AI agents have arrived – but are they assisting us or replacing us?

If you have been following the recent statements from the executives at companies like Salesforce and Meta, then you will know that 2025 is being considered as the “year of the agent” – the AI enabled services that are expected to work more independently than the current chatbots. A new model from China, Manus, made headlines recently after people claimed it could not only buy real estate and even create video games.

But as the training of the AI systems gradually reduces the human instruction frequency, a very particular issue comes up: what are the roles of people as the cognitive tasks are increasingly taken over by machines?

The Argument for the Enhancement Model of AI

LinkedIn co-founder and Microsoft board member Reid Hoffman has also given a rather optimistic view of the use of AI and its capacity to create employment. In his book Superagency, he claims that with the growth of AI, people will not become obsolete but will become more powerful. This view is in contrast to the increasing fear that the use of AI is eroding our critical thinking and independence. Hoffman compares the situation with the past, saying that the same concerns were expressed when the printing press was invented – it was predicted that this invention would undermine the memory of people. “Things we lose – like the ability to memorize and recite epic poems – are a small price to pay for the capabilities that AI brings to the table,” he said. Hoffman highlights the positive impact of AI, sharing examples of how chatbots have helped people in distress and even helped to prevent suicides. In the same way, people on X have begun to reply to @perplexity in their posts and ask AI to fact-check the claims – sometimes even the site’s owner. These examples demonstrate the potential of how AI can improve our ability to detect the truth.

Over Dependency as a Risk

However, progress is not always positive. Although technological innovations have brought many advantages to people’s life, they also have some side effects such as health issues due to sedentary lifestyle. The effectiveness of AI will therefore depend on the way it is used and managed. For instance, in China, AI is applied in the state control and this has raised issues on civil rights. According to Hoffman, if democratic countries continue to control the development of AI, then it can become a benefit. Nevertheless, the history tells that even in the West, the technological advancements – including the social media filters – have had unforeseen and at times negative consequences. A new challenge is the so-called ‘deference to the machine’, which was identified by Kester Brewin in the UK. The generation of young professionals, who are entering the labor market, is too dependent on the AI tools, which makes them question their own ideas and rely on the algorithmic solutions. This could become a bigger problem as more and more AI agents start to perform routine tasks of a white collar kind. While some people imagine that these systems will be used by humans to manage them as opposed to replacing them, research indicates that people’s performance deteriorates when they are in charge of automation. Some examples include; in aviation and healthcare, over reliance on technology has resulted to major mistakes.

A Defensive Strategy with Respect to AI.

These risks are avoided by pilots who work with a “defensive” attitude. They are trained to be ready for the worst at any time – for instance, keeping their hands on the stick during landing in case they have to attempt a go around. Luis Prato, an insurance executive and a private pilot, uses this approach in his business and advises that risk should be re-evaluated constantly. In a similar way, the governance of AI could use a protective strategy. We should not make the conversational platforms to rely on the AI generated responses completely but we should design the systems that will acknowledge the weaknesses. It should not try to create content where it is not sure what it is saying, this is to reduce what is called Hallucinations. Hoffman stressed that the notion of agency is a cultural construction. In this case, Hoffman and his critics are in agreement that the right way to relate with AI is not through acceptance but through skepticism. The purpose of AI is not to replace humans but to enhance their performance; therefore, the role of AI in autonomy is as much a function of the relationship that we have with the technology as it is of the technology itself.

—(c) 2025 Bloomberg LP.