A project manager opens an AI tool at 9:05 on Monday to draft a client update. The first version arrives in ninety seconds. It is decent, so she asks for a friendlier tone, then a shorter version, then a version with a table, then one that mentions last week's delay. At 9:50 she has seven drafts, no sent email, and a faint sense that something is still missing.
That scene is an example of a quiet shift. Starting a task has become almost free, and finishing has not. Without a clear picture of what "done" looks like, each new draft feels like progress while the clock keeps running. The forty-five minutes she lost is time that did not show up on any report.
This is a time issue because the saving at the start only counts if the task actually ends. A task with no stopping point will absorb whatever time is available, and AI makes it easier than ever to keep going.
------------- Context -------------
AI use at work is climbing fast. PwC's 2026 survey of around 50,000 workers found that daily use of generative AI rose from 14 percent to 22 percent in a year. That is a large behaviour change in twelve months, and most of us feel it in our own week.
The same research found that 44 percent of AI users report a heavier workload. That figure sits oddly beside the promise of saved time, and it is worth sitting with. If the tools make individual tasks faster, why does the week feel fuller?
One reason is that saved minutes rarely arrive as empty space. They arrive as a slightly faster task, and a slightly faster task invites a slightly higher bar. Ten minutes saved on a first draft becomes twenty minutes spent refining it, because refining is now easy and nobody told us when to stop.
The better framing is to treat "done" as something we decide before the first prompt, not something we discover at the end. A definition of done is a short statement of what the finished piece must contain, who it is for and how good is good enough. It takes about two minutes to write. It can save forty-five.
This is where the time thread is clearest. A defined finish line turns an open-ended task into a bounded one, and bounded tasks are the only ones whose savings we can actually bank at the end of the week.
------------- Cheap Starts Hide Expensive Middles -------------
When the first draft takes ninety seconds, we stop noticing how much happens afterwards. The middle of a task, the rounds of adjusting, comparing and second-guessing, has no cost label. It feels like work being done rather than time being spent.
Take a consultant who prepares a proposal. The first version appears in three minutes, which feels like a thirty-minute saving against writing from scratch. She then runs six refinement rounds of about eight minutes each, adding up to forty-eight minutes of tinkering. By the end the net saving is a little under ten minutes, and most of it came from the drafting she barely remembers.
The problem is not that refinement is wasteful. Some of it is exactly what makes the proposal good. The problem is that nothing separates useful refinement from drift. Without a definition of done, every round feels equally justified, so every round happens.
That is a direct time loss: roughly 48 minutes of middle absorbing 90 percent of the benefit of the first three. A sentence written beforehand, such as "done means three sections, one price, one next step", would have named the moment to stop. The saving comes from knowing the finish line, not from typing faster.
There is also a mental cost that never shows up on a timesheet. Each round of adjustment asks us to reread, compare and decide again, and those small decisions wear down attention. By the seventh draft the project manager from the opening is no longer judging quality. She is simply trying to remember which version she liked an hour ago.
That fatigue has a time price too. Decisions made late in a drift tend to be slower and less confident, so the last two rounds often cost more than the first two. A stopping rule written at the start protects the clearest part of the morning for work that needs it.
------------- Saved Time Gets Refilled Unless Something Claims It -------------
The 44 percent heavier-workload figure points at a second pattern. Even when a task genuinely finishes faster, the freed time often gets quietly refilled with the next request. Nobody decides this. It happens because open time looks available to everyone, including ourselves.
Consider a team of six that starts using AI for weekly reporting. The report drops from ninety minutes to forty. Nobody records that, and no one says what the fifty reclaimed minutes are for. By Friday each person has absorbed two extra small requests, and the week feels as heavy as before.
A definition of done includes the other half of the sentence: what happens with the time that remains. If the report is finished in forty minutes, the next fifty are either protected for focused work or deliberately assigned. That small decision, made in advance, is what turns a faster task into a lighter week.
The time effect compounds. Fifty minutes reclaimed on one weekly task is almost nine hours across a year per person, and across six people that comes to more than fifty hours. Those hours only exist if somebody claimed them. Unclaimed time does not stay unclaimed for long.
The pattern is easy to test. Pick any AI-assisted task this week and note the clock time at the first draft and at the final send. For most of us the gap between those two moments is the real cost of the task, and it is usually longer than the drafting itself. Seeing that number once tends to change how we start the next one.
------------- Done Is Easier to Write Than to Negotiate Later -------------
A third benefit is less obvious. When "done" is vague, the person who asked for the work and the person doing it often hold different pictures of it. The first version lands, the requester wants something slightly different, and the revision cycle begins. AI makes each cycle cheap, so we run more of them.
Imagine a marketing lead who asks a colleague for a one-page summary of a customer survey. Neither says whether it should include recommendations, charts or quotes. The colleague produces a clean page in ten minutes. The lead wanted recommendations. Another ten minutes. Then charts. Another ten. Thirty minutes of work for something a thirty-second conversation would have settled.
A shared definition of done moves that conversation to the front. It can be as simple as three lines: the audience, the must-have elements and the length. When both people can see it, there is far less to renegotiate, and the work tends to land in one pass rather than three.
This is a clean time gain. If a typical two-person handoff drops from three rounds to one, that is around twenty minutes saved every time it happens. Over a week with five such handoffs, it adds up to nearly two hours, with no new tool required.
------------- Practical Moves -------------
First, before the next AI task, write one sentence beginning "This is done when" and keep it visible. Two minutes of writing sets a finish line that can save forty-five minutes of drifting.
Second, cap the number of refinement rounds before you start, for instance two, and treat a third round as a deliberate decision rather than a habit. A fixed limit makes the middle of the task visible.
Third, for any task you hand to a colleague, share a three-line definition of done covering audience, must-have elements and length. That removes most of the back-and-forth that eats twenty minutes a handoff.
Fourth, decide in advance what happens to the time a faster task frees up. Name it as focus time, a break or a specific next job, so it does not drift into the inbox.
Fifth, once a week, compare estimated time against actual time on two AI tasks. A five-minute check shows where the middle is running long and which definitions need tightening.
------------- Reflection -------------
The headline numbers on AI at work, daily use up from 14 to 22 percent and 44 percent of users reporting a heavier load, tell a story that is easy to miss when we look at single tasks. Individual tasks are getting faster. Weeks are not getting lighter by default. The difference lives in what happens at the edges of each task.
That is why a definition of done matters so much for time. It marks where a task ends, protects the minutes that follow, and removes the renegotiation that eats into the handoff. None of it needs new software, only a sentence written at the right moment.
In the end, the AI shift worth paying attention to is not how quickly work begins. It is how clearly we can say that it is finished. Teams that can say it reclaim hours they would otherwise have spent drifting, and they keep those hours.
Where in your week does a task keep going long after it could have been finished? What would your "this is done when" sentence have been on the last piece of AI-assisted work you shipped? If a task finished fifty minutes early tomorrow, what would you want that time to be for?