AI Execution Trap: The Rise of the Third Kind of Work

The Third Kind of Work: What Bertrand Russell Couldn’t Have Predicted About AI and the Future of Labor

In 1932, Bertrand Russell published a short, razor-sharp essay called In Praise of Idleness. In it, he split all human work into two categories: first, altering the position of matter at or near the earth’s surface; second, telling other people to do so. The first kind, he noted, is unpleasant and poorly paid. The second is pleasant, prestigious, and capable of indefinite bureaucratic expansion.

Nearly a century later, this taxonomy still haunts us. Not because it’s wrong, but because it reveals a paradox that most founders, engineers, and knowledge workers feel in their bones yet struggle to articulate.

And now, AI is about to break it wide open.

The Paradox Nobody Talks About Here’s the version of the paradox that hits home for anyone who has ever scaled a company or moved from an IC role into leadership.

You start as a maker. You write the code. You build the prototype. You design the architecture. You are, in Russell’s terms, directly altering matter. The feedback loop is tight and gratifying, you push code, something works (or breaks), you iterate. There is a physical, almost visceral sense of creation.

Then the company grows. You hire. You delegate. Suddenly your calendar is wall-to-wall meetings. Your primary output is no longer code or designs , it’s emails, Slack messages, strategy decks, and one-on-ones. You have become Russell’s second type of worker: someone who tells other people to do things. And for many founders and senior engineers, this transition triggers a quiet existential crisis. A feeling that you’ve stopped creating and started merely talking.

This is what I call the productivity guilt trap. And it’s built on a philosophical error.

Communication Is Not the Opposite of Creation, It’s a Higher-Order Form of It Two strands of thought dismantle this guilt.

The first is philosophical. Hannah Arendt, in The Human Condition, distinguished between Labor (survival), Work (fabrication of durable things), and Action (speech, coordination, human interaction). She argued that Action, the interpersonal, communicative realm, is not subordinate to Work. It’s the only category of human activity through which we can aggregate individual capabilities into collective force. A coder writing a function is engaged in Work. A founder aligning fifty engineers toward a shared architectural vision is engaged in Action. Both create. They just create different things.

The second strand is economic. Ronald Coase asked a deceptively simple question in 1937: “if free markets are so efficient, why do firms exist at all?” His answer was transaction costs. It’s cheaper to coordinate people inside an organization through communication and management than to contract every micro-task through the open market. The manager who “tells others what to do” is not parasitic. They are actively bypassing the friction that would otherwise make complex production impossible. The economic value of that coordination equals the sum of the transaction costs the firm avoids.

Andy Grove made this operational with a single equation: a manager’s output equals the output of their organization plus the output of neighboring organizations under their influence. A one-hour meeting where you clarify strategic direction for ten engineers doesn’t produce one hour of value. It produces 400 hours of correctly-aimed labor. That’s a 1:400 leverage ratio. The speech act is the creation.

Enter AI: The Third Kind of Work Russell gave us two kinds of work. The knowledge economy blurred the boundary between them. AI is now introducing a third.

For the first time in history, we have a non-human agent that can both alter digital matter and coordinate the efforts of others, including other AI agents. An LLM doesn’t just answer questions. It writes code (altering digital matter), it drafts strategy documents (organizational communication), it orchestrates multi-step workflows (coordination), and it does all of this at near-zero marginal cost. This isn’t a minor efficiency improvement. It’s a categorical shift in the nature of work itself.

Consider what’s happening right now. AI coding agents write, test, and deploy features autonomously. AI project managers break down user stories, assign priorities, and generate sprint plans. AI research agents scour documents, synthesize findings, and produce analysis that used to take a team of analysts weeks. In Russell’s taxonomy, AI is simultaneously a first-kind worker (altering digital matter) and a second-kind worker (coordinating and directing the flow of effort). It collapses the distinction entirely.

What This Means for Humans “If AI can do both types of work, what’s left for us?”

The answer lies in the concept that Paul Graham identified as the deep tension in organizational life: the Maker’s Schedule vs. the Manager’s Schedule. Makers need long, unbroken blocks of time. Managers thrive in fragmented, high-context-switching environments. These two modes of working are fundamentally incompatible, and organizations spend enormous energy managing the friction between them.

AI eliminates much of this friction, not by removing one schedule, but by absorbing the administrative burden of both.

For makers, AI handles the boilerplate, the scaffolding, the repetitive implementation details. It compresses the cycle time from idea to working prototype. A developer who once spent 80% of their time on implementation and 20% on design can now flip that ratio. The human becomes the architect; AI becomes the builder.

For managers, AI absorbs the low-leverage communication overhead, status reports, meeting summaries, information synthesis, and routine decision support. A manager who once spent their days in information-gathering meetings can instead focus on the genuinely high-leverage acts: strategic judgment, cultural definition, and the kind of human coordination that no algorithm can replicate.

