What is AI? Here's a breakdown
The term ‘AI’ covers a broad range of technologies. They are not interchangeable. Understanding the key types prevents your organisation from buying the wrong thing, applying it to the wrong problem, or measuring it against the wrong expectations.
AI TYPE
General-purpose AI (LLMs)
What it does
Large language models trained on vast public data. Can write, summarise, answer questions, translate, draft documents.
Real-world examples
ChatGPT, Google Gemini, Claude. Useful for drafting correspondence, summarising documents, answering policy queries.
Relevance to you
Good starting point for productivity. Not suitable for sensitive data without enterprise controls in place.
AI TYPE
Enterprise AI (secure LLMs)
What it does
The same capability as general-purpose AI but hosted within secure, governed environments. Your data stays within your control.
Real-world examples
Microsoft Copilot (M365), AWS Bedrock, Google Workspace AI. Deployed behind your existing security perimeter.
Relevance to you
The right choice for most public sector use cases. Connects to your existing platforms and meets data governance requirements.
AI TYPE
Traditional machine learning
What it does
Learns patterns from historical data to predict outcomes or flag anomalies. Requires structured data and specialist setup.
Real-world examples
Fraud detection, demand forecasting, risk scoring in benefits or planning. Well-established in HMRC, DWP, NHS.
Relevance to you
High-impact but requires data readiness and specialist skills. Better suited to stage two of your journey.
AI TYPE
AI tools for developers
What it does
Code generation, test automation, and development acceleration tools that help technical teams build faster.
Real-world examples
GitHub Copilot, Amazon CodeWhisperer. Accelerates software delivery and reduces developer toil.
Relevance to you
Immediate value for any in-house development team. Low risk and measurable productivity gain.
AI TYPE
Agentic AI
What it does
AI that takes sequences of actions autonomously, not just responding to a single prompt but completing multi-step tasks.
Real-world examples
Automated case triage, document processing pipelines, intelligent scheduling. Emerging in government pilots now.
Relevance to you
Significant potential but requires mature governance. Plan for this as a longer-term capability, not a first step.
The most important distinction for public sector leaders
General-purpose AI tools like ChatGPT are designed for consumers. Entering citizen data, case notes, or policy drafts into these tools creates serious data protection and security risks.
Enterprise AI products, by contrast, operate within your security boundary, log activity for audit, and do not use your data to train public models. Always verify which category a tool falls into before allowing staff to use it.