AI was supposed to give corporate employees time back. For many leadership teams, the business case for generative AI has sounded straightforward: reduce routine workload, free up capacity and redeploy time into higher-value work.
Drafting, summarizing and debugging could all be handled faster, with fewer people stuck in the weeds. Yet a more complicated reality is beginning to emerge. In many environments AI is not shrinking the volume of work as much as intensifying it. For chief executives, the distinction matters. Treating faster output as the same thing as sustainable productivity risks hard-wiring fatigue and weaker decision-making into the operating model.
The promise was always plausible. Generative AI does remove friction from knowledge work. Tasks that once required a long run-up can now be initiated in seconds. Early pilots across industries have shown clear speed gains, and separate research from the London School of Economics suggests the productivity upside is real. Its global survey of nearly 3,000 workers and 240 executives found that employees using AI save an average of 7.5 hours a week—roughly the equivalent of one working day and about £14,000 per employee annually in productivity value.
For organizations under constant pressure to do more with less, those numbers are understandably attractive.
But acceleration changes behavior. An eight-month observational study at a US technology company of roughly 200 employees found that workers with access to AI tools did not simply finish earlier and log off.
They worked faster, took on a broader range of tasks and extended work into more hours of the day, often without being asked. Researchers, who observed teams across engineering, product, design, research and operations and conducted more than 40 interviews, repeatedly saw the same pattern: when work became easier to push forward, people simply pushed more work through the system.
This is the quiet mechanism behind what might be called workload creep. When the cost of starting a task falls, the number of attempted tasks tends to rise. AI eliminates the blank page and reduces the effort needed to make incremental progress. In practice that means the natural pauses that once structured knowledge work begins to disappear. Work that might previously have waited until tomorrow can now be advanced immediately. Over time, what initially looks like a productivity dividend can become the new baseline expectation for speed and responsiveness.
For senior leaders the risk is subtle but material. Sustained intensity can contribute to cognitive strain, decision fatigue and ultimately weaker judgment quality.
Most AI governance programs are rightly focused on privacy, bias, intellectual property exposure and security. Yet the evidence suggests that human performance risk belongs on the same dashboard. If the organization moves faster but thinking quality deteriorates, the apparent efficiency gain may prove fragile.
The pressure often appears first at the functional level. In the observational study, teams reported that AI made it easier to initiate and advance work across disciplines, encouraging employees to absorb a wider mix of responsibilities. Engineers spent more time reviewing AI-assisted output; product and operations staff found it easier to keep projects moving outside traditional working windows. None of this required formal instruction from management. The tools themselves made “doing more” feel accessible.
This is why generative AI is increasingly behaving less like a pure efficiency tool and more like a capacity-expansion engine. When friction drops, organizations rarely bank the time savings; they redeploy them. The result is that productivity gains quickly reset expectations for output. What begins as a helpful boost can become an invisible ratchet.
For chief executives, the strategic question is therefore shifting. The early AI conversation centered on cost removal. The more relevant question now may be how much additional human load is being created alongside the gains.
The London School of Economics research points to another constraint: 68% of employees reported receiving no AI training in the previous 12 months. Many firms are still early in translating access into disciplined, sustainable use. Without deliberate operating norms, faster tools can simply amplify existing organizational habits.
None of this diminishes AI’s genuine potential. Used well, it can unlock meaningful productivity and help teams focus on higher-value work. But the organizations most likely to benefit over the next few years will be those that treat AI as an operating-model shift rather than a simple automation layer.
That means defining where speed genuinely creates value, building deliberate pauses into high-stakes decision processes and protecting recovery time with the same seriousness applied to system uptime.
In practice, leaders should be wary of measuring success purely through responsiveness metrics, which AI will naturally inflate. Decision quality, error rates and talent sustainability are becoming equally important indicators of whether the technology is being absorbed healthily. The companies that move fastest are not always the ones that create the most durable advantage.
AI is clearly accelerating the modern enterprise. The open question for CEOs is whether that acceleration is translating into true capacity—or merely increasing the human cost of keeping up.
