Abstract
This paper reviews how definitional ambiguity in Artificial Intelligence, from rule-based automation to GenAI, LLMs, and agentic systems, complicates adoption in the UK public sector. It argues for an indicative, human-likeness-oriented characterisation of AI to stabilise terminology and support governance decisions. The review synthesises evidence that fragmented strategies, siloed organisational structures, and limited readiness impede implementation, while concerns about bias, black-box decision-making, and opaque evaluation systems undermine transparency and trust. Drawing on debates that distinguish structural from normative governance, the paper positions strategic planning as essential but insufficient under rapid technological change. It evaluates canvas-based approaches, including business, decision, and governance model canvases, and proposes that an agile governance paradigm is needed to align strategy, operational delivery, and stakeholder accountability. The discussion concludes that meaningful AI adoption requires flexible, iterative governance that preserves public legitimacy while adapting to accelerating innovation.