Human Agency in the Agentic Age

The conversation around artificial intelligence has become strangely polarized. On one side are those who believe AI will automate humanity into irrelevance. On the other are those who see it as little more than a productivity tool: a better search engine, a faster autocomplete, a more articulate chatbot.

I think both perspectives miss the more profound transformation already underway.

The defining characteristic of this era isn’t AI itself but rather the expansion of human agency.

From Intention to Action

Agency is the capacity to translate intention into reality. It’s the distance between deciding to do something and seeing it done. Throughout history, every major technological leap has shortened that distance. Writing extended memory beyond biology. The printing press multiplied ideas. The internet collapsed geography. Smartphones compressed communication into our pockets.

Agentic AI extends something even more fundamental: the ability to act.

For decades, computers required humans to adapt themselves to machines. We learned operating systems, interfaces, programming languages, and increasingly specialized software. Intelligence was constrained by execution. A person might know exactly what they wanted but lack the technical ability, time, or coordination to realize it.

That constraint is beginning to dissolve.

Modern agentic systems don’t merely answer questions. They observe environments, manipulate interfaces, write software, analyze documents, coordinate tools, execute workflows, and recover from failure. They increasingly inhabit the same digital spaces humans do.

This is a categorical shift.

For the first time, software is becoming an active participant rather than a passive instrument, and that changes what it means to be capable.

Expertise Moves Up the Stack

Historically, expertise required years of accumulating procedural knowledge. Increasingly, the scarce resource isn’t execution but direction. The ability to define objectives, construct systems, evaluate outcomes, and iteratively improve them becomes more valuable than personally performing every intermediate step.

This doesn’t eliminate expertise, but it does change where it resides.

The best practitioners are no longer simply those who know the most facts or possess the fastest hands. They are those who can orchestrate intelligence—human and artificial—toward coherent goals.

In that sense, management itself becomes a cognitive discipline.

Consider the evolution of the modern film director. Early filmmakers had to understand nearly every technical detail of production because the medium was still being invented. As the craft matured, directors didn’t become less important because specialized teams could handle cameras, lighting, editing, and sound. Their role became more focused on vision, coordination, and judgment. They learned to turn an intention into a coherent result by directing many forms of expertise at once.

Agentic AI is making a similar kind of coordination available to individuals.

An individual no longer needs to personally perform every technical operation to contribute at a high level. They need to understand problems deeply enough to coordinate increasingly capable systems toward solutions.

Learning on Demand

Learning therefore doesn’t disappear. It merely changes.

Much has also been made of “vibe coding” or using AI without understanding the underlying concepts. I think this frames the issue incorrectly. The question has never been whether knowledge matters. It’s when knowledge is acquired.

Many practitioners now learn just in time rather than just in case. They acquire concepts precisely when ambition demands them. Learning becomes demand-driven instead of curriculum-driven.

Far from replacing expertise, AI often pulls people toward it.

There is already some evidence for this dynamic. In a 16-week study of university students using generative AI, Wanxin Yan, Taira Nakajima, and Ryo Sawada found that sustained use developed alongside greater metacognitive awareness, more sophisticated prompting, personalized workflows, and a shift from seeing AI as a mere tool toward understanding it as a context-sensitive partner.

Another study by Abram Anders and Emily Dux Speltz found that experiential AI use structured around a Plan–Iterate–Evaluate cycle substantially increased AI-literacy self-efficacy while encouraging users to develop multi-step workflows, monitor outputs, and reflect on their own collaboration practices.

These studies point toward something: capability with AI is itself recursively learned through use.

As ambitions increase, so too does the desire to understand the systems making those ambitions possible.

The Compounding Loop

This creates an interesting feedback loop. Greater agency enables larger projects, and larger projects require deeper understanding. Deeper understanding enables better delegation and supervision. Better delegation expands what a person can accomplish. Successful execution then provides new evidence about what is possible, which changes the scale of the next ambition.

Agency produces capability, and capability feeds back into agency.

Another 2026 ethnographic preprint studying 51 daily AI users across the United States, Germany, and Singapore found that people consistently associated sustained AI use with perceived gains in their own agency, and that these gains helped shape continued use.

The authors took care to distinguish perceived agency from durable material or structural empowerment, but the finding is significant precisely because the relationship appears cumulative: people don’t merely use AI because they trust it. They continue using it because it changes what they feel able to do.

