Federal agencies are being asked to make meaningful artificial intelligence (AI) progress while managing competing priorities around security, compliance, modernization, budgets, and workforce capacity. Success depends not on chasing the next headline use case, but on strengthening the operational, technical, and governance foundations that allow agencies to apply AI in practical, mission-aligned ways.
MeriTalk recently sat down with Nina D’Amato, chief technology strategist at Lenovo and a former chief information officer (CIO) of the County of Santa Clara in the Bay Area, and a retired Marine Corps officer. She shared her perspective on how agencies can build the foundation for meaningful AI progress across the organization, the technology environment, and the broader partner ecosystem.
MeriTalk: You’ve used the term “VUCA” to describe what federal CIOs are facing. What makes this environment so difficult?
D’Amato: When I say VUCA, I’m talking about the reality federal IT leaders are operating in every day: volatile, uncertain, complex, and ambiguous. The technology is shifting fast, the cyber threat landscape is shifting fast, the policy environment is shifting fast, and none of that aligns neatly with long federal budget cycles or procurement timelines. CIOs are constantly reprioritizing in an environment that does not move at the speed of technology.
An AI pilot may be visible to the executives and exciting, but if the underlying operating model, data environment, governance, and infrastructure are not ready, it is very hard to move from proofs of concept to something the organization can scale and sustain.
MeriTalk: In that VUCA environment, when agencies look at moving AI from pilots to broader operation, what should they prioritize first — infrastructure modernization, security, data readiness, managed services — and why?
D’Amato: I always come back to capabilities. How will this investment enhance our ability to increase value to our customers or be more capable in service of our mission? Identifying the current state and the desired future state — how can these new technologies get us there?
The most important place to start is the operating model: how the organization delivers value. Taking the time to understand your purpose, the culture, the leadership team, how you are structured, the talent and skills, and the orientation of the culture toward technology. Are we ready to leverage new technologies to enhance our value to our customers and to ourselves? This is not easy work, but the sooner you dive into this, the more of an impact the organization will make.
To make a lasting impact immediately, I would evaluate the endpoint strategy. Agencies must start to transition to AI endpoints and devices that are ready for modern, AI-enabled applications. The AI PC can handle AI tools, features, or enhancements embedded in enterprise software. Otherwise, the device starts getting sluggish and overall productivity slows down or stops in some cases. Organizations can, and should, start this immediately to ensure their workforce stays productive. Another piece of this endpoint evaluation is whether to transition to device as a service in your operating budget versus the traditional capital buy model. Device as a service gives you more predictability and control in your budget and now there are many wonderful features embedded in this service — such as asset recovery and defining the refresh cycles to fit your budget.
Next, I would move to capabilities that the customers or organization needs and evaluate the infrastructure. Start with capabilities and back into the infrastructure. To run AI applications, you will need to invest in the right infrastructure with a chip set that can handle the demands of the application. Let the application and the capabilities gained drive your infrastructure decisions.
CIOs have conveyed to me that they are moving everything to the cloud to minimize complexity in the environment, but this is very costly and should be considered carefully. Are you sure you can afford this when cloud costs are already so high? Cloud makes sense for training and experimentation, but steady-state inference demands an on-premises infrastructure investment. Further, you maintain more control over your organization’s data.
MeriTalk: IDC research commissioned by Lenovo found that the government’s top AI priorities for 2026 are improving decision-making and regulatory compliance. Where are the most immediate opportunities to apply AI to those goals?
D’Amato: Better decision-making with the right data at the right time is something every agency wants. But to get there, agencies need to be honest about what is standing in the way. In many cases, it is not a lack of ambition. It is the burden of compliance, the weight of over-customized applications and broader legacy technology debt, and infrastructure that was not built for the AI era.
One of the most practical opportunities is using AI to help reduce the burden of the legacy environment. In this new era, we need to think about how the 10% of the IT budget for transformation can move the 80% of the operations budget that supports older technology.
A strong example is using AI to modernize legacy applications — translating or refactoring mainframe code such as COBOL into modern architectures and languages like Java or Python. This can reduce dependencies on legacy systems, expose data through APIs (application programming interfaces), and enable integration. Once you modernize, you can move easily into predictive analytics, machine learning, and AI-driven capabilities that are difficult to implement in tightly coupled legacy environments.
Another opportunity is using AI to analyze and synthesize data across environments so leaders can act faster and with more context. In California, one police organization is using a fantastic application called SkyAI to bring together data from sources such as 911, license plate readers, and case data so responders can make decisions faster. The application trains only on the police force’s data and gets smarter, faster, and able to draw conclusions and give recommendations. It can easily pull in everything associated with a case going back decades — from obvious associations to not-at-all obvious associations. Just as important, it uses a hub-and-spoke model that lets adjacent jurisdictions access the data. Federal agencies may have different missions, but siloed data is a very familiar challenge that AI-powered applications can solve.
MeriTalk: Many government leaders cited data quality and governance as priority investment areas. What are the first signs an agency’s data is not ready for AI, and what are the first steps to improve it without boiling the ocean?
D’Amato: Agencies need to stop aiming for perfect data, because that is not a realistic goal. The real question is whether the data is good enough, governed appropriately, and usable in the context of the mission and the risk profile.
The first sign the data is not ready is when it is too siloed to be coordinated across systems or functions. Another is when nobody is confident about how it is categorized, labeled, or governed. In federal environments especially, you have to know exactly what kind of data you are dealing with and what rules apply to it.
The first steps are practical. Use new technologies to help classify, categorize, and label data more effectively. Understand the governing rules around regulated data. And do not let the quest for perfect data prevent you from moving forward with a focused use case. You can improve the data while also putting it to work.
MeriTalk: The same IDC research highlights gaps in AI governance, risk, and compliance. What are the minimum guardrails agencies should put in place to scale responsibly and prevent shadow AI?
D’Amato: AI is a team sport. The CIO should not be doing this alone — and frankly cannot. Most of an organization’s problems are not technical in nature — they are business problems that the CIO can solve. Agencies need a governing group that brings together the CIO, legal, security, privacy, and business leads to evaluate proposed AI efforts, map them to a developed risk register or a risk scorecard, and decide where and how to move forward.
That group needs to ask the basic questions: What is the business problem we are solving? What data is involved? What capabilities will be gained? Is it increased productivity for our workforce? Is it better security? Where is the human in the loop? How does this improve the experience of the workforce or the public?
That is also how you reduce shadow AI: Make the process visible, give people across the organization a way to bring forward ideas, and create a mechanism to evaluate those ideas to move the organization forward while balancing risk.
MeriTalk: How does Lenovo help agencies get started and take concrete next steps over three to six months?
D’Amato: Lenovo helps agencies get started through an AI Readiness Assessment and AI Fast Start, which is designed to be consultative. Agencies bring forward their challenges, and we help them prioritize, put structure around governance, and, most importantly, work side by side with the change management effort. Then we call on our pool of partner applications to build and deploy a functional AI pilot with your data in 90 days. The engagement can deepen depending on what the organization wants to accomplish. Lenovo can support that work across devices, infrastructure, services, and partners.
I would also emphasize two things that are becoming more important. One is partnerships. The technology is moving too fast for agencies to work in isolation, and strong ecosystems matter. The other is as-a-service. In a volatile budget and pricing environment, as-a-service models can give agencies more control over operating budgets and help them avoid some of the staffing strain that comes with standing up and supporting everything on their own. That lets internal teams stay focused on higher-value mission work.
Nina will be speaking on the Agencies are reshaping tech delivery – a new playbook is taking hold panel at Shift Happens.