Automate repetitive admin tasks with AI-powered agents
Read the full detail of action 2.1.4 on GOV.UK. (opens in a new tab)
Year 1 progress
In Year 1, we have supported staff to explore agent capabilities within our enterprise generative AI tools. This has led to a growing number of agent-based solutions being developed across the department, particularly to automate low-risk tasks.
The department has established a growing registry of nearly 5,000 agents across the organisation. We have visibility of these agents within the Microsoft 365 ecosystem, alongside governance processes to track their development and usage. This provides a clear picture of how teams are using AI to support their work.
Year 2 priorities
In Year 2, we will shift our focus to targeting discrete, repetitive work tasks where more sophisticated agent deployments can deliver meaningful impact. This includes use cases such as form population, cross-system data retrieval, and workflow automation. To support this, we have put in place:
- A governance framework to ensure safe, auditable use of agents.
- A central approach to tracking and monitoring agents across MOJ.
- Defined processes for publishing and managing agents, including mandatory steps for higher-risk use cases.
- Support for scaling more advanced use cases, including those requiring forward deployed engineering capability to integrate across systems.
We will continue to take a test-and-learn approach, starting with well-defined use cases in controlled environments (using synthetic data where appropriate), while building the foundations needed to safely scale successful agents across the justice system.
Case studies (2)
RubyPilot
Ruby is an AI-enabled digital “red box”, bringing together ministerial commissions, submissions and other core private office workflows in a single platform rather than across multiple spreadsheets and trackers.
By making work easier to manage and track, it reduces administrative burden and improves coordination across teams, allowing private offices to focus more on supporting ministers and senior officials.
Early pilots demonstrated significant benefits, with teams reporting less manual tracking, better oversight of work and more time spent on supporting decision-making. Additional AI capabilities, including automated summarisation and a chatbot assistant, have been introduced to further streamline day-to-day tasks.
Ruby is now in production across most private office teams in MOJ and has become a core part of how ministerial and senior official offices manage their work. After a successful 1-month pilot in the Department for Work and Pensions, Ruby is also being explored by several other government departments, signalling growing interest and wider adoption across Whitehall.
OPG Investigations AssistantATRS — record published on the Algorithmic Transparency Recording Standard (opens in a new tab)Pilot
The Office of the Public Guardian (OPG) plays a vital role in protecting individuals who may lack the mental capacity to make decisions about their health and finances and have a lasting or enduring power of attorney (LPA/EPA) in place. When concerns are raised about the potential misuse of an LPA/EPA, OPG has the power to investigate concerns raised.
The continued growth in LPA demand has also seen a proportional increase in the number of concerns being raised with OPG and increasing numbers of investigations. Delays in investigations progressing can risk slowing the point of intervention, impacting the ability to intervene for the benefit of vulnerable individuals at the earliest opportunity.
To address this, the Justice AI Unit is working with OPG to design and test AI-enabled workflows that assist investigators with elements of the investigations process, reducing manual tasks. Investigators remain fully responsible for reviewing outputs and making decisions using their professional judgement, with AI supporting investigators with their work. Working in collaboration with Microsoft Forward-Deployed Engineering, the team has identified parts of the end-to-end investigations process where AI and automation can add value. This includes supporting time-intensive, repetitive manual tasks such as preparation for financial analysis, as well as practical tools to streamline day-to-day investigative work, improving quality and consistency.
A financial categorisation workflow is already in a controlled live pilot. The workflow processes transaction data within financial data sent to OPG, and automatically categorises transactions into structured groupings, providing investigators with a clearer starting point for financial review. During the pilot, performance is being monitored against key criteria including accuracy of categorisation, time saved compared to manual methods, and the consistency and usability of outputs in real investigations.
This work aims to reduce manual effort, improve consistency, and give investigators more time to focus on professional judgement and case-specific decision-making.
