OpenAI, Anthropic and Google are accelerating their development of a universal conversational assistant, capable of treating as well as purchasing. A strategic battle whose stakes go far beyond simple technological prowess…
In 2030, will ChatGPT be your doctor, your nutritional advisor and your personal purchasing director? The objective may seem excessive; However, it is no longer so unrealistic. Since the end of 2025, the American AI giants – OpenAI, Anthropic and Google in the lead – have entered a new phase. After developing models capable of complex reasoning, they are now focusing their efforts on a much more strategic project: connecting their AI to all third-party platforms. OpenAI has launched ChatGPT Health with connection to medical records via b.well in the United States and is forming retail partnerships with Walmart, Etsy and Shopify for instant payment. Anthropic, for its part, launched Claude for Healthcare in January, allowing Pro and Max subscribers to connect their medical records via HealthEx (which centralizes data from thousands of American hospitals and clinics), as well as Apple Health and Health Connect on Android. Google, finally, has developed a protocol for agent commerce (Universal Commerce Protocol) developed with Shopify, Etsy and Target. Everything speeds up simultaneously.
Still, this sudden convergence raises the question: why now, and why are health and retail emerging as the two priority verticals? The answer has less to do with a technological demonstration than with a precise strategy.
AI technological maturity
Why have publishers been accelerating in recent months on integration at all times with third-party services? The main explanation factor is that the models have reached a real milestone in reliability. “What Anthropic is offering today with Claude 4.5, we simply could not have done six months ago,” explains Clément David, CEO of Theodo Data & AI and Theodo Cloud. For almost two years, publishers sold dreams while knowing that their models were not yet ready for sensitive use cases. Hallucinations remained too frequent, context management too short. “As soon as we got to the fourth, fifth question, the AI was completely delirious,” he remembers. It is impossible, under these conditions, to entrust an AI with the analysis of medical files or the autonomous management of financial transactions. The technology first had to achieve sufficient stability to support critical processes.
Still, the turning point is not just about raw performance. A major behavioral change has taken place: models have learned to recognize their limits. For Clément David, this marker is decisive, particularly in B2C where trust conditions adoption: “The AI is starting to distinguish between what it knows and what it doesn’t know. It’s starting to say ‘I don’t know’.” This is precisely what makes it possible today to consider entrusting Claude or ChatGPT not only with information, but with decision-making processes.
Generate adoption
But technical maturity does not explain everything. If publishers are accelerating now, it is also because users are psychologically ready to take a step forward. “It’s a bit like the iPhone,” summarizes Adrien Gainié, data & AI partner at Saegus. And added: “When he arrived promising to do everything with a single device, it changed our relationship with technology. Tomorrow, we will no longer be able to imagine living without our assistant agent.” Users first asked questions (translations, factual research) then requested advice (product recommendations, document analysis, basic assistance) and are now ready to entrust entire processes to AI (travel reservations, email management, automated monitoring, etc.). “Before, people were not ready to do anything other than ask questions. Now, they are starting to entrust processes,” confirms Clément David. This mental evolution is not trivial, it opens the door to a dependence that is much more structural than a simple research tool.
And this is precisely where the real objective lies: the lock-in. Clément David spins the metaphor of Bill Gates’ “spaghetti dish”: once the cheese melts on the pasta, it’s impossible to remove one without dragging the others. The more a publisher multiplies the layers of integration, protocols, banking connections, health data, the more the output cost for the end user becomes prohibitive. Adrien Gainié underlines: “It is the Google model which is being replicated, with a consolidation of the market to keep a handful of players as much as possible.” The strategy is actually quite simple, generate adoption by proving utility, then create a deep enough customer footprint to make exit unthinkable.
Health and retail: strategic choices
If Anthropic, OpenAI and Google are investing massively in health, it is not out of altruism. “Once you have done health or defense, you can do what you want,” assures Clément David. Health is the ultimate flagship. The constraints are maximum: drastic regulations, zero margin of error, requirement for absolute reliability. Succeeding in such an area is equivalent to performance certification for all other use cases.
Retail has a more direct objective: capturing purchasing intent before the transaction. Clément David is explicit: the challenge is “to arrive at the moment when the user intends to buy, to be able to influence them”. The land is already ready. Ten years of cookies, behavioral tracking and journey analysis have mapped purchasing habits with surgical precision. A strategy which, however, seriously worries brands. Adrien Gainié observes a rise in anxiety among his luxury clients: “When we presented the issues of generative referencing to the executive committee of a large luxury group, it really worried them. They realized that it was no longer a gadget but a competitive risk.” The problem is simple, the traffic no longer goes through the brand sites, it remains captive in the assistant. Brands therefore find themselves in a race for direct integration, under penalty of disappearing from the purchasing journey.
In this mad race for general AI, consolidation is inevitable. Adrien Gainié is categorical: the market will only be able to support two to three dominant players, “like what has been done on web browsers or search engines”. In B2C, the question will therefore be: who will achieve the best lock-in via intimacy and personalization? Everything pushes publishers to create the deepest footprint possible before the market shifts. Whoever makes the exit cost prohibitive will win the battle. The question is therefore no longer whether these assistants will impose themselves, but who will impose them.
