Agentic AI is on the move. From Google Cloud Next in Las Vegas, an overview of the reality of current deployments and the transformations it promises.
Scalability, governance, security. From Google Cloud Next in Las Vegas, Google has attempted to respond to the three major issues facing businesses regarding agentic AI. Gemini Enterprise Agent Platform, eighth generation TPU, red teaming agents: the announcements have been numerous and the ambitions clearly displayed. Agentic AI is no longer limited to optimizing operational costs, it must now create new growth opportunities, transform business models and, beyond the business, tackle problems that humanity has not yet managed to solve. The promises match the investments: historic.
Two-speed deployments
On the French market, the deployment of agentic AI follows a contrasting pace. As for the large CAC 40 groups, the first agents in production exist, but they remain limited to very specific areas: supply chain, demand forecasting, logistics optimization. Anthony Cirot, VP EMEA South at Google Cloud, is direct on the subject: “There are not thousands of agents” among French customers. General management wants to move quickly, even “go to production tomorrow morning”, but operational reality slows down scaling. To date, organizations have on average 2 to 5 use cases in production, out of 150 to 200 cases identified.
The publication of Mythos by Anthropic a few days before the conference had already put customers under pressure, to the point that Google Cloud teams had to intervene upstream to reassure major accounts about the real implications. In this context, the clients present at Next, more than a hundred large European groups, particularly remembered the security aspect of the event. Announcements around agentic detection and remediation, the integration of Wiz and the ability to deploy security agents across the entire infrastructure generated immediate reactions, Google Cloud’s EMEA management assures us.
The company of the future will be agentic or not
Even more interesting, announcements on agentic governance are beginning to give a first idea of the business of the future. For Anil Jain, global managing director strategic industries at Google Cloud, the model that is emerging is that of an augmented employee, permanently assisted by an agent who himself orchestrates a network of specialized agents. “Not like a calculator, but like a real collaborator,” he sums up. And added: “The creativity, judgment, and empathy that humans bring to decisions will be maximized, because they will have more information, more guidance, and the ability to act more quickly.”
Asked about the possibility of billion-dollar companies emerging with no employees and only agents, Anil Jain is blunt: “It will be very niche and very limited.” For the expert, only a few very specific cases lend themselves to this: highly technical systems, based on infinitely repeatable and scalable processes, where the value generated does not depend on human interaction. But for the vast majority of legacy sectors, AI will augment the organization, not replace it.
The “breakthrough” moment for science and health
But agentic AI will not only transform businesses. It promises, on paper, to find solutions to problems on a planetary scale that no human has been able to solve. The discovery of new materials, the acceleration of drug development, the fight against climate change: so many areas where agents, backed by the scientific models founded by Google DeepMind, could compress decades of research into a few years. The ability of agents to orchestrate massive processing flows makes it possible to scale even more quickly on the research side. “There is so much information to process, so many ideas to test, that if you have super-powerful assistants processing all of that in parallel, imagine what we could accomplish,” enthuses Anil Jain, a scientist by training.
But this promise comes up against a constraint that Google now assumes publicly: demand for computing currently exceeds available capacity. It is in this context that the eighth generation TPUs announced at Next take on their full meaning. “Most people don’t understand what that really means,” acknowledges Anil Jain, “but the cost/performance stakes are considerable: you increase performance, you reduce energy consumption, you do more with less.” So even though overall demand continues to grow, the unit energy cost falls as per-chip performance increases.
TPU, agentic governance, security agent… Google answered almost all the technical questions in Las Vegas. But the main obstacle to adoption remains, according to Anil Jain himself, the management of change within organizations. Developing the skills of the teams, bringing the professions on board and creating trust… This is truly where the race is at stake. And in this area, companies (particularly French and European) still have a way to go.
