What AI gave me… and what it took from me

There is something I read everywhere on LinkedIn about artificial intelligence, and that I find more and more incomplete.

We are told that AI eliminates tasks with no added value. That it frees up time. That it finally lets us focus on what matters. That it increases productivity.

Fundamentally, that is true.

But it is only part of the story.

The real issue is not only the productivity gain, but the hidden human and economic cost.

I say this all the more freely because I am not speaking here as an outside observer. I am speaking about myself, my day-to-day, my work as a business leader and as a developer, and what AI has profoundly changed in the way I produce.

From developer to conductor

Like many people, I entered the phase now called “vibe coding”.

At first, it is exhilarating. You move faster. You prototype faster. You get unstuck faster. You can handle more topics in parallel. You feel like you are changing dimension.

And to some extent, that is true.

But this change comes at a price.

Before, when I spent several hours designing a clean algorithm, solving a tricky business problem, stabilizing complex logic, there was a very concrete kind of satisfaction. A small victory. A moment when you know exactly what you have just built and why it works.

Today, that feeling has shifted.

I no longer savor the small victories.

Not because I produce less. Quite the opposite. I produce more. Much more.

But my role has changed. I have become a kind of conductor. I have to frame, steer, arbitrate, review, correct, test, prioritize, document, check for regressions, and keep overall consistency across projects that are all moving faster than before.

I trust AI more to produce. I do more targeted code reviews. I rely on agents to run tests, including E2E, to guard against regressions. I steer more. I build less directly with my own hands.

Here again, this is not nostalgia.

There is a real gain.

AI shifts the mental load

But we have to be honest about what this gain brings about. AI does not remove mental load. It shifts it. And often, it increases it.

Because by taking away part of the execution work, it mechanically pushes us towards more complex, more abstract, more cognitively demanding tasks. Because by increasing our production capacity, it also increases the number of projects we take on. Because by giving us the feeling that we can go faster, it reduces our tolerance for the long view, for stepping back, for letting things mature.

And here, the problem is not AI alone.

The problem is also our own temptation to do more, faster, with less perspective.

As a business leader, I think we need to have the honesty to acknowledge it.

When you bring AI into your organization, you are not only bringing in a productivity tool. You are also changing expectations, rhythms, cognitive load, the relationship to quality, and sometimes even the relationship to work well done.

You are not simply removing tasks. You are reshaping work.

And that is precisely why I am increasingly uncomfortable with all the sales pitches that present AI as a purely positive given, as a kind of magic lever that would save time, cut costs and increase value with no trade-off.

That pitch is incomplete. And sometimes downright misleading.

The real price of a new dependency

The other thing that strikes me is the real price of this new dependency.

Here too, I am speaking from experience.

The trigger for this reflection, for me, was very concrete. I saw how sharply the cost of use could change depending on how you consume the models. Moving from a Claude Max subscription to using OpenClaw with Anthropic’s API opened my eyes.

I am not putting Anthropic on trial. On the contrary, this change of position forced me to face reality.

We have probably grown used to prices that do not yet reflect the true economic cost of AI.

And that is where many companies, freelancers and small businesses seem to me to underestimate the issue.

Because today, part of the ecosystem still reasons as if these tools were going to remain affordable at very comfortable price levels for the long term. As if the trajectory were bound to stay linear. As if the dependency effect would not, sooner or later, raise the question of real profitability again.

It reminds me of other models we have already seen elsewhere: services massively adopted at attractive rates, then readjusted once they became indispensable.

I think we still underestimate the fact that AI is, at this stage, largely driven by logics of investment, conquest and positioning. And that it would be naive to believe that the current price of use already reflects a stable and definitive truth.

In other words: yes, we save time. But how much time do we need to save to durably absorb the real cost of these tools when they grow in power, in consumption and probably in price?

And above all: what happens to an organization’s profitability if a growing part of its production chain depends on services whose pricing, usage rules and suppliers’ strategic decisions are all beyond its control?

I find this question widely underrated.

Many players today sell AI to their clients by talking almost exclusively about productivity leaps. Very few talk about the cost of structuring, the cost of learning, the cost of supervision, the cost of governance, the cost of dependency, and the human cost.

Yet that is precisely where a company’s maturity is decided.

Organizing rather than slowing down

For my part, this reflection does not push me to slow down on principle. It pushes me to organize better.

I still believe in AI’s immense potential. I use it every day. I see very clearly what it makes possible. I have no desire to go back.

But I also believe that the next step is not to “put more AI everywhere”. It is to rethink how work is organized around it.

• That means framing how it is used.
• That means deciding what should be automated and what must remain fully human.
• That means monitoring real profitability, not imagined profitability.
• That means accepting that extra productivity should not automatically turn into extra overload.
• That means investing in skills, not only in tools.

Reinvesting in people

That is also why I made a very concrete choice in my company: reinvesting in people, by hiring a junior developer whom I will train in these new tools and this new way of working.

This choice is not a step back from AI. It is exactly the opposite.

It is a way of saying that real transformation cannot rest solely on models, agents and subscriptions. It must also rest on passing on knowledge, on training, on people’s skills development, and on our ability not to create a training debt for newcomers to the job market.

Because in the end, the question is not only: “how much can I produce with AI?”

The real question may be: “what kind of organization am I building around it, and at what human, economic and professional price?”

---

I am curious to know how you experience it on your side.

Has AI made you calmer? Or simply faster? And do you, like me, sometimes feel that the real cost of this productivity is still largely underestimated?

Philippe