The Futurescot AI Challenge 2025 invites public sector organisations across Scotland to explore how Artificial Intelligence (AI) can solve real-world problems. While many are already familiar with AI tools like ChatGPT and Microsoft Copilot, a new generation of AI – called AI agents or Agentic AI – offers a major leap forward in capability, with exciting potential for public services.
This article is an explainer of what AI agents are, why they are important for public sector and how to think about them in context of your organisation and hopefully your AI Challenge application.
What are AI agents and Agentic AI?
AI agents are systems that don’t just respond to one-off prompts, but instead act more like digital assistants. They can plan, make certain decisions and perform multi-step tasks. Agentic AI refers to more autonomous systems that can take multiple steps without constant human prompts.
Why are AI agents important for the public sector?
Because AI agents can take on more tasks end-to-end, it means they can save time and free up people in the public sector for more complex work and value-added activities. AI is particularly well-suited to the tasks outlined below such as processing, evaluation, generation and extraction because it can rapidly analyse large volumes of unstructured data.
An AI agent can be developed by taking such tasks as part of a workflow, chaining them together to form an end-to-end process. This is illustrated by the following examples whereby these types tasks can be chained together to form an AI agent.
Complaints handling
- Single use AI: An LLM such as ChatGPT is prompted to assess complaints data and summarise key themes.
- AI Agent: Automatically categorises incoming complaints, drafts responses learning from previous examples, routes complex cases to staff and analyses patterns.
- Human-in-the-loop: Responses are reviewed by staff for tone, accuracy, and fairness before any responses are provided and any sensitive or complex cases are flagged for human assessment.
Transcribing videos to text
- Single use AI: Upload videos and get a text transcription.
- AI Agent: Continuously monitors new video content, transcribes the audio, overlays it with a structured taxonomy and integrates the results into a document store to make the content searchable.
- Human-in-the-loop: Staff audit transcripts for accuracy and handle content where AI may misinterpret specialised language.
Claims Assessment
- Single use AI: An LLM such as ChatGPT is prompted to review a claim form and supporting documents, summarise key details, and compare them against eligibility thresholds.
- AI Agent: Automatically extracts key data from claim forms and evidence, checks for completeness and consistency, conducts an initial eligibility assessment using local policy rules, and generates a draft outcome or request for further information
- Human-in-the-loop: Staff review the AI’s assessment and recommendations to ensure compliance with legislation and fairness. Complex cases are flagged for manual review and final approval.
Key considerations for setting up AI agents
Setting up an AI agent requires planning across three key areas:
- Human oversight
Define the tasks the agent will perform and identify where humans should remain in control. Agents work best when they handle routine or repetitive tasks, while sensitive decisions stay with people with human-in-the-loop checkpoints. - Technical infrastructure
To run AI agents you need more than just access to an LLM. You will also require technology to provide orchestration around the agent (e.g. managing the chain of tasks and maintaining memory) and mediating a connection to internal systems or knowledge repositories. - Governance, security and ethics
Public sector organisations should consider what governance is needed to be put in place to ensure agents comply with data protection laws, are secure, and operate transparently. For scenarios where data sensitivity, privacy concerns or specific regulations necessitates that data remains off public cloud providers, AI agents can be deployed within your own network using open-source models rather than cloud-based models. Ethical frameworks should guide deployment to avoid unintended bias or misuse, with regular monitoring of how agents behave in real-world scenarios.
Designing your AI agent
The Futurescot AI Challenge is a chance to explore how AI agents could drive transformation across Scotland’s public sector. To support public sector organisations understand how to design an AI agent StormID have put together this guide to designing an AI agent. To enter the challenge visit here.
