The paradox always starts with an oversimplified promise
AI was supposed to change everything.
It was supposed to accelerate teams, lighten workloads, improve decisions, increase productivity, transform jobs, free up time, create new margins, help organizations breathe again and, incidentally, make every employee feel like they were piloting a spaceship from inside a spreadsheet.
Then the numbers arrive.
The NBER study Firm Data on AI surveyed nearly 6,000 business leaders in the United States, the United Kingdom, Germany and Australia. It reports that about 69% of companies actively use AI. It also states that more than two thirds of executives regularly use AI, but only 1.5 hours per week on average. Finally, 89% of surveyed firms report no impact on productivity over the past three years. (NBER)
In other words: the rocket engine has been installed, but many companies are still pushing the aircraft by hand.
This result does not condemn AI.
It condemns organizational laziness.
Too many companies still confuse adoption with transformation. Installing ChatGPT, Copilot, Gemini or Claude inside the company does not transform anything by itself. It is like buying a grand piano and announcing that the executive committee now masters Rachmaninoff.
There is the object.
There is the use.
There is the practice.
There is the discipline.
And above all, there is what the organization decides to do with the time it frees up.
AI does not save time, it reveals your relationship with time
“AI saves time” has become one of the most comfortable slogans of the moment.
It reassures.
It sells.
It justifies budgets.
It gives leaders the impression that they are moving with the times.
But it carefully avoids the irritating subject: what do we do with the time saved?
If the time saved is used to produce more mediocre slides, answer useless emails faster, summarize meetings that should never have existed, generate three versions of a report nobody reads or accelerate an absurd procedure, then AI is not modernizing the company.
It is industrializing its flaws.
It is putting a turbocharger on the hamster wheel.
That is where the paradox becomes interesting.
The problem is not that AI does not work. The problem is that many companies insert it into organizations that have never redefined the value of work.
They ask AI to do things faster.
Rarely to do them better.
Even more rarely to do them differently.
Almost never to remove what no longer makes sense.
Solow’s ghost returns with a graphics card
This debate is not new. In the 1980s, economist Robert Solow famously summarized the computer productivity paradox by saying that the computer age could be seen everywhere except in the productivity statistics. The topic then fed decades of analysis on the gap between technological investment and visible economic gains. (Brookings)
Today, AI is reviving that old ghost.
We see AI everywhere.
In pitch decks.
In newsletters.
In conferences.
In strategy committees.
In consulting offers.
In roadmaps.
In internal announcements.
In sentences that begin with “we have launched a task force.”
But in operational results, the signal often remains blurred.
The NBER is not alone in observing this gap. A Federal Reserve Bank of Atlanta summary of the same study notes that more than 80% of surveyed firms report no impact from AI on either employment or productivity, while expecting larger effects over the next three years. (Atlanta Fed)
That is the core issue: AI is already present, but not yet structurally embedded everywhere.
It is used, but rarely integrated.
It is tested, but not always connected to strategy.
It is authorized, but not necessarily governed.
It is applauded, but still not practiced enough.
A powerful technology inside an old organization remains trapped
The OECD notes that generative AI can improve productivity by automating tasks and augmenting workers, but also stresses the need to adapt organizations, processes and strategies to fully leverage its potential. (OECD)
That sentence should be printed on the door of every executive committee room.
Adapt the organization.
Adapt the processes.
Adapt the strategies.
Not just distribute licenses.
Not just publish a policy.
Not just run a two-hour training session where everyone learns to write “act as a world-class supply chain expert.”
Productivity does not appear because a tool exists. It appears when the tool changes practices, roles, decision-making, quality standards, information flows and the way people work together.
A company that uses AI to accelerate badly designed procedures mainly obtains faster nonsense.
A company that uses AI to rethink how work is done can move into a different category.
The trap of reassuring small usage
The NBER study shows that executives who use AI do so for about 1.5 hours per week on average. (NBER)
One and a half hours.
That is less than a long episode of a series with opening credits, recap and end-of-evening guilt.
And yet these same executives must make decisions about how AI will transform their company.
There is a worrying mismatch here.
They want to lead a disruption they barely practice.
They want to evaluate a technology they touch only lightly.
They want to measure impact before creating the conditions for meaningful use.
Of course, not every executive needs to become a prompt engineer, model architect or agent developer. But without sufficient personal practice, their understanding remains abstract.
