AI Tools Added to the Workload. The Teams That Noticed First.

The uncomfortable part of the last two years is not that AI tools failed to help. Many people genuinely save hours with them. It is that a number of teams adopted several, measured carefully, and found the week had not got shorter.

"We bought AI and got more things to check."

What the teams noticed

The pattern was consistent enough to be recognisable. Each individual tool did what it promised: drafts written faster, summaries produced, code suggested. Individual satisfaction was high, and people would have objected to losing them.

What did not change was throughput. Work still took the same elapsed time to get through, and in some cases longer, because there was now more of it to review. The teams that spotted this early were the ones already measuring elapsed time rather than counting output.

The numbers, and what they do not say

Three findings from the research are worth holding together, because each is misleading on its own.

Regular users do report saving around eight hours a week, which is real and should be conceded rather than argued with. Productivity falls once workers are juggling four or more AI tools, so the benefit is not additive and at some point reverses. And two thirds of workers get no guidance at all on what to do with the time saved, which means the saving has no designated destination.

Put together they describe a specific failure: time is genuinely freed, nobody is told what it is for, and beyond a few tools the coordination cost of managing them exceeds the gain.

Checking is the new work

The most consistent complaint is that for every hour AI saves, people spend another checking what it produced. That is not a complaint about quality; it is a description of a new task category that did not exist before.

And checking has bad properties. It is hard to delegate, because verifying output requires the expertise that would have produced it. It is unsatisfying, so people do it less carefully over time. And it lands on the most senior people, who were already the most interrupted.

Before	After
Write the draft: 90 minutes	Generate the draft: 5 minutes
Review: 20 minutes	Verify every claim in it: 60 minutes
Wait for the decision: 2 days	Wait for the decision: 2 days

The bottom row is the point. The production time collapsed, the verification time grew, and the waiting, which was the largest number by an order of magnitude, did not move at all.

What the fix should be

If the waiting is the largest cost, the fix has to reduce waiting rather than increase output. That is a different kind of product, and it points the opposite way from the market's direction of travel.

StandIn takes that direction. It generates nothing new and it does not act on your behalf. It puts the work you have already done to work: you spend ninety seconds at the end of the day confirming a brief, mostly pre-drafted from what you actually did, and while you are off, your StandIn answers your colleagues' questions from it in your words, with a source under every answer, clearly labelled. It never guesses, so there is nothing to verify in the way a generated draft has to be verified: every answer points at the work it came from.

So the waiting falls, the interruptions fall, and no new output arrives needing to be checked. That is the whole of the contrarian position: the work you already did, finally working for you. See StandIn for leadership teams.

An audit worth running

Before buying anything else, spend a week on three numbers. How many AI tools does the average person on the team use in a week. How much time do they spend verifying output. And what is the ratio of elapsed time to active time on five recently completed items, as described in why projects slip on distributed teams.

If the tool count is above four, consolidation will help more than addition. If verification time is large, you have bought output rather than capacity. And if the elapsed-to-active ratio is high, your constraint was never production, so no generator will move it.

Common Questions

Do AI tools actually save time at work?

Regular users do report meaningful savings, and the effect is not additive: productivity falls once people are juggling four or more tools, and much of the saving is consumed by checking the output. The saving is real and the net gain is smaller than the headline.

Why did our AI rollout not make us faster?

Most likely because production was not your constraint. If work spends most of its life waiting for an answer, a decision, or a review, faster drafting changes a small share of the elapsed time.

How many AI tools should a team use?

Fewer than four, on the evidence. Beyond that the overhead of choosing between them, learning them, and reconciling their output begins to outweigh what each one contributes.

What should we do with the time AI saves?

Decide in advance and say it out loud, because two thirds of workers are given no guidance and the time quietly refills. An unallocated saving is not a saving.