Imagining AI agents running the company and negotiating with each other is more of a Planet of the Apes scenario than a credible digital transformation, far from the real uses of AI.
Will AI agents, our digital clones, more intelligent, tireless and reliable than us, replace us in the future and “do business” with each other? While we can debate the societal ethics underlying this vision, this anthropomorphic vision is above all as unfounded as the famous “Planet of the Apes” franchise! Certainly, certain science fiction paradigms have become reality. Autonomous combat drones developed by military powers are, nothing more, nothing less, than Terminators and Asimov’s three laws governing robotic ethics are strangely relevant. However, this vision of the “Agent Business” is marred by irrelevant anthropomorphism. On the one hand, reproducing human roles and organization is clearly not the most effective approach. On the other hand, AI agents cannot replace humans in making critical decisions or defining strategic directions.
An anthropomorphic “Agent Business” model circulates in the minds, conversations and offerings of certain AI SaaS. In this definition of “AI agents”, they would embody (and replace) personas in the company, interact with each other (within their firm but also outside its walls), to solve problems, make decisions, negotiate and even transact. A purchasing agent in company A would negotiate with a sales agent in company B, before the legal agents of the two companies validate the contract for the financial managing agent to approve it so that the supplier accounting agent triggers the payment of a deposit. The agents, respectively from the downstream and upstream logistics of the two companies, would then ensure the correct delivery and reception of the goods in the systems, all this against the backdrop of autonomously driving trucks and automated warehouses. Welcome to the Matrix!
The company would be duplicated by a digital twin, a kind of video game copying historical human roles and responsibilities, and playing out the same processes in the virtual world. Humans would be sent back to their beloved studies… or, we don’t really know what activities. This would be the end of work and the advent of universal income, which Singularity has been planning for decades. We find the concept of the article “we are almost certainly living in a computer simulation” by Nick Bostrom, published in 2003 but which received a wide response at the end of this decade: as soon as computing power allows it, humans will duplicate the real world, and observe the tribulations of their digital avatars on virtual planets. These simulations of the real world would multiply rapidly, almost infinitely. This article concluded, in all seriousness, that, if we accepted that computing power would become virtually infinite in a few years, then this scenario of the multiplication of digital simulations of the Earth would occur. Therefore, nothing allowed us today to think that we were evolving in the only embodied reality rather than in one of these numerous simulations. The whole Valley, in turmoil, was now convinced of already living in the Matrix. is not very realistic.
Why is this model fallacious?
On the one hand, the human organization of companies, that is to say the current division of tasks and roles, is very far from constituting the optimal solution for solving a problem. Why bring together agents who embody personas and let them debate to align with the marketing and sales plan for a product launch, develop the annual budget or establish the factory operating plan as part of the S&OP process, when a “super application”, unique for each of the subjects taken as an example in the list above, and supported by various technologies, including AI, can address its subjects? This “super application by subject” can issue recommendations (or even take actions in information systems, depending on the degree of autonomy given to it), integrating all the dimensions of each problem.
On the other hand, these “super applications” (for which it would be appropriate to reserve the name “agents”, rather than anthropomorphic doubles of human roles in a digital twin of the organization chart) support a transactional workflow, carry out complex optimization or analyze the dimensions of a decision and act under the control of humans. Systems serve humans, not the other way around.
AI agents (in this definition of super applications) can replace humans in tedious tasks (eg. the reconciliation of supplier invoices), carry out complex optimization (eg. the dispatch of intervention teams to customer sites), increase a human activity with added value (eg. teaching), or even provide “insight” for delicate decision-making involving risk assessment (eg a pricing strategy, the optimization of transport in periods of uncertainty, the allocation of rare stocks… even the military strategy, to cite a recently controversial subject). But humans remain in control.
The best example is undoubtedly the anecdote of the Netflix marketing teams who questioned the AI on the acquisition of the rights to the Korean Kpop series Demon Hunters: catastrophic score, negative recommendations from the AI. At the insistence of local teams, Netflix ended up making a restricted acquisition of the rights. We know the rest: a global hit! The human audience will never be predictable… As good as AI engines are, for value-added activities (including “busy” interactions such as negotiations that involve building trust between two organizations) and for delicate decisions in an uncertain environment (which involves taking a risk), human instinct and experience remain irreplaceable!
