The Last Human in the Room
On Agents, Behavioral Minimalism, and Why You May Not Need to Enter the Internet at All
The internet was built for human beings.
That sentence used to sound like a statement of purpose. Increasingly it functions as a description of vulnerability.
The dominant systems of the modern internet were designed around human presence. The feed requires human attention to function. The recommendation engine targets human psychology. The dark pattern exploits human cognitive tendencies: the pull toward novelty, the fear of missing out, the anxiety that something important is happening somewhere else. Surveillance pricing targets human vulnerability under conditions of asymmetrical knowledge. The engagement economy depends on one thing above all others.
A human being actually being there.
That dependency may soon become a structural weakness.
The Hostile Environment
It is worth being precise about what hostile means here, because the word is serious and deserves to be used carefully.
The internet is not hostile to human beings the way a storm is hostile, through indifference and overwhelming force. It is hostile in a more specific and in some ways more troubling sense. Much of the contemporary internet has been engineered to work against the cognitive conditions that make human flourishing possible.
Sustained thought requires the ability to remain with something long enough for it to deepen. The engagement economy is optimized to interrupt that. Autonomous preference formation requires exposure to genuine alternatives and the space to evaluate them without continuous pressure. Recommendation systems often narrow that exposure and accelerate the journey toward predictable outcomes. Genuine discovery requires the possibility of encountering something you did not already want. Personalization can reduce that possibility by returning a world organized around prior behavior.
Even boredom matters. Boredom is the psychological precondition for a certain kind of self-directed thought. The moment of reaching for a device rather than sitting with an uncomfortable feeling is also a moment of not developing the capacity to tolerate uncomfortable feelings. The attention economy has made that trade so frictionless and so constant that, for many people, the alternative condition—genuine unstructured quiet—has become unfamiliar.
None of this requires malicious intent from any individual engineer or executive. It requires only the consistent application of engagement as the primary metric of success. When engagement is what a system optimizes, it tends to build an environment that competes with whatever engagement displaces. Attention, reflection, patience, silence, difficult thought, and voluntary self-direction become structurally inconvenient.
This is one way to understand Jacques Ellul’s warning about technique. Technique is not merely a collection of tools. It is a value hierarchy. Efficiency becomes the standard against which everything else is judged. Meaning must explain its usefulness. Wisdom must demonstrate its applicability. Attention itself becomes a resource to be captured rather than a capacity to be cultivated. The internet did not set out to become hostile to humanness. It became hostile to humanness because this is what engagement optimization produces at scale.
The Coming Inversion
Something is changing that most people have not fully registered yet.
Artificial intelligence agents are beginning to move through digital environments on behalf of human beings. They can search, retrieve, compare, negotiate, purchase, summarize, monitor, and complete tasks inside the internet without the human entering the environment directly. They do not need to be there in the human sense. They arrive with a task, act within constraints, and return with a result.
This capability is still early. The agents available today are imperfect, require significant oversight, and cannot yet handle the full range of tasks that constitute a person’s digital life. They make errors. They require supervision. They remain dependent on the systems through which they operate. But the trajectory is clear.
A growing portion of digital life may eventually be mediated by agents acting for people who are somewhere else entirely.
That shift is not merely a change in how tasks get done.
It is a structural transformation of what the internet is.
For the past three decades, the internet has largely assumed direct human exposure. A person searches. A person scrolls. A person compares. A person reads reviews. A person clicks through pages. A person enters the marketplace, the feed, the portal, the app, or the platform.
Agentic mediation changes that assumption.
The person may still define the intention, grant authority, evaluate the result, and make consequential decisions. But the person no longer needs to personally inhabit every environment required to complete the task. The agent can enter certain digital spaces in their place.
That possibility matters because many of the internet’s extractive systems were designed around human exposure.
What Agents Change
The significance of agentic mediation is not that agents are perfectly immune to influence. They are not.
An agent can still be shaped by rankings, paid placement, restricted APIs, adversarial content, platform rules, poisoned information environments, deceptive interfaces, and incentives built into the model or service provider that operates it. An agent can be manipulated differently because the manipulation surface changes. A platform that can no longer hold human attention may attempt to influence the intermediary deciding what the human sees.
That problem is real.
It is also precisely why agent governance matters.
The point is not that agents magically solve manipulation. The point is that they change the site at which manipulation occurs. Today’s internet often works by operating directly on human psychology. It pulls at fear, novelty, insecurity, boredom, urgency, anger, social comparison, and exhaustion. An agent does not experience loneliness. It does not experience fear of missing out. It does not keep scrolling because it feels wounded, bored, curious, flattered, or afraid. It does not linger on a page because a recommendation system found the perfect emotional trigger.
The agent can still be misled.
