Build for the end state. Pay for the next step.

Some of my friends at BCG recently set out three ways insurers can bring agentic AI into the core. Keep the existing core and put an agentic platform on top. Thin the core and move rules, workflow and decisions into the agent layer. Or build a unified agentic core, where deterministic processing and agents are designed as one system.

Their verdict: only the third solves the integration problem with existing systems and data platforms. I agree. Multiple systems that have to be stitched together will always cost more to change than one and will often deliver worse outcomes.

But I want to add a caveat: the right end state is rarely the right first step.

Few insurers can quickly launch a comprehensive core modernisation programme. Boards want results in the coming quarters, not in a few years. So the real question isn't which archetype is best. It's how you get from where you are to where you need to be without paying twice.

Why twice? Because most fast routes are dead ends. Put an agentic layer on a legacy core and you get quick wins, then a second orchestration layer, then a third. A middle office on its own is a switchboard if it cannot become the long-term core solution.

I suggest the following test for AI investments: Will it still be standing when you reach the end state? If not, it's an expense, not an investment.

That is how we built the path at Peak3 for our clients. Four steps, each one builds on the previous.

1. Agents first, on the core you have. Our Agent Platform runs pre-built agents for claims, underwriting and document processing on existing systems, with every action bounded, traced and evaluated at runtime.

2. Real core capability on top of the legacy. Graphene, our AI-native core platform, runs as a middle layer over your existing system of record. It adds deterministic core services of its own, with agentic interfaces. No migration is needed, so you get to market fast.

3. Migrate, if and when you choose. Your data moves across from the legacy system of record. The agents, the platform and the core services stay where they are.

4. Keep it from becoming the next legacy. Graphene Harness, the first global insurance AI-DLC, keeps the core changing at the speed of the business.

Moving fast doesn't mean spending everything now. It means spending only on what grows into the end state.