In a market where the balance of power between models is reversed in a few days, relying on a tool rather than on skill is like building on sand.
Implement a “real AI strategy” before training the teams? Three years later, we see the result. Companies that waited until they had the perfect strategy before making a move were the ones that fell furthest behind. And those who built their entire approach around ChatGPT are now wondering if they should review everything, because their teams are switching to Claude. This is not a temporary problem. This is the very nature of the market.
The market moves faster than your internal processes
In just a few weeks, Claude has become the most downloaded free application on the American App Store, with a free user base increasing by +60% since January 2026. Companies that had taken out ChatGPT subscriptions for all their teams are in the process of canceling them. Not on a whim, but because on the ground, the quality of outputs is no longer comparable.
And it won’t be the last time. Models are progressing at a speed that no one anticipates. Developers themselves have become more productive thanks to AI, further accelerating the pace of releases. Not to mention the geopolitical dynamics that come into play on the subject: OpenAI’s recent choice to ally with the US Department of Defense, and Anthropic’s refusal to do the same, were enough to trigger a massive migration of users in a few days. Ideological decisions that change the balance of power in the market, overnight. In this context, building a rigid AI strategy around a specific tool means preparing to do it again in six months.
The bad reflex that we see everywhere
Knowing it is one thing, experiencing it in training every week is another. Companies who tell us: “we are waiting until we have chosen our tool before training the teams.” Or worse: “we are currently considering our AI strategy, we will come back to you when it is ready.” Meanwhile, their competitors are moving forward.
The truth is that a ChatGPT or Claude license can be purchased in five minutes and shut down just as easily. It is not an ERP. It is not an irreversible choice. The real question is not “which tool we choose” but “do our teams know how to use an LLM.”
Because someone who has mastered the fundamentals of prompting, who understands how a language model works, who knows how to structure a complex task to derive usable output… someone adapts in two hours when the tool changes. The other remains blocked with each update.
What we need to build is agility, not a doctrine
The most advanced companies that we support have not asked themselves the question of the tool. They asked themselves the question of uses. What do we want to do with AI? What tasks do we want to accelerate? What skills do we want to develop internally? Once these answers are asked, the tool becomes almost secondary.
This new paradigm requires something that our organizations do not yet know how to do very well: remaining flexible on means while being clear on objectives.
Training your teams now, on the fundamentals, on good business practices, on the logic of agents and workflows, this is the only investment that will not be obsolete in six months. It doesn’t matter if in a year everyone is on Claude, on Gemini, or on a model that we don’t yet know. The tool will change. The skill remains.
