If your product is information, AI isn't a strategy. It's survival.

Insurance looks and feels like a product business. But it isn't, not really. Strip it back and all we really do is process information: price risk, underwrite it, support claims decisions, stay inside our regulatory permissions and ensure we treat customers fairly. There's no factory and no warehouse. The raw material is information and so is the finished product.

That's worth sitting with, because it means a company like ours is more exposed to AI-driven disruption than almost any other kind of business. When the cost of reading, writing, reasoning and acting on information collapses, it doesn't nibble at the edges of what we do. It attacks the core. In today's world, for a business built entirely out of information, being "AI-first" isn't just a clever strategy; it's a matter of survival.

We've spent the last six months trying to work out what this actually means in practice. Not AI-assisted, not AI-curious, but a business where agents do a meaningful share of the work by default and people are deployed where they're genuinely irreplaceable. Here's where we've landed so far, with a big caveat: this is our current best thinking, not a finished answer. We'll get plenty of it wrong. But we think it's better to think out loud and be corrected than pretend we've solved it all.

It's an operating-model change, not a tooling change

The easy version of AI-first is buying licences and encouraging people to use them. We don't think that's the real thing.

The deeper shift is about where knowledge lives. For most companies our size, the important knowledge sits in three places: people's heads, scattered documents that nobody updates, and a few prompts written by whoever cared enough. That works for humans. You can always ping someone on Slack and ask. But it's useless for agents, which can only act on what's actually written down.

So the first real move towards becoming AI-first isn't deploying agents. It's making knowledge machine-readable.

Capture knowledge, properly

We've started moving everything that matters into a single set of version-controlled Operating Manuals: commercial terms, customer journeys, contracts, regulatory permissions, trading decisions. Owned by named people. Readable by humans and agents alike.

It's slower and less glamorous than buying a tool, and it's the part most companies will skip. It's also the part that actually unlocks everything else. And what's interesting is what emerges once this is done.

When knowledge is written down, the shape of the team changes

When knowledge lives in people's heads, you need teams of specialists for every domain. When it lives in a manual an agent can read, you need fewer.

The experts we do need increasingly lead a domain rather than carry it. They set direction and judge quality, with agents doing much of the drafting and execution underneath them. Below that, we lean less on deep specialists and more on talented generalists who work alongside agents, picking up a manual and acting on it: one problem space this week, another the next.

We think this buys us four things:

• Speed, because we can move people to where the work is.
• Resilience, because no one person leaving takes a domain with them.
• Leverage, because a small expert layer plus generalists plus agents cover far more ground than the same headcount in the old way.
• Quality, because agents help drive up the standard of what we deliver and how we show up for partners and customers.

There's an uncomfortable edge to this, and we'd rather name it than dress it up. Our rough estimate is that a fully-agentic Open could run with materially fewer people than it does today. That isn't a cost-cutting target and it isn't a plan we're executing to reduce headcount. It's simply where the technology points. The reality is that we're growing, in partners, customers, products and services, and that growth brings new challenges and plenty of complexity for our teams to solve. But we'd rather be straight about the direction of travel than let it surprise people later.

The hard part: you can't lead this if you can't build and direct agents

This is the part we keep coming back to, and the part we're least sure how to get right.

The usual way companies drive a change like this is to bring in consultants and change managers to coach people into adoption. We don't think that works here. It's hard to lead a team into a way of working you're not living yourself. If someone leading a function can't sit down and build, direct and judge an agent, it's difficult for them to credibly take their team there.

So we're asking leaders at Open to be technicians. Not software engineers, that's a different bar. But people who can write a prompt, design an agent's job, point it at the right knowledge, orchestrate repeating tasks and groups of agents, and tell whether the output is any good.

What it asks of everyone

For people earlier in their careers, this changes how you grow. The old path ran through slowly accumulating deep domain expertise. Increasingly the path runs through being unusually good at directing agents and producing strong work across several domains, quickly.

Open, in a couple of years, will suit a different kind of person than it did a couple of years ago. That's a better conversation to have up front than to discover halfway through.

What doesn't change

We don't want to oversell this. Some things stay firmly human, and our investment in them is going up, not down.

We're a regulated business across three jurisdictions, and human oversight of CX, pricing, underwriting, complaints and compliance isn't being thinned; if anything it's being raised. The same goes for the hardest human work: vulnerable customers, escalated complaints, complex problem-solving with partners, the conversations where judgement and empathy matter more than throughput.

The rules are simple:

• Agents draft; humans decide.
• Agents capture; humans approve.
• No agent acts on a regulated decision without a named person signing it off.

Done well, writing down how we work makes our controls stronger, not weaker. Every decision recorded, every change explained, every owner named.

What actually decides who thrives

If there's one thing we want to leave you with, it's this: the thing that separates the people who thrive in all this won't be raw talent. It'll be posture.

The tools are cheap. The documentation is public. Nobody needs to be brilliant at this overnight. We certainly aren't. What matters is a willingness to pick up a new tool and have a go before anyone's taught you how, to be a bit uncomfortable, and to keep going. The people who lean in early will have a real advantage, and it's better to be upfront about that than let it surprise anyone.

That's the bet we're making, on the business and on ourselves.