The insurance industry is experimenting widely with agentic AI. But turning a successful PoC into a secure, scalable, enterprise-ready solution is another challenge entirely.
In our latest webinar, Microsoft’s Jamie Taylor and Dufrain’s Fraser Black broke down how insurers can build production-grade multi-agent systems using Microsoft’s AI ecosystem and gave the live demo if it in action.
Here’s the practical guide summary, straight from the session.
1. Microsoft now offers a complete stack for agentic AI development
Jamie outlined Microsoft’s tiered agent-building ecosystem:
- Microsoft 365 Copilot – the “UI for AI”, ideal for simple agents and task automation
- Copilot Studio – low-code agent building for teams who want to go further
- Azure AI Foundry plus Agent Framework – for complex, multi-step, production-ready agents built with code-first engineering
This layered approach means insurers can start anywhere and scale up.
2. The new Agent Framework brings two critical capabilities
Jamie shared how the Agent Framework merges Microsoft’s previous tools (Semantic Kernel + AutoGen) into a single open-source framework that enables:
- LLM-driven agents that call tools, APIs and MCP servers
- Workflows that orchestrate multiple agents into complex multi-step processes with human-in-the-loop control where required
Foundational building blocks accelerate architecture design and reduce time-to-value.
3. Fraser’s live demo: underwriting automation with multi-agent orchestration
Fraser walked through a real underwriting example using:
- Copilot as the orchestration layer
- Azure AI Foundry for agent definition
- Azure Machine Learning, Storage, SQL and Search as supporting services
- A multi-agent workflow:
-
- Submission scoring agent
- Submission analysis agent
- Product suggestion agent
A submission was scored, analysed and matched to relevant insurance products all in under 30 seconds, without the underwriter lifting a finger.
The underwriter remains in control, but the workflow accelerates:
- Submission triage
- Information completeness checking
- Risk scoring
- Product recommendations
- Broker communications
This is exactly the type of high-impact, low-friction use case where insurers can unlock value immediately.
Watch the full demo here: Insurance Gets Agentic Webinar – Dufrain
4. The architecture is already available to insurers today
Fraser emphasised an important point: insurers don’t need net-new technology.
Everything demonstrated, storage ingestion, machine learning pipelines, AI Foundry agents, Copilot interfaces already exists within Azure ecosystems many insurers already use.
The real challenge isn’t tooling. It’s designing the right architecture, picking the right use case, and ensuring the right skills are in place.
5. Why build, not buy?
In response to audience questions, both Fraser and Jamie agreed:
- Off-the-shelf tools can be useful for simple tasks
- But true differentiation comes from building on your own data
- Internal capability-building gives insurers long-term competitive advantage
- Many off-the-shelf solutions are simply wrappers around the same Azure tech shown in the demo
Building creates IP, relevance, adaptability and advantage.
Agentic AI isn’t futuristic, it’s ready now
The session proved that insurers can rapidly move from ideation to deployment if they start with the right architecture and the right use case.
If you found this summary interesting, watch the full 30 minute webinar and demo here.
We’ll be sharing more information about the webinars in our upcoming blogs (blog one is here), follow us on LinkedIn for more, here.