Accelerating responsible AI: Enabling adoption in city-level government

As part of the Government AI Campus, Apolitical is publishing a series of articles featuring leaders from the 2026 Government AI 100, a list recognising public servants around the world who are pioneering AI adoption, capacity building and regulation.

Apolitical spoke with Andrew Ngui, an MIT-trained innovation leader delivering a citywide Data and Responsible AI Strategy that turns governance into day-to-day operational change across departments in the City of Kansas City.

How have you designed governance frameworks that allow high-impact pilots to move forward in practice?

So governance, in my mind, isn't about restricting movement. It's about defining the field of play so we can compete against expectations set by industry players, like Amazon, Apple and Google. We're not competing with other governments; the competition is expectations set by the tech industry.

Governance provides structure, boundaries and rules. You can play faster and with more confidence because the rules are defined. It establishes a safe environment to meet elevated expectations without compromising public trust and accountability, creating guardrails that enable experimentation rather than blocking it.

One key aspect is being inclusive by default. Our default position is opt out, meaning you're pre-enrolled. Innovation cannot be a closed-door executive function. It must be bottom up with top-down sponsorship, allowing staff to co-create and shape workflows, moving from adoption to ownership.

We reframe gaps as opportunities. We select pilots based on organisational pain: does it solve a real problem or add complexity?

At the heart of this is durable change. We often focus on project management and leadership, but forget the people side. Change management should be the starting point, not an add-on. It is the primary constraint we need to address.

Can you tell us more about the constraints governments typically face when enabling AI pilots? How can these constraints be managed without undermining responsible use?

We have to live with one foot in the future and one foot in reality. The shiny widgets of AI are exciting, but if we don't ground them in our existing workforce, data and infrastructure, we're doing ourselves a disservice.

There's a gap between executive optimism and frontline reality. Frontline staff are the experts, yet tools are pushed to them without consultation, creating friction. We need them involved from the start.

Data is another constraint. Governments have massive datasets, but much of it is trapped in silos or non-digital formats. Data is not just an asset, it's digital infrastructure. We must begin with diagnostic assessment and standardisation.

Legacy systems are also a challenge. We prioritise interoperability and security from day one, asking not just how it works but how it fits into existing systems and processes.

Finally, lack of context kills motivation. We must close the communication loop and amplify the practitioner's voice. When staff understand the why, they transform from passive users into active champions.

What most often breaks down when governments try to move AI from pilot to broader rollout, and what helps to overcome those barriers?

The breakdown happens when we mistake a technological success for an organisational one. Scaling is not about proving that the widget works, but aligning the entire system so that the organisation can survive the transplant.

"The organisation has an immune system built for stability. If innovation feels invasive, it will be rejected. This is why we talk about minimally invasive innovation: achieving maximum results with minimum organisational trauma."

This is not an IT project; it's a people project. Staff may have decades of experience. If change is dictated, it becomes a threat. We must reframe it as people-centred, with frontline experts leading change through co-creation.

Psychological safety is foundational. Pilots feel safe because they're temporary. Scaling feels permanent, and resistance comes from fear. Position AI as empowerment, automating tedious work so staff can focus on meaningful tasks.

Scaling also fails when systems don't align. If people and processes aren't ready, even the best technology fails.

From your experience working across agencies and sectors, what's been essential to making collaboration in AI effective despite different mandates, risk appetites and levels of maturity?

Collaboration travels at the speed of trust. The baseline is that doing nothing is not an option. That reframes the conversation from whether to act to how to act safely. Different agencies have different needs, but we face the same curve of obsolescence. It's about creating urgency without forcing risk-averse organisations to become startups.

Trust is our primary API. People don't block projects because they hate technology; they block them because they fear being left to hold the bag if it fails. We must show we are on their side.

Effective collaboration requires a protective partnership. Stakeholders need to know we understand their constraints and are solving for their safety. Empathy is critical. We must understand their context, their experience and their risks.

Initiatives like the Gov AI Coalition show how this works. Regardless of maturity level, agencies can share use cases, templates and lessons. There are many entry points, allowing organisations at any stage to engage and learn.

We must also ensure shared understanding. When I say apple, the colour in your mind should match mine. Without that clarity, collaboration breaks down.

What skills and capabilities are most important for public servants when working with AI?

Beyond tools like Copilot or ChatGPT, I feel we're in the age of creativity. The barrier to creativity has never been lower. It requires the ability to dream and imagine use cases we haven't yet considered.

We also need better models of thinking. It's not just about skills, it's about how people think, how they make sense of problems and how they design solutions. If common sense is not common, we cannot reach a shared understanding. Critical thinking, including second- and third-order effects, is essential. The key is defining use cases. The better you are at defining them, the more effectively you can leverage AI.

How do you see the balance between experimentation, governance and scale evolving in government AI in the next two to three years?

Everything needs to be human-centred. Not human-friendly, but human-centred, where everything revolves around the human.

"We need guardrails and safety nets that define boundaries. When boundaries are clear, people can explore and experiment with confidence."

We must rebuild curiosity. People have been conditioned not to question. We need to create space for curiosity and ensure their voices matter.

Teams should start asking more questions. Question-storming helps define what truly matters. At the heart of it, we are designing behaviour change. It must be intentional, not left to chance. Technology is black-and-white; people are not. We must design for that variability.

What excites you most about governments using AI, and what concerns you?

The opportunities are there, but it depends on whether we create systems that empower staff to do meaningful work. Culture shift is both an opportunity and a challenge.

A key concern is cognitive offloading and overreliance on AI. Using AI without checking outputs or assuming it understands context carries significant risk. We've seen this happen across sectors and countries.

We must also recognise the need for lifelong learning. It's not AI that will take your job, but someone who learns how to use it effectively. Learning must be continuous, supported from the top down and bottom up. Providing open access to training and tools enables staff to engage and adapt.