Between late 2025 and the first half of this year, GitHub — one of the critical technological platforms where most software is organized, reviewed, and published — began showing clear signs of structural stress, becoming completely unavailable for extended periods of time. In October 2025, the company itself acknowledged that it had launched a capacity expansion plan to increase its infrastructure tenfold. The problem, however, was not only left unresolved: it accelerated. By February of this year, that original estimate had already fallen short. The company admitted that it now needed to multiply its infrastructure thirtyfold to support its current load. So what is driving such a sudden shift? Artificial intelligence agents.
The importance of the GitHub case lies in the fact that it prefigures a test case for this new reality, in which AI agents begin to take on a dominant role. What appears, at first glance, to be a circumstantial problem should in fact be understood as something that will necessarily extend to any system built on a premise that, until now, seemed obvious: that the key user on the other side was human. Public APIs, free tiers, support flows, online stores, communities and social networks, authentication systems, wikis, corporate chats, and collaborative tools were all designed for a population of users who act according to a human cadence. But agents suddenly introduce the need to rethink and readapt the current operational regime. At scale, the Internet is beginning to receive users who never tire of trying, who autocomplete forms, who do not forget open tasks, and who do not stop when human attention, in general terms, runs out.
Much of the web and its systems have been anthropomorphized. This is expected. This is reasonable. Systems adapt to their users. Humans type slowly, switch between tasks, take time to interpret screens, then rest, then make mistakes, get bored, and abandon the task. For years, that fragility functioned as an invisible regulator of demand. Agents are coming to dismantle that regulatory scheme. Agents can open tickets, query APIs, compare results, generate variations, test alternative paths, and retry operations with a persistence foreign to the economy of human attention. That is why the crisis first appears at the level of infrastructure. But we should not be naive and think that this change will stop there. Eventually, it will reach every service whose design depends on a premise now being called into question: that the key end user is human. And this will necessarily continue across the rest of the constitutive layers of systems: infrastructure, interfaces, business models, and governance. The GitHub case is only the first tremor.
Compared with previous autonomous systems, the difference lies in the degree of delegated agency. Classical bots, for example, crawled pages, extracted content, published spam, or executed narrow routines. Agents, by contrast, interpret objectives, build plans, use all kinds of tools and applications, observe results, and create and synthesize the information they receive in order to update the plan and return to the task. They operate with complete naturalness in interfaces designed for humans. They search for us, code for us, converse for us, test vulnerabilities for us, and request support for us. And increasingly, each of these agents will find that the entity with which they interact — whether when receiving the result of a product search, a review of the code they generated, offers, or responses to support requests — will itself be another agent. On the horizon, we can glimpse a global-scale agent economy. Current systems will either adapt or disappear in favor of those capable of interpreting this new historical circumstance: the transition from the Anthropocene to the Robocene, as we explored in Chapter 4: “Obsolete.”
The ideas of the French urbanist Paul Virilio help us deepen and construct a perspective on this phenomenon. His thesis is that every technology invents its own accident: the ship brings with it the shipwreck; the airplane, the air crash; the agentic web, the saturation, degradation, and eventual retirement of anthropocentric systems unable to adapt to the new times. The accident reveals the kind of fragility contained within the promise. In the same way, the promise of agents consists in automating, at scale, virtually all the actions we perform in front of a screen. Perhaps we should ask ourselves about the inevitable future accident already taking shape. The GitHub incident then ceases to be a marginal exception and begins to function as a symptom of a much larger and unprecedented process.
The logical consequence is that the human will cease to be the central actor of the Internet. Humans will continue to define desires, objectives, and criteria of value, but many concrete actions will be carried out through agents. That displacement will transform every constitutive layer of distributed systems. First, it will fall upon infrastructure: more traffic, more events, more compute consumption. Then it will reach anthropocentric interfaces: what sense does it make to invest monumental effort in visual experiences optimized for humans if the key user will be an agent that prefers an API, a protocol, a semantic contract, or a structured channel of actions? Finally, it will reach business models: free tiers, usage limits, and engagement metrics will need to distinguish between the increasingly irrelevant human user and the far more active agentic user.
The GitHub drama thus becomes a loud warning that makes explicit the transition between two incompatible eras of the Internet: one organized around human visitors, and another that will have to organize itself around these invaders. And so, as Paul Virilio warns us, the key question seems to be how we are preparing for the inevitable accident toward which we are moving.