--- title: "The AI Work Paradox: The Real Constraint Is Absorption" type: newsletter date: 2026-08-21 source: linkedin summary: AI does not remove scarcity. It relocates it, from the people who produce work to the people who must absorb it. For decades, knowledge work was rationed by the cost of producing it. A market analysis took days. A prototype took weeks. A presentation required… newsletter: Technology Bytes draft: false --- **AI does not remove scarcity. It relocates it, from the people who produce work to the people who must absorb it.** For decades, knowledge work was rationed by the cost of producing it. A market analysis took days. A prototype took weeks. A presentation required research, synthesis, design, and revision. Exploring one more scenario carried a real price, and that price quietly governed how much work an organization chose to create. That governor is disappearing. Analysis can be generated in minutes, code in seconds, and synthesis on demand. Ten variations can now cost roughly what two used to. We call this productivity, and at the level of the individual task, it is. But organizations do not respond to cheaper production by producing the same amount and stopping earlier. They produce more. AI reduces the effort required for an individual task while increasing the total volume of work moving through the system. That is the **AI Work Paradox**. ### A Jevons-Like Effect, With a Different Bottleneck There is an obvious parallel with Jevons: when something becomes cheaper to produce or consume, demand often expands. AI is creating a similar effect in knowledge work. As the cost of producing analysis, code, content, and options falls, organizations ask for more of them. But the more interesting shift happens downstream. The constraint does not remain with production. It moves to review, prioritization, approval, and action. AI expands the supply of work much faster than human attention expands with it. The result is a relocation of scarcity: **from production capacity to absorption capacity.** ### The Bottleneck Moved Downstream Suppose a system produces twenty market scenarios instead of three. Someone still has to decide which ones matter. Suppose a coding agent generates five viable implementations. Someone still has to understand the architectures, weigh the trade-offs, test the behavior, and decide what reaches production. Suppose an assistant returns ten recommendations for improving a process. Someone still has to evaluate, sequence, and own them. The constraint has moved to the part of the system that did not accelerate at the same rate. Before AI, the scarce resource was often the ability to produce analysis. Increasingly, it is the ability to consume, evaluate, prioritize, and act on it. **Generation has scaled far faster than the human capacity for judgment.** Intelligence can now be produced faster than organizations can absorb it, turning what was once a production bottleneck into an attention bottleneck. ### Absorption Is Not Distributed Evenly This is where the paradox becomes an organizational problem rather than simply a productivity problem. The person who saves ten hours is often not the person who has to absorb the additional output. An analyst compresses three days of work into an afternoon and produces four times the material. But the reviewer, architect, editor, manager, or approving executive has not necessarily received the same productivity gain. Their throughput remains constrained by reading speed, working memory, meeting hours, and the finite number of decisions a person can make well in a day. The productivity gain is therefore captured locally, at the point of production, where it is visible and easy to measure. Some of the cost can move upward and outward toward the people who must review, evaluate, and decide. This helps explain why AI adoption can produce apparently contradictory experiences inside the same organization. The teams using it may feel dramatically more effective, while the people downstream of those teams experience a growing volume of material competing for their attention. Both can be true. An organization can record real productivity gains at multiple points and still create a new system-level constraint because the gains accrued to production while the pressure accumulated around judgment. ### Requests Became Cheap Before Outputs Did The expansion in work has another mechanism that is easy to miss. What collapsed first was not only the cost of producing work. It was the cost of **requesting** it. Asking a colleague for four additional variants once carried visible human friction. Someone had to spend the evening creating them. A manager knew that another request consumed someone else's time. That friction quietly governed behavior by forcing the requester to decide whether the additional work was worth asking for. Route the same request to an AI system and much of that friction disappears. “Give me five more options.” “Compare another ten competitors.” “Run three additional scenarios.” “Rewrite it for four audiences.” Each request is inexpensive to make and often reasonable on its own. Collectively, however, they create volumes of work that would never have survived the old economics of production. Some of what we thought was prioritization was actually being enforced by the cost of effort. When that cost falls, organizations need to replace implicit restraint with explicit discipline. ### Capacity