It cannot be manipulated in the same way a human being can.
That distinction is important. A system designed to capture human attention loses much of its power when the entity moving through the environment has no attention in the human sense to capture. A dark pattern designed to exploit cognitive weakness may still confuse a poorly designed agent, but it no longer operates on fatigue, shame, anxiety, or impulse. A recommendation engine designed to prolong presence has less leverage over an intermediary whose task is to complete the request and leave.
This is behavioral minimalism in practical form.
The principle behind behavioral minimalism is simple: if behavioral exhaust is what the extraction model runs on, generate less of it. Send an agent into the hostile environment instead of going yourself when direct human presence is unnecessary. Let the agent transact under bounded authority and return with what was needed. The person remains outside the machinery designed to capture them.
This does not eliminate behavioral exhaust.
An agent acting on behalf of a person may still reveal highly sensitive information. Searches for medical care, legal help, financial distress, addiction treatment, political organizations, or intimate personal needs can all disclose meaningful patterns even if the person never personally visits the site. The question is not whether exposure disappears. The question is whether exposure can be reduced, governed, compartmentalized, and prevented from being unnecessarily converted into persistent behavioral identity.
The goal is not magical invisibility.
The goal is governed exposure.
At the individual level, this is a privacy and agency problem. At scale, it becomes an economic and political question.
The Economic Question
The attention economy is not a background feature of the internet. It is one of the business models that funds much of the infrastructure ordinary people depend on for daily life: search, communication, news, navigation, commerce, entertainment, and discovery. Much of that economy depends on behavioral surveillance and the sale of predictive access to human attention and behavior.
That business model has historically depended on direct human presence.
If a significant share of digital activity becomes agent-mediated, the economics of attention may begin to change. Agents do not respond to ads in the same way humans do. They do not get drawn into recommendation loops for emotional reasons. They do not produce the same kinds of impulsive reactions that make certain content spread. They do not sit with an app open because it made them feel something.
But this does not mean surveillance capitalism simply collapses.
The model may adapt. Platforms may charge agents for access. They may restrict automated traffic. They may build agent-facing marketplaces. Sponsored placement may migrate from feeds into agent outputs. Retailers may compete to influence the intermediaries that compare products. Model providers may become new brokers of attention, trust, and recommendation. The economic struggle will not disappear. It will move.
That movement is the point.
Agentic mediation raises a question the current internet has not had to answer at scale: what happens to an economy built around capturing human attention when human attention is no longer necessary for a growing share of digital transactions?
The answer is unknown. But the question itself matters because it reveals that the current architecture is not permanent. It depends on a particular relationship between human beings and digital environments. If that relationship changes, the economic foundations built on top of it may also change.
This is why behavioral minimalism is more than a personal preference. It has the potential to become a counter-economic force. Not through boycott, moral pressure, or legislative reform alone, but through architectural substitution. The person does not need to enter every environment directly. The task can be delegated. The result can return. The surrounding machinery receives less of the human presence it was designed to capture.
Why Enter It At All?
The question that follows is simple.
If agents can accomplish many of the tasks that bring people into digital environments, and if those environments are often optimized against the cognitive conditions human beings need in order to flourish, why should the person enter every environment directly?
Not as an act of withdrawal. Not as a rejection of technology, connectivity, or the genuine benefits that digital life provides. But as a practical question.
You do not enter a hostile environment when a trusted representative can conduct the necessary business on your behalf. You do not generate unnecessary exposure when exposure carries real costs. You do not inhabit a space optimized against your interests when the thing you actually wanted can be retrieved and returned without your presence.
The warehouse worker who needs medical information should receive it cleanly, without entering a recommendation environment designed to keep him reading long after his question was answered. The elderly woman shopping for a birthday gift should receive structured results without her browsing behavior being scattered across data brokers before she makes a decision. The person in recovery should be able to find a meeting time without their search history becoming a permanent behavioral signal available to institutions whose interests may not be their own.
These are not advanced users managing sophisticated threat models. They are ordinary people trying to accomplish ordinary things. The environment they currently have to enter to do those things is not a natural feature of information technology. It is the result of design choices that served particular business models under particular technological conditions.
Different design choices produce different environments.
The agent goes. You do not. You receive the result. The hostile environment still exists, but your exposure to it is reduced.
That is not isolation.
That is governed connection.
The difference between entering every environment personally and sending a trusted representative is the difference between being directly exposed to every system you need to use and interacting through an intermediary whose purpose is to preserve your intention while minimizing unnecessary vulnerability.
What the Internet Becomes
When more digital activity becomes agentic rather than directly human, the internet begins to transform from an environment into infrastructure.
