From Checklists to Conversations: Agentic AI in Actuarial Work 

This event counts towards the two-hour GenAI CPD requirement under the revised PS 1.

Most actuaries have used AI as an advisor: it answers, and you still do the work. This session shows what changes when the AI becomes a worker — a chat model with tools, persistent state, and a loop that plans, acts, and corrects until the job is done. 

The talk answers the objection every actuary raises: "You want me to sign a number a language model produced?" The answer is no. The agent runs your pre-approved methodology as readable code; the program produces the number, and hard gates stop the cycle at every material step for human sign-off. Context engineering — the basis, the definitions, the documented method — is actuarial work, not IT work. 

The evidence is a live case study: eight agents in production at GrabInsure, covering pricing, reserving, monitoring, and regulatory analysis in Slack. The session closes with three honest lessons from the build — including how a wrong fact propagates at fleet speed — and the shift that matters: the actuary stops executing the procedure and starts specifying and reviewing it. Which is what the profession already trains for.