AI Anxiety and Early Capitalism: Who Controls the

Venkatesh Rao’s essay “Cognitive Mercantilism” compares today’s AI anxiety with the upheaval of early capitalism. That puts a different question on the table from the usual chatbot challenge: can the machine do my job? A system can change a job long before it can replace the person doing it. Someone still writes the final email, checks the code or answers the customer, but the employer may now decide which steps require human judgment and how much that judgment is worth.

That’s the useful edge of Rao’s analogy. Early capitalism reorganized who controlled productive tools, how people sold their labor and who collected the returns. AI could also change the bargaining relationship around a skill. If a worker’s hard-won expertise becomes one input in a company-controlled workflow, the worker may gain a powerful assistant, lose some autonomy or experience both on the same Tuesday.

The analogy has a limit: history offers a set of questions, not an itinerary. AI runs through digital infrastructure, a relatively small group of developers builds the most prominent systems, and organizations can add them to existing work without rebuilding a factory. The pace and the points of control differ. So what should a worker, manager or person paying for an AI subscription actually watch?

The Job Title Hides the Change

A job is a bundle of tasks. Consider a support agent who looks up policies, interprets a customer’s problem, drafts a response and decides when to escalate. An AI tool might speed up the search and drafting. It might also encourage a manager to expect more conversations per hour. Those are two plausible consequences of the same purchase, and a benchmark showing faster drafts cannot tell you how the workday changes.

That’s why AI deskilling deserves attention alongside head-count forecasts. When software takes over the first pass, people may spend more time reviewing and managing outputs. Review can demand expertise: spotting a plausible but wrong answer is work. Yet a workplace that reserves only approval clicks for humans may give newcomers fewer chances to learn how to produce the underlying answer themselves. The design of the workflow decides whether assistance builds skill or gradually sidelines it.

A displaced task also needn’t produce a displaced worker. Faster drafting could let a team handle requests it previously left unanswered, or it could let the organization shrink the team. Customers might get quicker service while workers face tougher quotas. Calling any one of those outcomes inevitable skips the choices employers make about staffing, prices and quality.

The more revealing question is what happens after the demo. Does the person doing the work have time to verify an answer? Can they challenge a faulty suggestion without being marked down for speed? Who is responsible when a polished response turns out to be wrong? Those decisions tell you more about AI at work than a side-by-side contest between a model and a human on one isolated task.

Productivity Has an Owner

The Industrial Revolution comparison extends beyond Rao’s essay. Economists’ concerns about the AI boom have also prompted Industrial Revolution comparisons. The attraction is easy to understand: a technology can increase what an economy produces while disrupting the people and places caught in the transition.

Productivity and pay, though, answer different questions. A business that completes more work per hour has options. It can lower prices, improve service, raise wages, shorten hours or retain the gain. More output alone doesn’t tell workers which option management will choose. The emerging debate over AI productivity gains therefore sits beside a distribution question: whose time becomes more valuable, and who can claim a share of the savings?

Ownership starts with the models and infrastructure, but it doesn’t end there. A large organization may buy access to a system and control its rollout while an individual worker has little say over the pace of adoption. A small business may get capabilities it could never have built itself, yet become dependent on a vendor’s pricing and product decisions. A customer may enjoy a faster answer without knowing how much human review went into it. The same tool can give one participant leverage while narrowing another’s choices.

A subscription price is only the visible sticker. Switching costs can include retraining a team, rebuilding routines and moving information between services. For a person deciding whether to pay for an AI assistant, the practical version of the ownership question is refreshingly unglamorous: can you export what you made, understand what you’re paying for and stop using it without losing your workflow? You don’t need to own a data center to care about lock-in.

Where the Historical Analogy Helps, and Where It Cheats

“Early capitalism” describes a long and varied transformation. Turning it into a single story with a guaranteed ending would flatten both history and AI. Workers’ circumstances differed across industries and places; the gains from new systems depended on institutions and power, not just on whether a tool performed well. That makes the comparison useful for asking who sets the rules. It cannot forecast how many jobs AI will create or remove, or when any one profession will change.

AI also arrives as software that an organization can insert into an existing process. A company can change an internal tool faster than it can build a new plant. That speed can help workers shed tedious steps, and it can leave little time to debate expectations before those expectations become the new normal. A rollout plan that specifies what humans should check, how mistakes get reported and whether staff can refuse unsuitable uses is a different proposition from simply handing everyone an account and raising targets.

The strongest optimistic case deserves its full weight. Cheaper access to capable tools could help people who lack a large team draft, research, code or organize work they otherwise couldn’t attempt. In a workplace, assistance could free experienced staff for harder problems and give beginners more guidance. Those benefits depend on cost, reliability and room for human learning. If the system’s answers require careful checking, the saved time may be smaller than the demo suggests. If access gets expensive or a worker never learns to work without the tool, the initial advantage can carry a longer-term cost.

The strongest concern is about control. An employer can use AI to increase output while workers absorb the pressure of verification and the risk of errors. A vendor can supply indispensable infrastructure while customers bear the difficulty of switching. Neither outcome follows automatically from a model’s capabilities. Both become possible through the contracts, budgets and workplace rules wrapped around it.

Questions to Ask Before the Rollout

For workers, the first question is concrete: which steps will the tool handle, and which decisions remain yours? Ask whether performance targets will change, who checks mistakes and how new staff will learn skills the tool performs for them. A promise to “save time” is incomplete until someone says who gets that time back.

For managers, a useful trial compares the whole workflow before and after adoption, including review and correction. An answer that appears in seconds can still cost time if it sends a colleague on a search for the error. Track whether staff can flag problems and whether customers get the service they were promised. Adoption numbers by themselves measure uptake, not benefit.

For buyers paying out of pocket, start smaller. Test a tool on a task you already understand. Check the result, the recurring cost and how easily you can take your work elsewhere. The best option may be a cheaper service, an existing feature in software you already use or no subscription at all. Paying for the flashiest model makes little sense if the work you need done fits comfortably inside a simpler tool.

Rao’s capitalism comparison earns its keep when it pushes those questions into the open. A machine’s ability to produce an answer is one event. Deciding who sets its price, checks its work, learns alongside it and benefits from the time it saves is the economic transition.