Past the proof of concept: the public sector's real AI adoption is underway

For several years, the public sector's relationship with AI has been a cautious one – pilots, proofs of concept and plenty of watching and waiting. But that tentative phase has passed. Experimentation and vague ambition is hardening into genuine, organisation-wide AI adoption, and the question has shifted from whether to use AI to how far and how fast to take it. Use of AI in the public sector is climbing steadily, and agentic AI – systems that carry out multi-step tasks with limited supervision – is beginning to climb with it. 

How far and how fast AI is deployed depends heavily on scale; a small local authority and a national health system face very different problems with very different resources, and the sector holds cautious and mature pockets side by side. But the direction of travel is consistent. Where earlier automation followed rigid scripts, agentic systems can orchestrate multi-step workflows, adapt to context and hand back to a human when judgement is required. AI technologies enable smaller teams of well-equipped people respond to requests as they arise rather than vanishing under the backlog. 

Moving beyond the low-hanging fruit requires a clear governance strategy 

That shift in ambition is where things get interesting. Much of the low-risk, high-reward work – the obvious low-hanging fruit – has been picked in early implementation. The value that remains sits in medium-risk, high-reward territory: finance, compliance, casework and other areas where the people costs are high and the potential returns higher still. But moving into that territory is a governance question before it is a technology one.  

The guiding principle is that high returns matter more than risk level, provided the governance is proportionate. Data protection here is not a one-off exercise but an ongoing discipline, built into the process rather than bolted on afterwards, and governance frameworks must be baked into operating processes, with employees educated and enabled accordingly.  

That foundation is far from a given: in a recent report from Artefact, surveys indicate that just 12% of executives believed their data infrastructure was genuinely ready for AI. This means the most impactful use cases must be approached iteratively, not in a single leap. Small, bounded proofs of concept, kept under real human oversight, let teams refine systems safely before scaling them. 

AI-enabled productivity and retention in public services 

This is where the value conversation is often misread. In the public sector, cost is still the language of value, and it always will be. But the returns that matter most flow from productivity. Administrative burden has been building for years as legislative change and compliance requirements accumulate, and highly specialised workers, whose skillsets are needed for essential services, increasingly spend their days on essential but low-value busywork. Relieving that burden lets them add more value per person and, just as importantly, protects them from burnout. Retention is the underrated prize here; reduced burnout means reduced turnover, preserving immense value for public sector organisations.  

For example, UK physicians have begun using AI tooling to allow them to be more present with patients and families during exams and consultations. One cardiologist is quoted saying that he sees a 3-5 minute time savings per consult, adding up to possibly an additional patient each day, helping to reduce wait times significantly across his 50-person department. 

AI-powered AI: the future of scaling 

There is a compounding effect worth naming: AI is now enabling AI. Teams increasingly use generative and agentic tools to build faster, more deterministic systems, then run those systems in production, reaching high-value outcomes far sooner than a traditional build would allow. This means high-value use cases are closer than they were even a year ago, and is why starting small and smart now matters so much: modest, well-governed pilots today are what make rapid, confident scaling possible tomorrow. 

The organisations pulling ahead tend to get three things right at once: 

• They set a centralised strategy for the tools being deployed, so effort is not duplicated across silos.  
• They implement right-size governance of inputs, outputs and usage, treating said governance efforts as continuous rather than a launch-day gate. 
• They invest in enablement, helping people understand and navigate both the technology and the governance around it.  

As reported by Artefact, 71% of public sector employees said they felt unprepared for AI. This means reskilling and job redesign is essential rather than optional. Enablement is also the best defence against shadow AI, which spreads not only through unsanctioned tool usage but also through personal, unstandardised processes involving approved technologies that never reach the shared workflow. The temptation is to treat oversight as a brake, but the opposite is true. Governance is a feature, not friction. Clear accountability, impact assessments and continuous monitoring are precisely what make autonomy safe, and safety is what makes scale possible.  

The public sector does not have to choose between ambition and responsibility. Handled well, agentic AI lets it have both. 

Contact us