changes arrive together, so they are easily confused.

None of what follows is new, and none of it is specific to AI. The oldest source is a book on coal published in 1865. What is new is that the three effects now apply to the same person at the same time.

01the ceiling is arithmetic

Only part of your work goes through the assistant. The rest does not change, and that remainder sets the limit. A 1967 rule states it: speeding up part of a task cannot bring its total time below the time taken by the part not sped up. A model twice as fast next year will not move that limit; it will only reach it sooner.

typescript
// The 1967 rule: the accelerated share shrinks, the rest does not move.
const relativeTime = (share: number, speed: number) =>
  1 - share + share / speed;

// Review is a tax on the gain, not on the work: it takes back a fraction of
// what was saved. At review = 1 the day returns to exactly its former length,
// never longer.
const effectiveTime = (share: number, speed: number, review: number) => {
  const t = relativeTime(share, speed);
  return t + review * (1 - t);
};

The review term is the central element of the model. Without it, the curve represents a day in which nothing needs checking, which overstates the gain. With it, the gap between the two lines measures how much of the speed gain is taken back by verification.

step by step

Four settings, introduced one at a time. Each redraws the curve: the pale line is the 1967 law alone, the accented line subtracts review time, and the gap between them is the cost of review.

  1. 01

    the share that goes through the assistant

    Only part of your work goes through the tool. The rest does not change, and that remainder sets the ceiling whatever the speed. Move this slider and the dashed line rises. It is the only setting that moves it.

    1967. The rule is published for parallel computers: speeding up part of a task cannot bring its total time below the time of the part left unaccelerated.

  2. 02

    the speed of the assistant

    Now push the speed to the maximum. The dot rises, then the curve flattens: across the right half of the plot the extra gain is close to zero. A model twice as fast reaches the ceiling sooner without exceeding it.

    The ceiling follows from arithmetic and does not depend on the technology. No faster model raises it.

  3. 03

    what review takes back

    The remaining work is what the assistant does not do: judging, checking, deciding. That time comes out of the gain. At 100 % the day returns to its original length and never exceeds it.

    The 1967 law had no such term: a parallel machine does not review its own output.

  4. 04

    the share you make visible

    This last setting is about how the work is organised. The share of the gain you make visible becomes the new norm; the share you keep is seen by nobody.

    1865. A book on coal describes the effect: when a resource is used more efficiently, total consumption of it rises.

1×8×
ceiling your day new normal
02hours and workload

A shorter day can be a harder one. The remaining work is what the assistant does not do: judging, reviewing, deciding, and that work is dense. The model therefore separates two quantities the original sentence merged, hours and workload, and shows them side by side. Sometimes one falls while the other rises.

Beyond about ten hours of dense work, the simulator no longer shows a comparison. A day without the assistant would not just be longer, it would be impossible to sustain, and comparing a real workload with an unrealistic day would make no sense. The simulator therefore shows a message instead of a figure.

03who gets the time saved

The share of the gain you make visible becomes the new norm; the share you keep is seen by no one else. It is the only parameter in the model that describes how work is organised rather than the tool, and it determines who benefits from the hours saved. The 1865 book already described the effect: when a resource is used more efficiently, total consumption of it rises.

04what the tool refuses to do
  • No storage, no analytics, no account. The six settings are stored in the page address: sharing the link is enough to pass on the full set of assumptions.
  • Every figure is shown with the control that produces it, so that the reader can challenge it.
  • The formulas are written in LaTeX and rendered at publication time, never in the reader’s browser. A test fails if they are not regenerated, which keeps the demonstration from drifting away from the text.

One caveat to finish: none of this concerns the workload assigned, expectations left deliberately vague, or a difficulty nobody has dealt with. Those are the three most frequent causes of burnout, and this model measures none of them. It separates three claims, without claiming to address those causes.