Becomes the New Baseline The paradox compounds because productivity gains rarely remain exceptional for long. They become expectations. Once a report can be produced in a day instead of a week, a day becomes the expected turnaround. Once ten scenarios become feasible, three can begin to look insufficient. Once every customer interaction can be personalized, personalization stops being a premium capability and starts becoming normal. The productivity gain did not vanish. It was absorbed into the operating model and reissued as a requirement. AI increases capability. Greater capability increases capacity. Capacity resets expectations, and those expectations generate new demand. The organization becomes more capable, but the additional capability quickly changes what the organization considers normal. ### More Work Is the Point, When It Compounds More work is often exactly what we want. If AI makes it economically feasible to test twenty drug candidates instead of three, that additional activity may create enormous value. If an engineering team can evaluate ten architectures before committing to one, the resulting system may be better. If a company can serve millions of customers with the level of personalization once reserved for thousands, the additional work becomes part of the value proposition. The problem, therefore, is not volume itself but **undifferentiated volume**: additional work that creates more material to consume without producing proportionally more value. The question is not whether AI creates more work. It is whether the additional work compounds into better outcomes or simply into more material requiring someone's attention. That distinction becomes more important as the marginal cost of knowledge work continues to fall, because producing something cheaply is no longer evidence that it was worth producing. ### "Hours Saved" Measures Only Half the Story Most organizations currently measure AI productivity through hours saved. If a tool returns ten hours a week to an analyst, the gain is easy to report. But the number tells us very little unless we know what happened to the recovered capacity. ![](https://media.licdn.com/dms/image/v2/D4D12AQH7i3bElax7Ig/article-inline_image-shrink_1500_2232/B4DaAjBpYVKIAY-/0/1787294029581?e=1790812800&v=beta&t=I9edOJoY5tXPqIciQQZLTShwC-_vUFf5_qdNUp0342E) Hours saved reports the same productivity gain across all four cases, even though their organizational outcomes can be completely different. Ten hours removed from a process is not the same as ten hours used to generate thirty additional outputs. And neither is the same as ten hours reinvested in better judgment. The more useful question is therefore not simply how much time AI saved, but **what the organization converted that recovered capacity into.** That moves AI measurement from efficiency toward outcomes. ### Abundance Requires a Different Discipline For most of the digital era, organizations focused on increasing production. More data, more software, more analysis, more content, and more insights were generally treated as signs of greater capability. AI makes many forms of production abundant enough that maximizing output is no longer a sufficient management objective. When generating another analysis becomes almost effortless, deciding whether that analysis should exist becomes more important. When twenty alternatives can be produced on request, identifying the three that deserve attention may create more value than producing the twenty. When an agent can execute hundreds of actions, determining which actions matter becomes more important than maximizing how many occur. The scarce capability is shifting from generation toward judgment. And judgment does not necessarily scale simply because we give people more material to judge. In many cases, it improves when we give them fewer, better choices. That creates a different organizational discipline: **deliberately producing less than you are capable of producing, so that what you do produce can actually be absorbed.** ### The Advantage Moves to Absorption For decades, competitive advantage in knowledge work came partly from production capacity. Organizations that could analyze faster, build faster, and ship faster had an advantage. AI is rapidly making more of that production capacity broadly accessible. Almost every organization will soon be able to generate more analysis, code, content, recommendations, and alternatives than it can realistically use. That changes where differentiation can occur. The advantage may increasingly move downstream, toward the organization's ability to evaluate what it produces, discard what does not matter, concentrate attention on what does, and convert intelligence into action. **Production capacity can increasingly be purchased. Absorption capacity has to be built.** And that changes the AI productivity question. The issue is no longer only how much more an organization can produce with AI. It is **how much additional intelligence the organization can actually absorb and convert into value.** --- **If you found this article valuable, subscribe to Technology Bytes.** Each edition explores the intersection of Agentic AI, software architecture, cloud, sustainability, and enterprise innovation—sharing practical insights, original frameworks, and thought leadership to help leaders build AI systems that are not only intelligent, but also efficient, responsible, and scalable.