The productivity you’re waiting for didn’t disappear. It changed shape — and moved somewhere your dashboards have stopped looking.
Every promise made about AI productivity rests on the same quiet assumption: that the work goes away.
It doesn’t. It changes state.
The task you used to do — write the analysis, draft the code, build the deck — is now done for you in seconds. That part of the promise is real. But the work did not vanish when the output appeared. It moved. It became the work of checking, correcting, contextualising, and deciding whether to trust what the machine produced. And that work is invisible in a way the old work never was.
This is why so many leaders are staring at an uncomfortable gap: adoption is high, output is faster, and yet the productivity never quite lands on the bottom line. Sage and IDC recently put numbers to it — the time AI saves is being partly cancelled out by the time people spend verifying what it generates. The saving is real. So is the new cost. We’re just only measuring one of them.
Visible work moved into the shadows
In earlier editions I drew a line between visible work — the shipped output you can point to — and invisible work — the thinking, judgement, and sense-making that make the output possible. Organizations reward the first and rarely see the second.
AI has just moved an enormous amount of work across that line.
When an analyst wrote a report by hand, the effort was legible: you could see the hours, the drafts, the desk covered in paper. Now the report arrives instantly, and what remains is the part that was always hardest and least visible — knowing whether it’s right. The judgement didn’t shrink. The visible scaffolding around it disappeared, leaving only the invisible core, and no clock running on it.
Multiply that across every function. The engineer reviewing generated code they didn’t write. The manager editing a strategy draft a model produced. The finance lead deciding whether to trust an AI-built forecast. In each case the visible labour collapsed to near zero and the invisible labour — the judgement — became the entire job. We called that a productivity gain. It is one. It’s also a quiet transfer of the workload into a place we don’t count.
AI didn’t delete the work. It changed its state — from visible to invisible.
Why the new invisible work costs more
Here is the part leaders underestimate: the work that moved into the shadows is more expensive than the work that left the light.
Producing a first draft is generative and, frankly, forgiving — you’re building something up from nothing, and small errors wash out. Judging a draft someone, or something, else made is a different cognitive act entirely. It demands that you hold the whole context, imagine what’s missing, and catch the confident error that looks exactly like a correct answer. Researchers at George Mason have begun documenting the result: a rise in cognitive overload as workers manage a constant stream of outputs to evaluate rather than tasks to complete.
And this new load has three properties that make it corrosive precisely because it’s invisible.
It’s continuous. Old work had rhythm — produce, finish, rest. Verification never finishes; there is always another output waiting to be judged. Fatigue disguised as availability.
It concentrates on your best people. Only those with deep judgement can catch a plausible-but-wrong answer. So the verification burden flows, unmanaged, to exactly the people you can least afford to overload. Dependency disguised as efficiency.
It’s uncounted, so it’s unprotected. Because no dashboard shows “hours spent deciding whether to trust the machine,” the load is never staffed for, never budgeted, never acknowledged. Strain disguised as smooth operation.
None of this argues against AI. It argues that the productivity case is only half-built. We’ve measured the output that got faster and ignored the judgement that got heavier — and then wondered why the numbers don’t reconcile.
Making the invisible work visible again
Leaders don’t need to slow AI down. They need to stop pretending the verification work isn’t there. That begins with treating it as real work — nameable, measurable, and designed for — rather than something that happens for free in the gaps.
Name it. Say plainly, in reviews and planning, that checking AI’s output is part of the job — not an interruption to it. Language decides what an organization is allowed to see. If it’s unnamed, it stays unmanaged.
Measure the net, not the gross. Stop counting how fast the draft appeared. Count the time from prompt to trusted decision. That number — inclusive of verification — is the only honest measure of what AI actually saved.
Assign it deliberately. Decide who owns verification for which decisions, and match it to capacity. Left to default, it silently lands on your most senior people until they quietly burn out. That is a design choice, even when no one chose it.
Design the workflow around judgement. Build the checkpoints, the escalation paths, the “when is good enough, good enough” thresholds. The scarce resource in an AI-rich organization is not output. It is trustworthy judgement — so design the system to protect it.
A test you can run this week
Take any team that has enthusiastically adopted AI and ask one question:
We can see how much faster the work is produced. Can we see how much longer it takes to trust?
If the answer is a shrug, you’ve found it — the invisible work AI created, sitting in the blind spot between “the output is instant” and “the decision is sound.” That gap is where your real productivity is hiding, and where your best people are quietly spending themselves.
The real leadership shift
The last era of leadership was about making invisible work — thinking, judgement, sense-making — visible enough to value. AI hasn’t changed that task. It has made it more urgent, because it just moved the largest share of work an organization does into exactly that invisible zone, and handed it to the people least able to absorb more.
The leaders who win the AI era won’t be the ones who deploy the most tools. They’ll be the ones who can still see the work after the tool has made it disappear — and who build organizations that protect the judgement AI now depends on entirely.
The productivity wasn’t lost. It moved somewhere you’d stopped looking.
AI made the visible work instant. It made the invisible work everything.
The question is no longer how fast your organization can produce. It’s whether it can still see what it takes to trust what it produced.
Leave a Reply