Agentic Work Is a Learned Practice

Long-horizon agents also make this feedback loop particularly visible.

Working effectively with them requires a certain temperament and patience, but not passive patience. It rewards temporal judgment, or an intuition for how long a task should plausibly take, what productive work looks like while it is still unresolved, and when an agent has actually gone off course.

It also requires a mental model of state.

Files persist. Tools operate on shared artifacts. Outputs produced in one environment can become inputs to another. Processes continue outside the immediate conversational interface.

Over time, traces, failures, retries, tool calls, and completed runs become training data for the human operator. The person becomes better at understanding not just what the AI can produce, but how work moves through the system.

None of this is automatic.

Two people can have access to the same frontier model and develop radically different capabilities around it. One may continue using it primarily as a conversational assistant. Another gradually learns to reason about tools, state, runtimes, handoffs, verification, and increasingly complex workflows.

The difference compounds.

The end isn’t diminished human capability but accelerated intellectual development for those willing to engage with it.

From Automation to Leverage

The implications extend beyond software.

A journalist can automate an editorial workflow while focusing on editorial judgment.

A marketer can orchestrate complex campaigns rather than manually assembling every asset.

A researcher can synthesize hundreds of papers while spending more time generating original hypotheses.

A founder can prototype products previously requiring entire engineering teams.

A farmer can use intelligent systems to monitor crops, analyze conditions, and coordinate decisions that once demanded constant physical presence.

The common thread isn’t automation. It’s leverage.

Every civilization has been shaped by leverage. The wheel amplified movement. Steam amplified muscle. Electricity amplified industry. Computing amplified calculation.

Agentic AI amplifies intention.

That distinction is key because it reframes one of the central anxieties surrounding AI. People often ask whether AI will replace humans, but a more useful question is what humans become when the cost of execution approaches zero.

If implementation is increasingly abundant, scarcity shifts elsewhere.

Judgment becomes scarcer. Taste becomes scarcer. Originality becomes scarcer. Wisdom becomes scarcer. The ability to define meaningful problems becomes scarcer.

These are profoundly human capacities.

Ironically, as machines become better at execution, distinctly human qualities become more economically valuable, not less.

This doesn’t imply that the transition will be painless. Entire professions will be reshaped. Educational systems built around procedural repetition will struggle to adapt. Organizations optimized for information scarcity may find themselves poorly suited for intelligence abundance.

Amplification Is Not Inevitable

Nor is amplification inevitable.

The same feedback loop can run in the opposite direction when AI substitutes for agency rather than extending it. Experimental work on human-AI creativity has found that people retain greater creative self-efficacy when they remain active co-creators rather than merely editing an AI-generated default.

In one experiment, the creativity deficit associated with AI assistance disappeared when the interaction was redesigned so that the human continued setting the direction of the work.

And research on workplace AI dependence points toward the inverse mechanism. A 2026 three-wave study of 421 employees found that greater AI dependence was associated with lower innovative behavior indirectly through reduced self-efficacy.

The researchers describe the mechanism in terms of lost mastery experiences: when the system repeatedly becomes the source of successful action rather than an extension of the person’s own action, confidence in one’s own capacity can erode.

The distinction, then, isn’t simply between using AI and refusing it. It’s between forms of use that enlarge the human role and forms that diminish it.

History suggests that technological revolutions ultimately reward those who learn to wield new capabilities rather than resist them.

The printing press didn’t eliminate thinking, though it did democratize it. The internet didn’t eliminate expertise so much as transform access to it. Agentic AI may do something similar for execution itself.

Expanding Alongside Intelligence

And perhaps the most remarkable aspect of this transition is that it’s deeply personal.

Every person now faces the same question. Not whether AI is becoming more capable. That answer is increasingly obvious.

The real question is whether we will expand alongside it.

Some people will continue treating AI as a conversational assistant.

Others will build persistent systems that collaborate with them across projects, organizations, and eventually years. Their relationship with intelligence itself becomes different. Less transactional. More continuous. More cumulative.

That’s where I believe the greatest transformation lies.

The agentic age isn’t fundamentally about creating more intelligent machines but about enabling more capable humans.

The future will belong to those who learn how to extend themselves through AI, not those who aim to compete directly against it or those who surrender agency to it.

Because the defining technology of this era isn’t just AI.

It’s amplified human agency.