From ChatGPT to AI agents: the intelligent automation revolution that affects all industries

‍A new quantum leap in Artificial Intelligence — OpenAI and Nvidia are convinced: AI agents are a new major step in generative AI and will have a very strong impact on businesses as early as 2025.

An AI agent is a standalone software that aims to achieve the objective that is given to it. The AI agent writes its own action plan and interacts with the other systems. It is THE solution not to be missed in 2025 to design Support Agent, Knowledge Agent, Talent Sourcing Agent and other Social Media Agents.

This article does the stitch on the recent progress of AI agents and how this is a major topic for businesses in 2025.

The concept of autonomous software becomes concrete thanks to Artificial Intelligence

Since the 1980s, multi-agent agents and systems have been conceptualized, they are able to intelligently distribute tasks between several entities and to bring about collective intelligence. But the Agent developments are complex, and they are quickly limited in their ability to interact with other systems.

However the recent advances in advanced language models (LLM) made it possible to give autonomous agents a New boom because generative artificial intelligence makes it possible to give them great autonomy in formulation of action plans, decision-making, but also interaction with other systems.

In 2023, the first experiments impressed the tech community, such as AutoGPT and BabyAGI, which paved the way. These tools, based on GPT-4 and a Python environment, demonstrated that an LLM could act as an agent, if given the ability to call APIs and remember.

While the result was promising, it was still imperfect: hallucinations, fragile reasoning, and limitations in task performance due to the complexity of the APIs of other systems.

In 2024, companies like Relevance AI or CrewAI have started offering comprehensive solutions for creating AI agents, while frameworks like LangChain or Google Cloud Vertex AI Agent Builder have facilitated the integration of AI agents in corporate contexts. More general public, OpenAI's formulation of the concept of Function Calling paved the way for GPTs, autonomous assistants capable of interacting with other systems.

And it was in the fall of 2024 thatOpenAI gave a completely different dynamic to AI agents presenting them as the next major evolution of ChatGPT. So Sam Altman, during the T-Mobile Capital Markets Day, presented the five levels of evolution of Artificial Intelligence, announcing level 1 (chatbot) as the past, and the transition from level 2 (reasoner) to level 3 (agent) as imminent. This announcement from the OpenAI leader had a catalytic effect on the ecosystem, pushing startups and businesses to reconsider their approach to Artificial Intelligence.

‍

‍

This announcement was followed at the beginning of 2025 by the promotion of agents by Jensan Huang, CEO of NVIDIA, during a keynote speech, going so far as to present it as a multitrillion-dollar opportunity.

From now on, the objective is no longer simply to assist, but toact proactively and independently, sometimes even if it means roughly reformulating what was presented as Generative AI at the beginning of the year into an AI Agent, without really changing the underlying technology.

‍
Structuring and industrialization

In any case, in 2025, the landscape of AI agents will be structured both thanks to the use of sophisticated tools for designing AI agent chains, and also under the impetus of OpenAI, which announces a complete evolution of its ChatGPT solution to an agent solution (even if this is still quite conceptual). Anthropic, the publisher of Claude.ai, a direct competitor to ChatGPT, published a very comprehensive guide to the design of AI agents at the beginning of 2025: Increased LLMs integrating memory and contextual tools, orchestrated workflows to chain and parallelize tasks, and finally truly autonomous agents able to plan and execute complex actions.

OpenAI As for him, prepare in the coming days for the launch of”Operator“, an AI agent integrated into macOS, designed to interact directly with the user environment. An advance that already raises questions about the security And the trustworthiness of these technologies.

A future to be built

Indeed, despite this enthusiasm mixed with real progress and gross hype, major challenges remain. Beyond the major topics of reliability, governance and security of Artificial Intelligence, the industrial design ofAI agent requires a rethinking of development paradigms and technical architectures, which requires skills both in the design and in the strategic orientation of projects.

So to go beyond pure concept and obtain real answers, only pragmatic and solution-oriented approaches will make it possible to overcome hype and create truly impacting uses.

‍
At 321, we support decision-makers from large undertakings :

to define their innovation strategy on AI agents and quantify the impacts on their top or bottom line
to identify and integrate AI agents developed by early stage companies
to build tailor-made AI agents

So contact us !