And abstraction is dangerous.
It produces two caricatures.
On one side, the worshippers of magical AI.
On the other, the comfortably immobile skeptics.
Between the two, there are serious leaders: those who test, compare, question, identify use cases, examine friction, measure quality, speak with teams, accept learning and make small mistakes quickly.
The two questions that change everything
In chapter 14 of my book, dedicated to applying artificial intelligence, I propose two simple questions to get out of the fog:
How can I increase, thanks to AI, my productivity and the quality of what I deliver within my current role and objectives?
What can I now do thanks to AI that was impossible before?
Any use of AI that does not answer one of these two questions is a distraction.
These two questions are powerful because they prevent the company from reducing AI to a gadget.
The first question improves the existing work.
The second opens new territory.
The first asks: how can I do better what I already do?
The second asks: what new capability becomes accessible?
The mistake is to remain stuck in the first one.
Yes, AI can summarize, write, correct, synthesize, classify, prepare, rephrase, translate, search and generate.
Fine.
But the organizational shift begins when AI allows teams to test hypotheses faster, personalize customer relationships at scale, simulate multiple strategic scenarios, create prototypes without waiting for three framing meetings, compare complex options, surface dispersed knowledge or give a team a level of preparation it could not previously reach alone.
Saved time must not feed the absurd task machine
Time saved by AI can follow three paths.
First path: do more of the same.
This is the most common one.
More emails. More documents. More variants. More internal noise. The company thinks it is gaining productivity while mainly increasing the volume of internal circulation.
Second path: do better.
Improve the quality of an analysis. Prepare a decision more seriously. Review a contract with more attention. Anticipate objections. Clarify a message. Improve the precision of a deliverable. Raise the standard.
Third path: do something else.
Explore scenarios that were previously too expensive. Give small teams capabilities that were once reserved for large organizations. Invent new services. Transform customer relationships. Create faster learning loops. Reconfigure the job itself.
The problem is that many companies do not choose.
They let the time saved be swallowed by the existing system.
And the existing system loves to feed.
It always creates one more meeting.
One more report.
One more validation.
One more procedure.
One more final_final_V12 version.
The productivity paradox is managerial
We talk a lot about models, GPUs, tokens, copilots, agents, automation, RAG, fine-tuning, security, sovereignty and hallucinations.
Fine.
But in many organizations, the central knot lies elsewhere.
It is managerial.
Who decides the priority uses?
Who measures the quality of work produced with AI?
Who arbitrates between time savings and quality gains?
Who authorizes the removal of a task that has become useless?
Who protects employees who experiment?
Who prevents AI from becoming a surveillance tool rather than an augmentation tool?
Who defines what still deserves human attention?
Companies want new results with old reflexes.
They want the future with validation circuits inherited from a 2007 spreadsheet.
They want autonomous teams, then require three signatures to test a twenty-dollar subscription.
They want innovation, but keep processes that punish experimentation.
In my book, chapter 14 emphasizes the creation of a multidisciplinary exploration team made up of early adopters and ambassadors, able to test tools, validate relevant uses and identify organizational friction.
That is exactly what many companies need: not a grand AI ceremony, but a team able to learn quickly and spread useful practices intelligently.
Useful AI is not the one that replaces humans
The obsession with replacement is an intellectual dead end.
It attracts clicks.
It feeds fears.
It entertains forecasting firms.
But in real companies, the most useful question is often augmentation.
How do we free humans from tasks that prevent them from thinking?
How do we reduce mental load?
How do we improve decision preparation?
How do we make information more accessible?
How do we give teams more analytical power?
How do we increase the time available to listen, create, negotiate, arbitrate, learn and transmit?
The AI that matters is not only the one that writes for you.
It is the one that helps you think better before writing.
It is not only the one that summarizes a meeting.
It is the one that forces you to ask why that meeting exists.
It is not only the one that generates a report.
It is the one that reveals that nobody makes decisions from that report.
Oops.
Adoption without work redesign produces little
The OECD observes that AI adoption in companies is progressing, but remains highly variable depending on countries, sectors and company size. It also highlights that larger firms adopt AI more easily than smaller ones. (OECD)
This point matters.
Large companies have more resources, but also more inertia.
Small companies have fewer resources, but sometimes more speed.