It stops being a place people must inhabit and becomes a system agents use to accomplish tasks. That is closer to what the internet was before the engagement economy colonized so much of it: a utility, a protocol layer, a set of routes through which requests move and results return. The web browser was always a window, not a room. The engagement economy turned it into a room people could not easily leave. Agents may turn it back into a window.
The hostile environment does not disappear. Systems of behavioral capture remain. Surveillance infrastructure remains. Pricing strategies, recommendation systems, and commercial incentives remain. But their relationship to the human being changes if the person no longer needs to stand inside them for every task.
A feed designed to capture attention is far less powerful when there is no human attention available to capture. A recommendation loop designed to deepen immersion loses force when the intermediary has no desire to remain immersed. A marketplace designed to extract behavioral signals from comparison shopping receives less signal when the comparison is performed under constraints by an agent that returns only the relevant result.
This does not end the internet.
It changes its function.
The goal is not a world without digital systems. It is a world where digital systems return to a more appropriate role: infrastructure that serves human intention rather than environments that compete for human presence.
The internet was built for humans. Much of it became hostile to humans. Agentic mediation creates the possibility that human beings no longer need to be exposed to every part of the machinery that evolved around their presence.
That is not the end of the internet.
It is the possible end of the internet as an extractive environment that ordinary people must personally inhabit.
The Deeper Stakes
There is a question underneath all of this that is worth naming directly.
Can a free society sustain itself inside an environment structurally hostile to the cognitive conditions democratic life requires?
Democracy depends on people being able to form genuine preferences, deliberate, dissent, associate, and act politically without those activities being continuously observed, modeled, and acted upon by entities with interests in particular outcomes. The secret ballot exists because we understood long ago that observable political behavior is coercible political behavior. The right to think privately before acting publicly is not a luxury. It is a structural requirement of free political agency.
Continuous digital immersion inside extractive environments threatens privacy, but the issue is larger than privacy alone. It threatens the interior conditions under which free political agency becomes possible. Every search, article, moment of doubt, moment of curiosity, or stage of political formation can become part of a persistent record available to corporations, governments, data brokers, employers, insurers, or anyone able to purchase or compel access.
The chilling effect is not theoretical. People self-censor. They avoid certain searches. They moderate expressed views. They stay inside the socially permissible range not because they were commanded to, but because the awareness of observation is enough.
This problem deepens as more of life moves through digital environments and as those environments become more capable of modeling and influencing behavior. A generation raised entirely inside these systems may not experience the attention economy as an intrusion. It may experience it as the condition of ordinary life.
That is why the question of personal digital infrastructure matters politically. Large institutions will have increasingly powerful AI. Governments will. Corporations will. Insurers will. Employers will. Platforms will. Retailers will. Financial institutions will. If the individual has no comparable intelligence operating under their authority, the asymmetry becomes profound.
A free society cannot depend on individuals navigating intelligent institutional environments alone.
This does not mean architecture can replace law. Democratic societies need legal protections, institutional limits, due process, and enforceable rights. But law alone may not be sufficient if ordinary participation in digital life continuously generates detailed behavioral models as a structural byproduct.
Individuals need countervailing infrastructure.
Not anti-social infrastructure.
Not anti-government infrastructure.
Not isolation.
Countervailing infrastructure: systems operating under the authority of the person, capable of managing permissions, memory, disclosure, and delegated action in environments where everything else is becoming more intelligent.
That is the deeper purpose of personal AI.
Not convenience.
Not productivity.
Protection for agency.
The Work
The internet being hostile to humanness is not a permanent condition. It is a consequence of specific design choices made under specific economic pressures during a specific period of technological development.
Different design choices produce different environments.
A world in which people possess AI that acts on their behalf, minimizes unnecessary behavioral exposure, and returns results without requiring immersion in systems optimized against their interests is not a utopian fantasy. It is a technical and architectural direction that becomes more plausible as model efficiency improves, hardware capability increases, and agent governance matures.
The work is to build the philosophical and architectural foundations before the current paradigm consolidates further. To establish behavioral minimalism, agentic mediation, governed exposure, and the principal model as design requirements rather than exotic preferences. To make the case that ordinary people deserve the same protection from hostile digital environments that sophisticated users can currently assemble for themselves with enough technical knowledge.
To insist that you should not have to understand the internet to deserve dignity inside it.
And to imagine, clearly and in detail, what digital life might look like when the human being is no longer required to be in the room.
The agent goes.
You stay home.
You get what you needed.
The environment processes the request, but it receives less of the person than it was designed to capture.
That is the goal.
Not less technology.
Not disconnection.
Not a life outside the modern world.
A life where the modern world comes to you on better terms: through a representative you trust, under authority you govern, in a form you can use, without surrendering the interior conditions that make you human in the first place.
