TLDR: Knowledge workers are increasingly delegating not only tasks but also thought processes to AI agents. What is marketed as a productivity gain is, in part, a creeping loss of competence. Those who no longer distinguish between delegating execution and delegating thinking are quietly and gradually eliminating their own value. The crucial question is not how to use agents, but when not to.
The relationship has been reversed
For a long time, the relationship between humans and tools was clearly defined. The tool waits. The human decides. A spreadsheet does the calculations when you feed it numbers. A hammer lies there until someone picks it up.
AI agents have reversed this logic. According to an analysis from the Tech for Impact Summit 2026, an agent today receives a goal, breaks it down into subtasks, executes them, evaluates the result, and iterates until the goal is achieved. It doesn’t wait. It works.
That's not a metaphor. It's a system architecture with consequences.
A PwC survey from 2026 shows that 79% of companies are already using agentic AI. The Time Magazine predicts that by 2050, most people will lead digital workforces larger than today’s corporations. This isn’t science fiction. It’s a question of organization and power that we’re currently addressing, whether we realize it or not.
The Actual Optimization Problem
We optimize for throughput. More output, faster, cheaper. That’s legitimate. But it’s not the only factor, and right now it’s overshadowing all the others.
„We underestimate how valuable friction is.“ Kartik Hosanagar, professor at the Wharton School
When AI takes over the initial stages of design—summarizing, researching, and structured thinking—humans lack productive resistance. Not as a romantic concept, but from a neurobiological perspective. ScienceInsights describes this in an analysis on cognitive AI delegation from November 2025: Those who avoid intellectual friction bypass the very process that generates understanding in the first place.
The consequences are measurable, and they come from an unexpected direction. Kartik Hosanagar, a professor at the Wharton School, documents a medical finding with structural transferability: Endoscopists who regularly used AI for polyp detection became less proficient at finding polyps as soon as the AI was turned off. Adenoma detection rates dropped from 28% to 22%. The delegation of tasks altered clinical behavior without the individuals involved even noticing.
This isn't just a one-off. It's a mechanism with a name.
A March 2026 arXiv study, „The Cognitive Divergence: AI Context Windows, Human Attention Decline, and the Delegation Feedback Loop,“ describes this phenomenon precisely: As AI capabilities increase, the cognitive threshold at which people begin to delegate decreases. They are also increasingly delegating tasks that would actually help maintain cognitive capacity. The resulting losses in ability are neurobiologically detectable and not spontaneously reversible.
What's the point of the prompt?
That is the strategic question that is rarely asked.
When I ask an agent to format a client report, I save time. When I ask him to structure my line of reasoning, I save myself the trouble of thinking. This isn't a gradual difference. It's a categorical one.
ScienceInsights clearly states the implication: Anyone who accepts AI-generated outputs, code, analyses, or texts without being able to understand the underlying logic loses the ability to verify the results.
„An agent manager who can no longer judge whether the output is good is not a manager. He is just a relay station.“
What remains is agreement without understanding. It looks like decision-making authority. It isn't.
Where Is Automation Headed?
In its *Future of Jobs Report 2025*, the World Economic Forum estimates that 39% of workers’ core competencies will change by 2030. AI and data analysis top the list of in-demand skills. According to the WEF, creative thinking, resilience, and judgment remain critical.
That sounds reassuring. It is a normative statement, not an empirical guarantee.
In 2025, the Stanford SALT Lab, in collaboration with the Stanford Digital Economy Lab, analyzed 1,500 workers across 104 occupations. Findings: Approximately 80% of U.S. knowledge workers will be affected by LLMs in at least 10% of their tasks, and 19% see more than half of their core responsibilities as potentially changing. The question is no longer whether, but how consciously we shape this transformation.
No one can guarantee that we will retain our critical human abilities unless we actively practice them.
Three observations that I consider relevant to the decision
Delegation is not a neutral act. Every task I delegate is a task I no longer practice myself. Employees grow through delegation. Agents do not. The difference is not trivial.
„The “Agent-Manager" role is valuable only if the ability to exercise sound judgment is maintained. This presupposes that the person delegating still understands the person to whom the task is delegated. Anyone who merely orchestrates without understanding is not irreplaceable. That person can be replaced by better orchestration.
We need to distinguish between delegation of execution and delegation of thinking. The former relieves the burden. The latter causes atrophy. Most claims about productivity do not make this distinction because it makes sales more difficult.
The real question
„What kind of thinking do we want to continue to be personally responsible for?“
This isn't a technical decision. It's one we make today, tacitly, through every prompt we send, through every summary we read, without the original.
The most important skill in the coming years isn't knowing how to use agents. It's knowing when not to.
Disclaimer
This article was compiled manually based on my own knowledge and supplemented by research using AI (Perplexity.ai and Claude), and then simplified using Deepl.com/write. The text is then reviewed and critically evaluated by two people of my choice. The image is AI-generated (Ideogram/Adobe Firefly). This article is purely educational and does not claim to be exhaustive. Please let me know if you notice any inaccuracies—thank you.
Questions about AI and digital business models? #fragRoger Would you like to learn more? I’d be happy to visit you—whether at your home, your company, your ERFA group, or your association—and help out with a workshop or a keynote presentation.
Feel free to reach out to me anytime for a no-obligation chat. So if there's anything I can do to help, please contact me at #fragRoger and subscribe to my official newsletter.
And that’s why I remain convinced: The future doesn’t lie in delegating at any cost, but in making a conscious decision about what we, as humans, will keep doing. Not because machines can’t do it, but because we have to be able to do it. So if you want to talk, and if you want to work with me, feel free to contact me at www.rogerbasler.ch
Sources (APA)
Hosanagar, K. (December 2025). AI is causing you to lose your skills. Here’s how to prevent it. Substack. https://hosanagar.substack.com/p/ai-is-deskilling-you-heres-how-to
PwC & Time Magazine. (January 2026). AI changed the world of work forever in 2025. Time. https://time.com/7342494/ai-changed-work-forever/
ScienceInsights. (November 2025). Is AI making us lazy or redefining what it means to work hard? https://scienceinsights.org/is-ai-making-us-lazy-or-redefining-effort/
Stanford SALT Lab & Stanford Digital Economy Lab. (2025). The Future of Work with AI Agents: Assessing the Potential for Automation and Augmentation Across the U.S. Workforce. https://futureofwork.saltlab.stanford.edu/
Tech for Impact Summit. (March 2026). The Future of Work: Human Talent, AI Agents, or a Post-Work Society? https://tech4impactsummit.com/blog/future-of-work-human-talent-ai-agents-2026/
World Economic Forum. (2025). Future of Jobs Report 2025. https://www.weforum.org/reports/the-future-of-jobs-report-2025/
Zheng, Y., et al. (March 2026). The Cognitive Divergence: AI Context Windows, Declining Human Attention, and the Delegation Feedback Loop. arXiv. https://arxiv.org/html/2603.26707v1
