LinkedIn
AI THAT THINKS AHEAD
One direct example of that shift is the state’s push to modernize its enterprise resource planning (ERP) system. In an interview, Darby described a recent request for information (RFI) for a new ERP system that doesn’t just ask vendors to pile on AI features, but to bring “native AI architecture” into the core of the platform. The RFI makes clear that the state expects this architecture to be “extensible, secure and auditable.”
The reason is simple: Darby doesn’t just want automation that follows instructions, but systems that can spot problems before humans must step in. She described the kind of solution that could detect anomalies in payroll, flag procurement bottlenecks, surface compliance risks or even identify potential fraud.
At the same time, Darby emphasized that autonomy has limits in government. Any system with that kind of power, she said, “has to be bounded and auditable.” The RFI reinforces this by requiring explainability features, auditable chain-of-thought tracking, and controls that allow the state to restrict or mask AI reasoning for compliance purposes.
However, as this type of technology implementation moves forward, the CIO stressed that the focus is not on removing people, but on changing the nature of their work. With agentic AI, staff are being moved away from acting like “data entry clerks” toward roles that rely on human judgment, oversight and decision-making — letting AI handle repetitive or analytical tasks while humans focus on interpretation, strategy and governance.
The state’s broader adoption of AI shows it is no longer treated as a technology for standalone use cases. As Darby puts it, AI is being “absorbed into the normal workflows,” with integration occurring earlier in both decision-making and technology planning. She described 2025 as a year spent building guardrails and shared language around AI, while 2026 is shaping up as the year to put those frameworks into practice.
Darby also discussed the practical forces shaping state AI strategies, including funding realities and competing priorities. Even as AI becomes more visible, she framed it as something that has to connect to measurable value. As in other states, the goal here isn’t to deploy AI for the sake of appearances, but to apply it in ways that improve outcomes.
MODERNIZING MORE THAN MACHINES
For the state’s broader modernization work outside of AI in the last year, Darby pointed to the successful rollout of Tennessee’s unemployment insurance system, which went live in late 2025, alongside ongoing modernization efforts in offender management and child welfare systems slated for completion in 2026. The offender management system is on track for an October go-live, while the child welfare system is expected to launch this summer, according to the tech leader.
But any successful modernization, she said, brings change management challenges that cannot be solved with software alone.
“The hardest part is never technology. It’s usually untangling decades of embedded assumptions and policy procedure workflows — the ‘this is the way we’ve done it,’” Darby said. “As we’re doing any new system, we need to be able to explain that this is the requirement or the functional need and this is the outcome we’re looking for.”
SECURITY AND CLOUD BUILT IN, NOT BOLTED ON
As Tennessee expands its use of AI-enabled and cloud-based systems, Darby said cybersecurity is not being treated as just a final approval step; she described their security team as integral to every project. As she put it, “Cybersecurity should be at the table in the beginning — not a gate at the end.”
That approach has translated into concrete changes in how systems are evaluated and built. The CIO pointed to the use of preapproved design patterns for AI and cloud deployments, along with risk tiers tailored to specific technologies instead of one-size-fits-all controls.
A similar emphasis on flexibility and guardrails extends to the state’s cloud strategy as well. While Tennessee remains cloud-first, Darby described the approach as intentionally pragmatic rather than rigid. Some workloads — including low-latency systems, specialized environments or certain health-care technologies — continue to “make more sense on-premises due to operational, regulatory or risk considerations.”
Rather than centering decisions solely on where applications run, according to Darby, the state is prioritizing consistent outcomes across environments. The emphasis is on standardized security controls, consistent identity and access management, portability and resilience, avoiding vendor lock-in, and keeping total cost of ownership in check.
“Cloud-first remains the goal across the state,” she said, “but cloud-only isn’t realistic and we don’t think it’s necessary.”
‘EXPONENTIALLY,’ NOT ‘INCREMENTALLY’
The central message Darby kept returning to is that small, incremental changes won’t be enough to drive the next phase of government technology in Tennessee. She described states as being at an “inflection point,” and said the ones that move forward will be those that “think exponentially and not incrementally.”
In her view, that transition is not only about adopting new AI tools or modernization efforts. It is about changing what the government can do. Darby described a future where government becomes “predictive versus reactive,” where policy impact can be measured faster, and where new service models can take shape instead of simply updating old ones.
