Back to the journal

Adoption IA

The biggest AI revolution is not technological

Studies show that, used well, AI makes people better. Why so many programs disappoint, and how to steer AI adoption around people.

Author
Tomorrow Solutions
Published
November 16, 2026
Reading
9 min
The biggest AI revolution is not technological

It is human. What the research shows, what I see on the ground, and what it means for those who run an AI programme.

In executive committees, one question comes up more and more often: how many jobs will AI allow us to save? Some leaders are already asking why they should keep hiring, since AI can do part of the work.

The question is legitimate from an economic standpoint. It is not the one that drives me. After more than twenty years of transformation work at Sony, Mars and Mastercard, and two years supporting AI adoption in large organisations, I hold one conviction: no technology changes a company for good unless its people grow with it.

So the real question is this: what can we become thanks to AI?

What the research says: AI first makes people better

Two landmark studies shed more light on this debate than most speeches.

The first was conducted by Harvard Business School with Boston Consulting Group. It covered 758 consultants, or 7% of the firm's consultants. On tasks representative of their work, those using AI completed 12.2% more tasks, 25.1% faster, with results of 40% higher quality (VentureBeat, Mi-3).

Yet that is not the most interesting finding. The researchers described a "jagged frontier": outside it, AI output is unreliable and can even degrade human performance. In other words, AI makes people better when they know when to use it and when to distrust it. Human judgement becomes more valuable (The Harvard Crimson).

The second study, by Erik Brynjolfsson (Stanford) and his MIT colleagues, followed 5,179 customer support agents. Access to the tool raised productivity by 14% on average, and by 34% for novice and less-skilled workers, with minimal effect on the most experienced. The authors note that the model spreads the best practices of top performers and helps newcomers learn faster. The assistance also improved customer satisfaction and employee retention (NBER).

Both studies say the same thing: used well, AI is first an accelerator of learning and skill. It lifts people up. That is the exact opposite of a replacement logic.

The paradox: plenty of tools, little transformation

If AI makes people better, why do so many programmes disappoint?

The MIT NANDA report published in 2025 made headlines: despite 30 to 40 billion dollars of enterprise spending on generative AI, 95% of organisations see no business return. The authors attribute this failure to brittle workflows, weak contextual learning and a mismatch with day-to-day work. They add that these tools mostly improve individual productivity rather than company performance (Virtualization Review, Think Digital).

BCG's annual AI at Work study, conducted in 2025 with more than 10,600 people, offers a complementary reading. More than three quarters of leaders and managers use generative AI several times a week, but regular use among frontline employees stalls at 51% (BCG).

The levers it identifies are all human. When leaders clearly support AI, the share of employees who view it positively rises from 15% to 55%. Only a quarter of frontline employees say they receive that support. On training, 79% of those who received more than five hours are regular users, compared with 67% of the others (BCG, Unleash).

So the blocker is rarely the tool. It lies in the support, the meaning given to the programme and the trust of the teams.

Why a replacement logic makes programmes fail

Teams quickly understand the intent behind a programme. If AI is presented as a way to cut headcount, they protect themselves. They test little. They keep quiet about what does not work. They do not bring their real use cases, the ones that would create value. The programme then moves forward in the reports, much less in real work.

Fear is not limited to the teams. According to BCG, the fear of losing one's job to AI is even stronger among managers and leaders (43%) than among frontline employees (36%). A programme that leaves its intentions unclear therefore loses its natural champions (IT-Online).

Counting jobs also makes us forget what AI does not do. It does not carry the responsibility for a decision. It does not know a client's history, the tensions of a negotiation or a regulator's expectations. It does not hold a relationship. All of that remains human, and that is often where the value lies.

Three capabilities AI can develop

My work comes down to one sentence: helping those who make decisions become better thanks to AI, without ever replacing them. In practice, I see three gains.

More perspective to understand. An expert drowning in urgency processes cases but no longer thinks. When AI takes over part of the research, monitoring and synthesis, the expert gets time back for judgement, which is why they were hired in the first place. On the ground, it often starts with simple gestures: summarising a forty-message email thread before a meeting, drafting a briefing note from three documents, doing a first review of a long contract.

More clarity to decide. A well-summarised file, disagreements spotted earlier, options laid side by side, risks listed before the meeting rather than after: the decision stays human, and it is better informed. The "jagged frontier" sets one condition: check what AI asserts. Clarity comes from combining the speed of the tool with the rigour of the person.

More capacity to act. This is the most underestimated gain. In a legal department I support, lawyers with no technical background have built their own agents: litigation monitoring, alert tracking, summaries of major contracts, monitoring of legal reforms. They use them in their real work. No one has been replaced. They now do what they could not do yesterday, and their colleagues in other countries now want the same tools.

This last point matters: when people grow, they become the programme's best ambassadors.

The triad that works: leadership, HR, IT

AI adoption is a company project before it is an IT project. What works best is a triad in which each party plays its role.

Leadership sets the direction. It says why the organisation is investing in AI, what it expects from it and what it does not. It leads by example by using the tools itself. The BCG figures above show how much this visible support weighs on buy-in.

HR carries the business lines and keeps managers in the loop. HR connects AI to skills, career paths and tomorrow's roles. The manager is the decisive link: the one who allows, encourages and rewards experimentation in the team.

IT owns governance and the usage framework. Clear rules on data, security and authorised uses do not slow teams down. They give them the confidence to try. In regulated environments, legal and compliance must be involved before deployment, not after.

When one of the three is missing, the programme limps. Direction without a framework creates shadow usage. A framework without direction creates indifference. Direction and a framework without managers create licences that sit idle.

Measure what people can now do

An AI programme that only looks at the number of logins misses the point. I recommend tracking four levels, separately:

  1. Participation: who attends training and sessions, by team and by country.
  2. Regular use: how many days of use over a fixed period, month after month. An active user is not necessarily a regular user.
  3. Adoption: which methods, agents and gestures have truly entered daily work, and are shared.
  4. Impact: concrete cases described by the people themselves, with what changed in their work. Behind these four levels, one question should guide steering: what new capability has the team acquired? A usage metric that is not linked to any new capability says nothing about return on investment.

Five decisions to put people at the centre

  1. State the intent clearly. Write down what the organisation expects from AI, and what it does not. Silence about intentions is read as a threat.
  2. Start from the tasks that weigh most. Ask each business line which tasks take the most time or energy, rather than presenting the tool's features.
  3. Set simple usage rules. One page, approved by legal and compliance, that anyone can read in five minutes.
  4. Rely on champions within the business. People who show what works with their own files. Nothing convinces a lawyer more than another lawyer.
  5. Track capabilities, not just usage. Report to sponsors on what teams can now do, with real examples. Each of these decisions is modest. Together, they move the programme from technical demonstration to daily work.

Four questions to ask yourself this week

  • Do our teams know why we are deploying AI, and what it means for their role?
  • Have our managers been trained themselves, and do they know how to support their teams?
  • Do we know how many people use AI regularly, not just how many hold a licence?
  • Can we name three things our teams do today thanks to AI that they could not do six months ago? If an answer is missing, that is where the next lever lies.

It is people who will change AI

We often ask how AI will change people. I believe the opposite: people will change AI, provided they are given the means to do so.

More perspective to understand. More clarity to decide. More capacity to act.

What new capability has AI already given your teams?

Is your AI programme moving more slowly than planned? Let's talk: we will look together at where usage is stuck, and how to get it moving again.

Sources

Turn this analysis into a decision.

A 30-minute diagnostic identifies the main friction point in your AI program and the next useful checkpoint.

Book a diagnostic