In both cases, the real lever is the ability to turn uses into useful routines.
One training session is not enough.
One hackathon is not enough.
One enthusiastic message from the CEO is not enough.
An AI strategy must translate into daily work:
- how a meeting is prepared;
- how an offer is written;
- how a market is analyzed;
- how a customer is answered;
- how a new employee is trained;
- how a project is documented;
- how a decision is made.
As long as AI remains “one more tool,” it creates fatigue.
When it becomes a new way of working, it starts creating value.
The wrong metric: number of licenses
Many companies measure AI maturity with reassuring indicators.
Number of deployed licenses.
Number of completed training sessions.
Number of shared prompts.
Number of workshop participants.
Number of identified use cases.
Number of pilots.
All of that can be useful.
But these are activity indicators, not necessarily transformation indicators.
More serious indicators are elsewhere:
- how many decisions are better prepared?
- how many tasks have been removed?
- how many deadlines have truly been reduced?
- how many deliverables have improved in quality?
- how many errors have been detected earlier?
- how many customers received a more relevant answer?
- how many employees say they have regained focus?
- how many managers have changed their decisions thanks to AI?
The trap of technological deployment is that it gives the impression of action.
But the action that matters is the one that changes work.
AI as a mirror of the company
AI acts as a revealer.
It reveals organizations that know how to learn.
It reveals managers who know how to trust.
It reveals teams that know how to experiment.
It reveals processes that deserve to be deleted.
It reveals curious employees.
It reveals leaders who truly practice.
It also reveals companies that merely repaint old habits with futuristic vocabulary.
When a company says “we use AI,” we should ask:
- where?
- for what?
- with what result?
- with what measurement?
- with what change in method?
- with what evolution of the human role?
- with what removal of useless tasks?
Without these questions, AI becomes decoration.
Expensive decoration.
Sometimes very pretty.
Still decoration.
Saved time must become a strategic choice
The most important topic is not speed.
It is the allocation of freed time.
Time saved thanks to AI can be returned to employees to reduce mental load.
It can be invested in quality.
It can be used to train teams.
It can help explore new markets.
It can strengthen customer relationships.
It can improve strategic preparation.
It can help people make better decisions.
It can also be immediately taken back by the organization to produce more volume, more control and more noise.
That is why AI is not only a technology issue.
It is a management issue.
A culture issue.
A courage issue.
A choice issue.
AI transforms nothing if management does not transform the use of saved time.
It accelerates what already exists.
If the company is clear, it accelerates clarity.
If the company learns, it accelerates learning.
If the company is drowning in its own procedures, it accelerates administrative drowning.
With a very nicely designed life jacket.
Conclusion: AI does not forgive absurdity
The NBER study cools down the prophets of instant productivity, but it does not condemn AI.
It condemns the illusion that technology alone creates transformation.
AI does not magically save time.
It gives us an opportunity to look at what we do with time.
And sometimes, the spectacle is uncomfortable.
In your company, is the time saved thanks to AI used to do better, to do something else, or simply to do the same absurdity faster?
Because AI without reflection on work is just a photocopier on steroids.
References
(Developpez) = https://intelligence-artificielle.developpez.com/actu/382362/Des-milliers-de-PDG-admettent-que-l-IA-n-a-eu-aucun-impact-sur-l-emploi-ou-la-productivite-ce-qui-conduit-les-economistes-a-ressusciter-un-paradoxe-vieux-de-40-ans/
(NBER) = https://www.nber.org/papers/w34836
(Atlanta Fed) = https://www.atlantafed.org/research-and-data/publications/working-papers/2026/03/24/03-firm-data-on-ai
(OECD) = https://www.oecd.org/content/dam/oecd/en/publications/reports/2025/06/the-effects-of-generative-ai-on-productivity-innovation-and-entrepreneurship_da1d085d/b21df222-en.pdf
(OECD) = https://www.oecd.org/content/dam/oecd/en/publications/reports/2025/05/the-adoption-of-artificial-intelligence-in-firms_8fab986b/f9ef33c3-en.pdf
(Brookings) = https://www.brookings.edu/articles/the-solow-productivity-paradox-what-do-computers-do-to-productivity/
(McKinsey) = https://www.mckinsey.com/capabilities/tech-and-ai/our-insights/is-the-solow-paradox-back
