show index hide index
At the crossroads of theory and practice, agentic AI emerged in 2025 as a truly turbulent revolutionary force among emerging technologies. This pivotal year marked a decisive turning point, transforming models previously confined to laboratories into integrated infrastructures, visible and tangible in everyday life. As innovations continue to pour in, particularly with the emergence of communication protocols and integration tools, 2026 is shaping up to be a crucial year. What new advances do they promise us in the field of AI agents ? What social and technical challenges will we have to overcome to guarantee a future that is both bright and responsible?
For over sixty years, artificial intelligence was primarily confined to laboratories and scientific publications. However, the year 2025 marked a decisive turning point with the emergence of AI agents in the public sphere, transforming the way we interact with technology. As we approach 2026, new innovations are on the horizon, promising to further this revolution.
The Transition of AI Agents in 2025 Before 2025, AI agents were merely stirring in the shadows, reserved for researchers and development professionals. In January 2025, with the launch of DeepSeek-R1, a Chinese open-source model, the landscape changed dramatically. This model enabled open testing and independent applications, sparking growing interest in the alternatives offered by the laboratories of the time.
Throughout the year, the American scene saw the emergence of more advanced models, while China enriched an already popular catalog. The trend was clear: Chinese models attracted more users than their American counterparts, sparking a real craze around the use of AI agents in everyday products.
Key innovations of 2025
Among other major breakthroughs, Google introduced the Agent2Agent protocol, which simplifies communication between agents, while Anthropic launched the MCP, focusing on the efficiency of agents’ use of tools. This collaboration between the two giants was a step toward open standards, as both protocols were submitted to the Linux Foundation at the end of the year. This has led to the emergence of agent ships—navigators integrating these advancements—that handle everyday tasks such as booking and comparing services, demonstrating that AI can now be an assistant in our lives. The outlook for 2026 While 2025 laid the groundwork, 2026 is shaping up to be the year of thestructuring of AI agents. We are seeing adaptations in evaluation methods; traditional benchmarks are no longer sufficient. Researchers
will need to venture beyond simple results to examine the processes by which agents execute tasks.
This shift aims to guarantee the
reliability of AI systems and strengthen user trust. In parallel, the governance and organization of workflows are becoming crucial issues. Companies will need to integrate these agents effectively to minimize the potential risks associated with their use. Challenges to overcome
Despite the enthusiasm, socio-technical challenges stand in the way of this progress. Concerns range from the environmental impact of data centers, which place a heavy strain on electrical grids, to the issue of automation and potential job losses. Fears surrounding over-regulation in some countries, compared to the more relaxed approach in the United States, also raise concerns. Experts warn of the dangers related to the security of connected systems. The interconnection of models and the accumulation of agents can increase the risk of malicious injections, hidden on the web and potentially harmful. A systemic approach for the future To address the complex challenges of agentic AI, rigorous engineering, thoughtful design, and clear documentation are essential. By placing agents within a socio-technical system framework, it is possible to build a secure and innovative AI ecosystem capable of meeting current and future challenges. Expectations for 2026 are high. Innovations in this field could redefine the standards of artificial intelligence, offering a multitude of new opportunities as we explore the capabilities and implications of AI agents in our daily lives.
