The most common reason for AI agent failure is not the intelligence of the model but its inability to access the correct, timely, and well-governed data.
Modern AI applications demand access to a vast and highly distributed data landscape, requiring connectivity to hundreds of sources, which makes traditional data consolidation impractical.
Real-time data is a non-negotiable requirement for many AI agents, with a significant portion of organizations stating that data even one minute old is too stale for their use cases.
A unified semantic layer, enabled by data virtualization, is the most effective way to solve AI's data challenges by providing a single point of access, governance, and security without costly data replication.
Enterprise investment priorities have rapidly shifted, with revamping data architecture for AI now being the top priority, surpassing direct investment in models or other parts of the AI stack.
Pre-AI Boom
Pan describes Denodo's origins in academic research on data management, with its core technology being data virtualization, before the recent AI-driven acceleration.
2025 (Cited Report)
Pan cites an IDC report from 2025 where 'revamping data architecture for AI' was a mid-level investment priority for organizations, ranked 5th or 6th.
Last Year
According to Pan, the rise of AI has served as a 'great accelerator' for Denodo's business, indicating a significant increase in demand and strategic focus.
Present Day
Pan states that Denodo's product roadmap is now 'completely dominated' by AI-related features, reflecting a full strategic pivot to address the AI market.
April 2026 (Cited Report)
Pan references a recent IDC report from April 2026 which shows 'revamping your data architecture for AI' has risen to become the top priority for AI investment, signifying a major shift in enterprise strategy.
▶The Data-Centric Failure of AI AgentsJul 2026
Alberto Pan argues that the primary reason AI agents fail in enterprises is not due to flawed models but because of inadequate data infrastructure. He cites survey data indicating that most failure modes are related to the inability to access the right data at the right time.
This perspective shifts the focus of AI investment and troubleshooting from model optimization to foundational data management, suggesting a significant market opportunity for platforms that solve data access, governance, and real-time delivery.
▶The Challenge of Extreme Data FragmentationJul 2026
Pan highlights that AI agents in large organizations need to access an average of 400 different data sources. This extreme distribution of data creates immense complexity for integration, security, and maintaining a consistent semantic understanding for the AI.
The scale of this fragmentation suggests that traditional data consolidation methods like ETL into a central warehouse may be too slow and rigid, creating a strong business case for agile, federated technologies like data virtualization.
▶The Imperative of Real-Time DataJul 2026
A recurring theme is the critical need for fresh, real-time data for AI agents. Pan claims that for 70% of organizations, data that is even one minute stale is considered invalid for some AI applications, and a similar percentage report a lack of real-time data as a significant problem.
This positions data latency as a key performance indicator for enterprise AI readiness, potentially devaluing data platforms that rely on periodic batch updates and increasing the value of solutions that can query data live at its source.
▶Unified Governance in a Distributed WorldJul 2026
Pan emphasizes the difficulty of enforcing consistent governance and security policies when AI agents access hundreds of sources. He presents the solution as a unified data access layer where rules, such as for GDPR, can be defined once and applied universally.
As AI usage and regulation grow, the ability to demonstrate and enforce consistent, auditable data governance will become a critical component of risk management, making centralized policy enforcement platforms essential for compliance.