Fivetran Reveals the Hidden Challenges of Agentic AI in 2026

Massive Investments Hampered by Deficient Infrastructure

Companies are investing considerable sums, reaching tens of millions of euros, to integrate agent-based artificial intelligence into their operations. However, despite these investments, a recent study highlights a technical reality often overlooked: data infrastructures are not up to the ambitions. By 2026, the maturity of data pipelines is identified as the determining factor between the success and failure of AI projects.

The transition from traditional generative AI to autonomous systems is profoundly changing the expectations of IT departments. They are no longer satisfied with merely generating text but are looking to delegate complete tasks to software agents. Yet, according to Fivetran, information infrastructures are failing to keep pace with the growing financial ambitions.

A Growing Gap Between Ambition and Reality

Fivetran's report paints a grim picture for decision-makers in the tech sector. While 60% of global organizations are heavily investing in these technologies, only 15% of them have solid foundations to support them. In France, this figure drops to 12%, revealing an even weaker preparedness. Companies are racing to implement cutting-edge technologies on fragile infrastructures.

George Fraser, CEO of Fivetran, emphasizes that the problem does not lie in the choice of computing models but in the systems' inability to provide reliable real-time information. According to him, building on fragile foundations can only lead to more frequent and severe failures.

The Hidden Challenges of Software Autonomy

For a software agent to make informed decisions, it must have complete visibility of its environment. However, industry professionals identify two major obstacles to this autonomy: the lack of traceability and the constraints of sovereignty and compliance. These barriers hinder large-scale deployments.

Nearly 40% of experts surveyed believe that these governance gaps turn pilot projects into dead ends. The quality of systems then becomes a crucial issue for economic survival, as uncontrolled agent-based AI can quickly go off the rails.

Openness as the Key to Performance

The most advanced companies in AI utilization have understood that the key to their success lies in openness. By adopting interoperable architectures, they avoid becoming dependent on a single vendor.

For data leaders, a system's ability to communicate with others is now an essential criterion. Gartner warns that more than half of initiatives could fail without adequate preparation. By 2026, the priority is no longer to acquire new tools but to consolidate the data flows that feed them.