Brian Chesky calls this “Founder Mode”, the practice of staying deeply embedded in the details of your product while simultaneously architecting the organization at the highest level. Before AI, Founder Mode was brutally difficult because one person’s bandwidth is finite. AI expands that bandwidth dramatically. It lets founders and senior leaders maintain the unbroken chain between strategic vision and ground-level execution that organizational theorists have identified as the single greatest predictor of company health.

The Real Shift: From Doing to Deciding Here’s the uncomfortable truth that most AI discourse avoids: the nature of valuable human work is shifting from execution to judgment.

In the pre-AI knowledge economy, your value as a professional was partly in what you knew and partly in what you could produce. Knowing how to write a SQL query, design a React component, or draft a legal brief was valuable because the production itself was the bottleneck. AI removes the production bottleneck. When any competent professional can use AI to produce a first draft of almost anything in minutes, the scarce resource is no longer the ability to produce; it’s the ability to judge. To know whether the output is good. To understand the second-order consequences of a strategic decision. To feel the subtle wrongness of a product direction before the data confirms it.

This is where the Communicative Constitution of Organizations (CCO) theory becomes predictive rather than merely descriptive. CCO scholars argue that organizations don’t just use communication; they are literally constituted by it. The organization exists only insofar as its members continuously create shared meaning through four flows: membership negotiation, self-structuring, activity coordination, and institutional positioning. AI can execute within all four flows. But it cannot originate them. It cannot decide what the organization should become, who belongs, or what the institution stands for in the world. Those remain irreducibly human acts, acts of judgment, values, and identity.

The Entropy Argument Information theory provides the cleanest frame for understanding why this matters.

Every organization is an open system fighting entropy. the natural tendency toward disorder, miscommunication, and decay. Management, at its core, is the continuous injection of negative entropy into the system. Clear communication reduces the number of possible chaotic states. A well-articulated strategy constrains the solution space, focusing energy rather than scattering it.

AI is an extraordinarily powerful entropy reduction tool. It can process, synthesize, and distribute information at scales no human can match. But entropy reduction requires more than processing power. It requires intent. Someone has to define what order looks like. Someone has to decide which of the infinite possible organized states is the right one for this company, this market, this moment. That’s the work that remains. And it’s the most demanding kind of work there is, not because it’s physically hard, but because it requires the rarest combination of domain knowledge, contextual judgment, and the courage to commit to a direction under uncertainty.

The Future Isn’t Jobless. It’s Leveraged. Russell argued for a four-hour workday. He imagined that technology would eventually reduce the necessary hours of labor, freeing humanity for leisure, culture, and personal growth. He was half right.

AI won’t eliminate work entirely, but it will ruthlessly bifurcate it. It will compress the time required for execution so dramatically that the baseline expectations of a productive day will skyrocket. Instead of spending eight hours producing one outcome, a knowledge worker will be expected to use AI amplification to produce five outcomes in that same timeframe. While the optimistic view promises hours freed up for judgment, strategy, and creative exploration, the economic reality is that the demand for execution will scale to consume the newly freed time. The leverage ratio gets multiplied by an order of magnitude, but as a result, the value of pure execution plummets toward zero. Strategy, taste, and relationship-building will indeed become the only defensible human skills left, but a company only needs a handful of true strategists. Consequently, the future isn’t about working less; it is about smaller, hyper-productive teams managing a relentless firehose of AI-generated output, where the baseline pressure to deliver has been permanently accelerated.

For founders, this means the Maker-Manager paradox dissolves, but it is replaced by an intense new pressure. You don’t have to choose between building and leading because AI gives you the bandwidth to do both. However, because you can do both at scale, the market will demand that you do. The speed of iteration becomes relentless, and the bottleneck is no longer how fast your team can build, but how fast you can make accurate decisions.

For individual contributors, the shift is a stark ultimatum. Since the sheer volume of execution is now handled by machines, the ICs who remain won’t be the ones who can write the most boilerplate or produce the fastest reports. They will be the rare few who act as high-level editors: judging quality, defining the right problems, and steering the AI. The jump from “executor” to “strategist” is no longer a mid-career milestone; it is a survival requirement.

For organizations, the broken chain, the fatal disconnect between leadership vision and ground-level execution, shrinks. When AI handles information synthesis and coordination overhead, the signal from the factory floor reaches the boardroom instantly. The cybernetic steering mechanism that Norbert Wiener described in the 1940s finally operates at the speed the theory always promised. But this also means that a company steering in the wrong direction will hit the wall faster than ever before.

The Question That Remains Russell’s paradox was never really about whether management creates value. It does. Economics, philosophy, and organizational theory all confirm it.

The real question, the one AI forces us to confront, is deeper: If machines can now both alter matter and tell others to do so at near-zero marginal cost, what is the uniquely human contribution to work?

I believe the answer is this: humans are the ones who decide what’s worth doing. Not what’s efficient, not what’s optimal, not what the training data suggests, but what actually matters. That’s a judgment no algorithm can make, because it requires something algorithms do not possess: a stake in the outcome.

The third kind of work isn’t altering matter. It isn’t telling others to alter matter. It’s deciding which matter is worth altering in the first place. And in a world where we can build almost anything instantly, deciding what matters has just become the hardest and only job left.