Lloyds Launches Agentic AI Research with University of Glasgow

Lloyds Banking Group and the University of Glasgow launched a 4-year applied research programme to evaluate how agentic AI driven by LLMs can assist software and data engineers within a large financial-services environment. The collaboration will embed academic researchers with Lloyds' engineering teams, run structured experiments, and fund a PhD studentship, a Masters by Research, and a post-doctoral position to study real-world adoption, measurable outcomes, and governance questions. Lloyds serves 28 million customers and intends to use the findings to scale agentic approaches across its engineering organisation.

The project focuses on integrating semi-autonomous agentic tools, often implemented as orchestration layers over LLMs, into day-to-day engineering tasks. Researchers will design empirical software engineering experiments that track both qualitative and quantitative signals. Key implementation and measurement topics include:

• •Funded research roles: PhD, Masters by Research, post-doctoral associate embedded with engineering teams
• •Evaluation metrics: output quality, development velocity, defect rates, and task completion time
• •Methodology: empirical software engineering techniques such as data mining of repositories, A/B style experiments, controlled task assignments, and observational studies
• •Tooling and platforms: pilot integrations with developer-assist systems (team-reported use of tools like GitHub Copilot), internal knowledge assistants, and agentic orchestration layers that coordinate multi-step workflows

The study will run recurring cycles where engineering teams pair with agentic counterparts to solve assigned tasks; results will be tracked quarterly to observe learning curves and aggregation effects.

Large organisations are accelerating adoption of generative tools for coding and operations, but robust, large-scale evidence on agentic AI impact in enterprise software engineering is scarce. Lloyds has already reported material value from GenAI deployments and substantial internal usage of developer-assist tools. This collaboration gives academic researchers rare access to production-like settings, enabling evaluation of how LLMs and agentic orchestration perform when embedded inside established engineering processes, compliance regimes, and risk controls.

This is significant for three reasons. First, the research shifts the conversation from proof-of-concept pilots to longitudinal, organisation-level evidence that can inform rollout, retraining, and governance. Second, the partnership explicitly ties technology evaluation to workforce upskilling and process change, addressing one of the main practical barriers to scaling agentic systems. Third, results will likely influence vendor choices, internal platform design, and regulatory discussions about auditability and control for semi-autonomous developer assistants.

The study's measurement design and published findings. Expect to see detailed metrics on defect density, developer productivity, and how team roles evolve. Watch for reproducible protocols and data-sharing agreements that let others benchmark results. Also monitor governance artifacts the partners produce, such as recommended safety checks, audit logs, and human-in-the-loop policies, since those will be the path to enterprise adoption at scale.