The more everyday tasks people delegate to AI agents, the greater the risk of losing skills they no longer use, Anthropic engineer Addy Osmani has warned. In his view, automating work should not mean giving up independent thinking: even if AI can perform tasks better than humans, people still remain responsible for the results.

Osmani is not calling for people to stop using AI tools, including in software development. He believes the problem arises when people stop checking algorithms’ output, understanding their decisions, and coming up with ideas on their own, TechRadar reports.

How “cognitive debt” builds up

Osmani admits that his own daily work now involves far fewer situations requiring critical thinking: many processes are now handled through AI agents. He says skills that people stop using gradually fade away—perhaps faster than one might expect.

The engineer describes this process as the gradual accumulation of “cognitive debt.” At first, a person carefully reads and checks an AI response. Then, because the previous result was good, they examine the next recommendation less thoroughly. Over time, they settle for skimming the overall structure of the response without looking into the details.

As a result, users may find themselves unable to explain a particular line of code or a technical decision—even with help from the very AI agent that produced the result.

The danger lies not in a single algorithmic error, but in a gradual change in human habits. The more decisions are made automatically, the fewer opportunities remain for independent practice.

Why responsibility cannot be handed over entirely to AI

One argument made by advocates of widespread automation is that there is no point in retaining skills for tasks that AI will eventually perform better than humans. Osmani disagrees with this approach.

Even if artificial intelligence has done a substantial share of the work, a person will still have to answer for the result. This is especially important in programming and other fields where mistakes can affect users, infrastructure, or business processes.

That is why, the engineer believes, human involvement in decision-making must be preserved. Delegating individual operations to AI is perfectly acceptable, but people must understand exactly what the system is doing, why it chose a particular option, and how well-founded its output is.

In other words, automation should not turn into blind trust in an algorithm.

How to retain skills in the age of AI agents

Osmani suggests several ways to avoid what he calls “intellectual surrender”:

· Practice core skills regularly. If AI writes code, analyzes information, or solves problems on a person’s behalf, it is important to do that work independently from time to time.

· Formulate your own ideas first. Before turning to AI, it helps to define the problem yourself, think through a possible solution, and only then compare it with the model’s suggestion.

· Examine the results carefully. Do not just look at the final answer; work to understand its content, structure, and details.

· Understand AI decisions. If an agent chooses a particular approach, the user should be able to explain why it was used and what its limitations are.

Osmani emphasizes that some skills may indeed gradually disappear as technology advances. The question is which ones people are willing to consciously hand over to machines and which they consider essential to retain.

The engineer’s central point is not that AI itself makes people less intelligent. The risk arises when the convenience of automation displaces the practice of independent thinking. And the more decisions people entrust to agents, the more important it becomes to retain the ability to evaluate their work and take responsibility for the consequences.