Imagine that a task that used to take four hours now takes one with artificial intelligence.
The immediate reaction is to say that you are four times more productive. But that calculation is incomplete. You finished one task faster; we still do not know whether the organization produced more value, whether the result improved, or what happened to the remaining three hours.
That leads to the important question:
Who gets to keep the time AI frees?
The worker may keep it as a less exhausting day. The company may capture it through greater output or lower costs. The customer may receive it through a lower price, better service, or faster delivery. Or it may disappear beneath a new layer of meetings, reviews, and content nobody needed.
Technology saves time. The organization decides, explicitly or implicitly, what happens to it.
Using AI, working faster, and creating value are not the same
It helps to separate three levels that are often confused:
- Adoption: a person or company uses AI tools.
- Task efficiency: a specific activity requires less time or effort.
- Real productivity: the same resources produce a better and valuable outcome.
The jump from the first level to the second can be immediate. Reaching the third requires redesigning processes, measuring quality, and deciding how to reuse the capacity that was released.
A presentation completed in twenty minutes instead of two hours is an efficiency gain. If nobody needed that presentation, it created no value. A hundred lines of code generated in seconds are not productivity either if they require hours of review or add unnecessary complexity.
Speed and value can coincide, but they are not synonyms.
The gap between the individual and the firm
This distinction already appears in the data.
McKinsey's 2026 global survey found that 37% of respondents attributed a positive EBIT contribution to AI in their organizations, while only about 6% qualified as AI high performers. Adoption has grown faster than broad financial impact.
Another study, based on surveys of nearly 6,000 executives in the United States, United Kingdom, Germany, and Australia, found that 89% saw no AI impact on their firm's productivity over the previous three years. That does not prove AI is useless. It shows that accelerating isolated tasks does not automatically become a visible organization-wide improvement.
The paradox is not that one person cannot save time. It is that thousands of small savings can become trapped inside a system that continues to operate in the same way.
When producing more becomes work inflation
AI dramatically lowers the cost of generating text, images, analysis, presentations, and code. That makes it possible to create more, but it also makes it easier to request more.
If a report once took a week, few reports were probably commissioned. If it now takes an hour, ten reports, five versions, and a meeting to compare them may appear.
The released time fills up again.
We can call this work inflation: the volume of deliverables rises without a proportional increase in value. More documents to read. More features to maintain. More messages to answer. More options to evaluate.
Productivity is not about filling every available minute with another task. Sometimes it means removing a task, simplifying a decision, or avoiding the production of something that changes no outcome.
From personal assistants to organizational systems
Personal copilots allow each worker to accelerate part of their activity. Organizational agents imply a deeper transformation: they can receive work, use tools, coordinate steps, and deliver results within a process.
But that transition adds questions a good prompt cannot answer:
- Who defines which work should be done?
- Who verifies quality?
- Which permissions does the agent receive?
- Who is accountable when it makes the wrong decision?
- How do we measure value instead of merely counting outputs?
Once AI enters the complete workflow, the advantage no longer depends on the model alone. It depends on architecture, data, controls, and above all, organizational design.
Saved time has no automatic owner
Let us return to those three hours.
If the company maintains the same salary and demands four times as many deliverables, it captures almost the entire benefit. If it shortens the workday without reducing income, part of the benefit reaches the worker. If it lowers prices or improves quality, the customer participates too. If the time is dedicated to learning, thinking, or experimenting, it can become future capacity.
None of these outcomes is built into the technology.
Distribution depends on bargaining power, contracts, competition, metrics, and management decisions. It can even change over time: at first, a worker uses AI to reduce their burden; later, that new speed becomes the minimum expectation.
The productivity question is therefore also political and economic. Asking how much time was saved is not enough. We need to ask who has the power to decide what happens to that time.
Price tells a story too
Suppose a service used to cost a thousand dollars because it required many hours of work. With AI, the same result can be obtained in a fraction of the time.
Should the price fall?
There is no universal answer. The client may be paying for hours, but they may also be paying for the outcome, accumulated knowledge, assumed risk, or accountability for the result. Even so, when marginal cost falls and competitors learn to use the same tools, some of the efficiency will often flow into the market.
This forces us to reconsider business models built around selling time. If the value was in executing a repeatable task, automation puts pressure on price. If it was in understanding the problem, making decisions, and taking responsibility for the outcome, AI may expand capacity without removing the value proposition.
How to tell whether productivity is real
Before declaring a gain, measure the whole system:
- Define the valuable outcome. Not “generate more content,” but improve decisions, sales, learning, or customer satisfaction.
- Establish a baseline. Record time, cost, quality, errors, and rework before introducing AI.
- Measure total cost. Include human review, corrections, tools, coordination, and failures.
- Watch the bottleneck. Accelerating one stage does not help if work waits at the next one.
- Choose the destination of saved time. Fewer hours, greater capacity, better quality, learning, or a lower price.
- Review the distribution. Identify who actually receives the benefit and who absorbs the new risks.
If we cannot say what improved and for whom, we are probably measuring activity rather than productivity.
Productivity is a design decision
AI can draft text, analyze data, or produce code much faster. In many contexts, that saving is real. But value does not appear through the automatic accumulation of released minutes.
Those minutes must be converted into better outcomes.
We can use them to produce more, rest, learn, serve a customer better, or solve problems that previously did not fit into the schedule. We can also waste them by generating more work for everyone.
The great debate is not only about how much AI can accelerate. It is about the kind of organization and life we want to build with that speed.
Because if a machine gives me four hours back, the decisive question is not whether I became faster.
It is who had the power to decide what happened to them.
References
- McKinsey & Company, The State of AI: Global Survey 2026.
- National Bureau of Economic Research, Firm Data on AI.
This piece grew out of a script about the difference between using AI, completing tasks faster, and producing real value. You can watch the original video on YouTube.