Agent-based AI: Data and Integration, a Crucial Challenge

The Crucial Importance of Data for Agent-Based AI

For an agent-based AI project to truly succeed, it is imperative to rely on quality data, rigorous governance, and seamless integration. These elements form the essential foundations without which even the most advanced systems may fail to achieve their objectives.

As companies gradually embrace generative AI tools, their interest increasingly shifts towards agent-based AI. These systems are specifically designed to plan, analyze, and execute actions within predefined contexts, thus offering a new dimension of autonomy and efficiency.

A Growing Market but Full of Pitfalls

Investments in agent-based AI are following this upward trend. According to a McKinsey study, the market could grow from $5 to $7 billion in 2024 to an impressive $199 billion by 2034. However, this enthusiasm should not overshadow the inherent complexities of implementing these technologies.

Many companies struggle to transform their pilot projects into fully operational large-scale solutions. The statistics are revealing: Gartner predicts that more than 40% of agent-based AI projects will be abandoned by the end of 2027.

At the same time, a survey conducted by Qlik indicates that while 97% of companies have allocated a budget for agent-based AI, only 18% have taken the step towards full deployment. The potential is clearly identified, but implementation remains a challenge, particularly in complex business environments.

Often Inadequate Data Infrastructures

Unlike generative AI tools that assist employees with specific tasks, AI agents interact directly with the company's workflows. They can, for example, detect financial anomalies, adjust supply chains, or autonomously trigger operational processes.

This autonomy requires flawless precision, as errors can have significant consequences. A recommendation generated by AI can be verified before implementation, but an automated flow requires complete trust in the accuracy of the data and the underlying logic.

It is often at this stage that companies realize their data infrastructures are not ready to support this level of responsibility. The main obstacle lies in the lack of data maturity. Agents require a consistent and reliable view of information at the enterprise level.

Many organizations still operate with fragmented data, duplicated sources, or poorly defined ownership. Unstructured data further complicates the situation. Internal documents, emails, and knowledge bases often contain crucial context. However, their owner is rarely identified, making it difficult to verify their accuracy or relevance.

When agents interact with operational systems, these weaknesses quickly become apparent. They can even undermine trust in their decisions. Therefore, data consolidation is an essential prerequisite for successful deployment.

The Challenges of Governance and Integration

When AI makes decisions, several crucial questions arise: Who is responsible for the data? Who validates the actions? When is human intervention necessary? Clarifying these responsibilities allows teams to trust the system while minimizing the risks of errors.

Regulation also plays a structuring role in this context. In Europe, AI legislation aims to frame transparency, accountability, and risk management from the design of systems. Although these rules may seem restrictive, they provide a useful framework for more controlled adoption.

The proliferation of AI tools presents another challenge. In many companies, each team uses its own solutions, whether they are analytical platforms, internal systems, or external assistants. For agents to function effectively, they must have access to reliable data and be able to interact with these various tools.

Towards Harmonious Interconnection

The importance of common standards then becomes evident. Technologies like the Model Context Protocol (MCP) enable the connection of assistants to enterprise systems while respecting access and governance rules. Rather than multiplying custom integrations, organizations can offer consistent interfaces usable by different tools.

This approach simplifies deployment and lays the groundwork for large-scale adoption, allowing companies to fully leverage the benefits of agent-based AI.