Agents this, agents that. Sort your house out first. — Problem Hacker #26

I was chatting recently with a mate who’s spent most of his career in the insurance industry. Good operator. When I asked how far AI has really penetrated his business, he sort of shrugged. He reckoned it’ll take decades before AI tools are meaningfully embedded across insurance operations.

At first I sort of laughed. Decades??? But once I started digging into the data, I realised there’s truth in what he was saying.

A recent McKinsey survey found that 88% of companies report using AI in at least one business function. But only about a third have scaled AI across the enterprise in a way that meaningfully alters core workflows. BCG’s research finds that just 7% of insurance organisations have successfully implemented AI at enterprise scale.

So here’s what I think is going on:

Insurance isn’t slow to experiment. It’s slow to delegate.

Almost all insurance firms are using AI somewhere, but very few have fundamentally changed how decisions get made. That’s because delegation demands accountability infrastructure, not just models. You need auditability, identity, rollback, observability, decision boundaries. Most legacy systems don’t have those.

This mirrors a bigger pattern. At Davos 2026, PwC’s global chair said that over 50% of companies are getting little to no value from their AI investments, largely because foundational infrastructure and organisational readiness are missing.

Everyone’s Talking Autonomy. Few Are Ready For It

Retailers call it agentic commerce. Banks talk about agentic cores. Cloud vendors advertise autonomous execution layers. All basically the same thing: systems that plan, act, retry, and coordinate without human intervention.

Almost no one shipping these narratives is shipping governance at the same pace.

Agents don’t fail loudly, they fail politely. A small optimisation here, a tweak there, thousands of tiny decisions that no one monitors until suddenly no one understands why the system behaves the way it does.

The Hack: Map Where Decisions Already Happen Without Humans

Before you buy another “agentic AI suite,” do something more revealing:

Write down every existing decision in your business that happens without explicit human review.

Not just AI decisions. Any automated decision that runs itself.

Auto-approvals. Dynamic pricing. Threshold triggers. Retry logic. Escalations. Rules that “handle it unless something goes wrong.”

Those are agents in everything but name.

Now ask one absolute question for each:

If this decision quietly optimises the wrong thing for six months, who notices, and who’s accountable?

If the answer is “no one,” you don’t have an AI readiness gap. You have an accountability problem.

The Bit Everyone’s Avoiding

Insurance might feel like it’ll take decades before AI becomes core. But that’s not because the technology isn’t moving fast. It’s because no organisation scales delegation without fixing the plumbing first.

And fixing the plumbing (the identity, audit trails, governance bit) is harder than building the agent.

So the real advantage won’t go to the company with the smartest agent.

It’ll go to the one with the clearest answers to the questions that currently nobody’s willing to ask.

The Problem Hacker

Smart takes on strategy, growth, and uncomfortable truth.

From Mark Jefford // jefford