Why agencies must strengthen the data foundation before scaling AI
Federal agencies are entering fiscal year 2027 with a familiar challenge: moving forward with artificial intelligence projects while updating old systems, improving cybersecurity and working within tight budgets.
Jeremy Wilson, chief technology officer for the public sector at EDB, said data connects all these priorities. Agencies are finding that successful AI and cybersecurity projects depend less on the technology added on top and more on the quality of the data foundation underneath, he said.
“AI is a data problem before it’s a model problem,” Wilson said. “Agencies are recognizing that AI progress is determined less by access to models and more by the readiness of their data.”
Wilson said many agencies still struggle with data spread across different systems and environments, making it hard to use on a large scale. This challenge has led government technology leaders to ask where AI tasks should run and how AI systems can best work with the data needed for government missions.
“Our whole thesis is: Bring the AI to the data, not the other way around,” he said. “If your mission data is in Postgres, and increasingly it is, you can stand up vector search, retrieval-augmented generation and AI agents on that same governed platform.”
Critical need #1: Keeping control of data
As agencies expand their use of AI, concerns about security, privacy and control over government data are leading to greater interest in AI systems that keep data under the agency's control.
Wilson said agencies cannot afford to lose track of where sensitive data is stored or who can access it.
“Agencies can’t send citizen data, law enforcement data or classified-adjacent data to whatever inference endpoint happens to be cheapest,” he said. “Sovereignty means you control where the data sits, where the models run and who touches both.”
To address these concerns, EDB has adopted an approach that brings AI models to the data. This lets agencies run AI models where their mission data already lives instead of sending the data to outside services. Wilson said this design allows agencies to build AI systems that can take actions within controlled environments and reduces their need to rely on large, publicly available language model services.
He added that keeping control of AI and data is increasingly important for business reasons as well as security.
“Our Sovereignty Matters research found organizations that are deeply committed to sovereignty realize up to five times the return on investment on AI,” Wilson said.
Critical need #2: Protecting and managing data in the AI era
Federal discussions about zero-trust security often focus on user identities, devices and networks. Wilson said agencies can overlook the importance of protecting the data itself.
“Data is one of the five pillars in both the Cybersecurity and Infrastructure Security Agency’s maturity model and the Defense Department’s reference architecture. Yet, it’s often the least mature,” he said. “You can have perfect identity and network segmentation, but if the database itself isn’t enforcing least privilege and protecting data at rest, you haven’t finished the job.”
Wilson said protecting data requires strong rules about who can access it, encryption, tools to track and review activity, and compliance with federal security requirements. These safeguards have become more important as agencies pay closer attention to software supply chain security and compliance.
“Post Executive Order 14028, agencies are accountable for their software supply chain,” he said, pointing to the growing importance of software bills of materials, which list the components used in software, and tools that check compliance automatically.
Critical need #3: Planning for long-term modernization
Efforts to update old technology systems are also leading agencies to reconsider their long-standing investments in database software.
Wilson said many federal organizations still rely on proprietary technology platforms that raise costs and make modernization harder. He pointed to growing interest in open-source technology as agencies look for more flexibility and lower running costs.
“We built deep Oracle compatibility into EDB Postgres Advanced Server precisely so agencies don’t have to rewrite every application to escape proprietary licensing,” he said.
Wilson pointed to the Forest Service as an example of an agency that used modernization as a chance to switch to a lower-cost, better-performing platform. He said the result was an open technology foundation that can be moved between environments and did not simply replace one form of dependence on a vendor with another.
Critical need #4: Building a foundation for fiscal year 2027 priorities
Agencies are under pressure to show clear results from their AI investments. Wilson said federal technology leaders should focus less on individual tools and more on the underlying data systems that support them.
Wilson said starting with a strong data foundation will determine how well agencies turn technology spending into results for their missions over the next several years.
“Every initiative on the federal agenda — from AI adoption and zero trust to customer experience and legacy retirement — lands on the data layer eventually,” he said. “If that layer is open, secure, sovereign and AI-ready, everything above it gets easier and cheaper.”
Bringing AI to the data: A different approach to agency AI
Traditional AI systems often require agencies to move data to a separate platform for analysis and model training. According to Jeremy Wilson of EDB, this can create extra challenges for managing data, protecting it and meeting security rules.
An approach that brings AI models to the data reverses this process:
Step 1: Keep mission data where it is already stored.
Step 2: Apply rules for data management, security and access where the data is stored.
Step 3: Run AI tools, including vector search, retrieval-augmented generation and AI agents, within that controlled environment.
Step 4: Provide AI-generated insights while keeping control over sensitive government data.
Potential benefits for agencies:
- Less data movement
- Greater control over data
- Simpler data management and oversight
- Lower security risks
- Faster deployment of AI tools