For ten years, IT infrastructure has continued to transform: from virtualization to cloud-native platforms, including containerization.
And if this evolution has largely shaped the current technological landscape, a more discreet transformation is profoundly changing the way organizations operate: that of agentic operations.
With the growing maturity of AI, the biggest opportunity is no longer simply to develop applications faster. It is to use intelligent agents to manage, operate and evolve the infrastructure itself. In 2026, I think we will see a real shift: from AI as a development support tool towards operational autonomy driven by AI, particularly in complex environments such as AI factories and sovereign clouds.
So far, the excitement around AI has mostly focused on development assistants and productivity gains. But when we look at the operational side of the business, with networking, storage, virtualization, bare metal provisioning, or cluster lifecycle management, the adoption curve is still very embryonic, and I don’t think it will last.
With the deployment of increasingly complex AI infrastructure stacks, manual operations simply will not scale. AI factories require automation from the physical layer to orchestration and application workflows. The real question businesses are starting to ask is no longer “How can AI help me write code?” ”, but “How can AI help me run everything?” »
Agentic systems provide a path forward. Rather than thinking of automation as a collection of scripts and dashboards, companies are starting to consider autonomous workflows capable of triaging incidents, adapting policies and executing corrective actions without ongoing human intervention.
We are, in fact, moving towards an infrastructure that can largely manage itself, guided by human intent, but executed by intelligent agents.
Why protocols will become decisive
One of the most important developments in this evolution is the emergence of standardized ways for agents to interact with systems, with the industry converging towards a common approach to connecting tools, workflows and AI decision-making.
Historically, organizations have built highly customized automation systems tailored to specific environments. But with the rise of generalist AI agents, another model is emerging: rather than creating bespoke agents for each domain, teams can enrich generalist agents with specialized tools and capabilities that allow them to operate in infrastructure environments.
This shift is transforming the innovation economy: organizations are leveraging advances in large AI models and general-purpose agent frameworks rather than rebuilding intelligence from scratch, resulting in faster iteration, less operational overhead, and a more flexible ecosystem that evolves in step with the overall AI landscape.
The end of tailor-made agents
Today, many teams are experimenting with specialized agents, which are bots designed to perform specific operational tasks. These experiences have their value, but most companies will ultimately opt for generalist agents augmented with tools specific to their field.
The reason is simple: general agents inherit rapid improvements from the broader AI ecosystem, and when models advance, reasoning capabilities expand, and integrations multiply, organizations automatically benefit… without having to rewrite their automation stack.
In practical terms, infrastructure workflows could soon look like something very different from what we know today. A general-purpose agent could generate operational logic, deploy it in an execution environment, and manage long-running tasks asynchronously. Rather than static pipelines, we will see adaptive systems where agents continuously evolve the code that drives the behavior of the infrastructure.
Imagine a production incident triggering an autonomous triage process: rather than relying on pre-written runbooks, an agent would analyze the telemetry, generate the remediation logic, and execute corrective actions while learning from the results over time. And it’s no longer just theoretical, since experimental systems already demonstrate this model.
The reality of corporate adoption
Despite the enthusiasm, enterprise adoption remains the main challenge. Any new technology goes through a classic cycle of validation, security audit and organizational transformation, and agentic systems are no exception.
Security teams are understandably cautious. AI introduces new attack surfaces and new compliance constraints. Companies must evaluate how agents access systems, how decisions are audited, and how risks are controlled. The good news is that many cybersecurity teams are already developing a detailed understanding of these technologies, and the questions they ask today are much more advanced than those in the early days of the cloud.
The other challenge is increasing skills. Not in the traditional sense, but in learning how to apply these tools effectively. AI agents excel at certain types of reasoning and automation, but they are not a replacement for deterministic software. Organizations need to rethink their workflows rather than inserting agents into existing processes. True value emerges when companies ask themselves from the ground up: How can our operations evolve when AI agents are part of the system?
Where to start
For those navigating this transition, the first step is simple: start using technology.
Organizations that have successfully adopted cloud-native have done so by giving their teams the freedom to experiment, learn, and integrate new tools into real-world workflows. And the same logic applies to agentic operations. Top-down guidelines can accelerate adoption, but deep transformation happens when teams anchor AI in their daily work, not as a curiosity, but as a core operational capability.
This also implies a change of posture: AI agents are not simple productivity tools but really open the way to entirely new workflows. Marketing, development, operations and customer support will all evolve with the arrival of agents as active participants in decision-making processes. The competitive advantage will not go to the first to deploy agents, but to the first to have reorganized their operations around them.
From this perspective, our industry’s priorities remain anchored in two realities. Enterprises continue to need reliable, scalable AI infrastructure, particularly for sovereign and hybrid deployments, and demand for GPU capacity continues to outstrip supply, making operational efficiency critical.
And the technology must deliver tangible business value: the goal is not automation for automation’s sake, but the identification of high-impact use cases where agents can improve reliability, reduce operational burden or accelerate innovation.
Crossing the AI chasm
Each major technological breakthrough forces organizations to reinvent themselves: the transition from virtualization to containers has transformed application delivery, the rise of cloud-native has redefined infrastructure management. Agentic operations represent the next leap in this same trajectory.
In 2026, we will see the emergence of an infrastructure that is no longer content with running workloads, but which actively participates in its own management. The combination of general-purpose agents, evolving protocols, and enterprise experimentation will push the industry beyond incremental automation to something far more transformative: systems that continually adapt, optimize, and improve. Organizations that embrace this shift today will not simply adopt AI, but will completely redefine how modern infrastructure is built and operated.
