Agentic AI for Actuaries: Practical Use Cases with Claude Code

Actuarial work is full of tasks that are technically routine but stubbornly manual: rebuilding a Solvency II balance sheet, regenerating year-end reporting and QRTs, reconciling a cashflow model against the numbers it is supposed to reproduce, refreshing an MI dashboard, or standing up a small tool to answer a one-off question from the business. Each is individually straightforward, yet together they consume a large share of an actuary’s time and are a recurring source of key-person and spreadsheet risk.

Agentic AI describes an emerging class of coding tools that do more than answer questions. Rather than returning a snippet to paste, an agentic tool reads your files, writes and runs code, inspects the output, and iterates towards a working result under your direction. Used well, it lets an actuary automate genuinely complex workflows – end-to-end reporting, model reconciliation, assumption-setting, validation harnesses, model and process documentation, and internal tools – without a dedicated software-engineering team. This session uses Claude Code, one example of this class of tools, to demonstrate the ideas throughout; the principles transfer to comparable agentic tools.

All examples are illustrative and deliberately generic – the focus is on the techniques and the working method, not on any organisation’s proprietary models, data or results. The emphasis throughout is practical and honest – what works, what does not, and how to keep agentic AI safe, validated and defensible in a regulated actuarial setting.

After this web session, participants will be able to:

• Explain what agentic AI coding tools such as Claude Code are, and how they differ from general-purpose chatbots.
• Identify actuarial workflows – reporting, reconciliation, modelling, assumption-setting and internal tooling – that are strong candidates for this kind of automation.
• Follow a repeatable pattern for scoping a task, directing the tool, and reviewing its work.
• Use the tool to generate and maintain model and process documentation that stays in step with the underlying code and satisfies Solvency II evidence requirements.
• Apply validation and governance techniques (fail-closed checks, reconciliation to a trusted source, golden-master tests, documentation) so that AI-assisted work is safe and defensible.
• Recognise the limitations and risks – model risk, key-person risk, professional and regulatory responsibility – and judge where agentic AI is and is not appropriate.

The approach is applied and example-led, grounded in real actuarial applications. Demonstrations are kept short and purposeful – each chosen to illustrate a specific Claude Code capability in a couple of minutes – and any longer-running task is shown through its prepared inputs and results rather than by watching the tool work, so the session keeps moving. Throughout there is candid discussion of what works, what does not, and the pitfalls to avoid. No prior software-development experience is required, though familiarity with actuarial modelling will help participants get the most from the session.