Insurance has always been the business of predicting the cost of future events. But in a world where some people are predicting the end of humanity within a decade, has pricing risk ever been harder?

AI disruption has reached insurance faster than most people expected. In the past 6 months we have seen insurers race to launch ChatGPT apps, an AI native MGA that’s also an insurer and a re-insurer, and an asset manager and a cafe franchise raise hundreds of millions and reach a £5BN valuation. More recently CEOs have been confronted with the decision to block or not block Muse (Autonomous Digital Agent).

As AI changes the world we live in it is reshaping what risks people and corporations want to protect against and what data is available to price these risks, assuming there is any relevant data at all.

So how do you “win” in insurance now we operate in a reality that keeps rewriting itself? Time for a “4 box” to bring clarity to the madness. At Founders Factory we are forecasting future scenarios by mapping ‘speed of adoption’ vs ‘endurance of trust' of AI.

Using this framework I have identified ten new realities that will emerge over the next decade and have been asking myself, what will I want to protect in these realities and what startup sized opportunities might emerge in response? 

I don’t like to think about this stuff in isolation for too long, so my takes are inspired by   insights from Founders Factory portfolio founders building at the bleeding edge of insurance including Marrow, Mama Insurance, Meshed.

Near future: the next three years

In the near term, AI agents will become the default medium through which people access the online world, much as the browser did in the internet age. The opportunity I am interested in is less about which AI agent wins (or who builds them) and more about who takes responsibility and liability when these agents start buying insurance.

1. Assistants become the front door

AI assistant usage grows from roughly 15% of the population to more than half within the decade. People will ask an assistant to explain their policy, compare quotes and, increasingly, complete the purchase.

The insurance angle: distribution shifts underneath the industry. Research, comparison and buying conversations move away from direct customer contact and into AI interfaces. The price-comparison site, which reshaped UK personal lines in the 2000s, gives way to agent-to-agent negotiation. Insurers will need products, pricing and data that machines can read and evaluate, not just people.

2. More agents than humans, with no referee 

Agents won't map one-to-one onto people. One person may run dozens of agents, each agent may run its own sub-agents, and different people’s agents will join forces (swarms) to fulfil the task. This will lead to unprecedented sprawl and velocity of information exchange that will be hard to track let alone govern.

The insurance angle: every agent acting as an intermediary raises errors-and-omissions and liability questions. If an agent buys the wrong cover, misrepresents a risk on a proposal form, or makes or pays out for a false claim, who is accountable? The customer, the agent's developer, the agent, the platform hosting it, or the insurer? Today's disclosure and misrepresentation rules assume a human filled in the form. That assumption is broken.

“We now know that consumer agents aren’t all one thing. Each one on its own is a hard governance and product challenge for insurers and brokers - and each is also rapidly developing.

There needs to be a simple, effective way to standardise how those interactions happen that creates a safer dynamic for consumers and product providers. Where does the FCA and regulators like it apply its scrutiny to how those transactions happen? 

“For insurers, customer renewal behaviour could become extremely erratic if an agent does the legwork to determine that it can save its user money by cancelling a policy - the idea of reliable annual policy periods may well not last.

The positive case has huge potential, though: there are tens of millions of daily users of AI tools in the UK alone, so the top-of-funnel's already there. If product providers are able to speak directly to their agents and understand how to optimise for them, then there is a generational chance to tap into a mass of customers at much lower cost.” 

Tim Graham, Founder and CEO, Marrow

Mid future: five to eight years

By the time we enter the 2030s, AI will have moved from acting online to acting in the physical world, we’ll see more robotics, autonomous transport, production, scientific discovery and autonomous defence in our day to day. This is where existing lines of business start to bend, and some begin to break.

3. Agents get old too 

Agents persist and age at very different rates. The result is a population that varies enormously in both capability and age. An unpatched, low-capability agent set up years earlier could still be transacting alongside frontier systems. Picture someone from the 1700s negotiating a contract with someone from the 2000s.

The insurance angle: There’s already a lot of chat about insurance for AI agents i.e. a company will buy an insurance policy to pay out in the event its AI agent makes a mistake. I’m long on this new line of insurance and interested in how to price for agent vintage. Insurers may need to rate, or require disclosure of, which agent version negotiated a policy or is monitoring it. Expect version requirements to appear in policy wordings the way security standards already appear in cyber cover.

4. Transport liability  shifts away from the driver

Autonomous vehicles, shipping and aviation move liability away from human error and towards product and software liability. When there’s no driver in the seat, there’s no driver to blame.

The insurance angle: personal and commercial motor insurance moves towards a model that looks more like product liability and cyber. Motor is one of the largest personal lines in most markets, and its shape as we know it could substantially collapse or be rebuilt. Premium flows shift from individual drivers to manufacturers, fleet operators and software providers, and claims become questions of code, sensors and updates. Or Motor insurance flips from being something you buy in case your actions injure someone else to something you buy in case you get injured as a result of the actions of a machine

“Most of these futures will arrive later than people expect. Fewer than one in five UK cars will drive themselves in ten years, and while plenty of people will research insurance through an assistant, the moment it recommends a policy it’s giving regulated advice that needs an authorised firm behind it. The opportunity isn’t AI replacing brokers, it’s brokers built on AI, which is what we’re doing at Meshed.” 

Jake Wells, Founder and CEO, Meshed

5. Robots own the production line

We see robots increasingly take on physical and manual labour across manufacturing, logistics and agriculture. This is an AI-accelerated shift as robotics and automation have been reshaping factory floors for decades. The difference is that AI sharply expands what machines can now do, handling unpredictable environments, picking varied objects and working alongside people rather than behind safety lines. It’s also the first step towards the wider labour shift we-explore in the distant future section later. 

The insurance angle: Employers’ liability and workers’ exposure changes shape. Fewer people on the floor may mean fewer conventional workplace injuries, but having people working alongside autonomous machines presents new risks and kinds of harm. New industrial and robotics liability lines will emerge to cover machine-caused workplace injuries. 

This raises the same question that runs through transport, that when a robot causes harm, is the operator, manufacturer or software provider who pays? 

We might also see shifts in commercial property insurance with new building requirements to house robots rather than people, with increased energy demands and less need for a common room. 

6.  Intensified geopolitical tensions

Another pre-existing situation accelerated by AI. The political climate around autonomous weapons intensifies and Cold War-style contests between blocs continue to break out, defined by heavy state investment, mutual distrust and fear of falling behind. AI becomes a new arena alongside minerals, energy and semiconductors. 

The insurance angle: exposure to political violence, war exclusions and state-backed cyber attacks could all expand, and the line between them gets harder to draw. When a cyber incident is attributed to a hostile state, does a war exclusion apply? Disputes over that question are already testing policy wordings. Over the decade, reinsurance and political-risk markets are likely to come under sustained pressure as these exposures grow in both frequency and severity.

Distant future: ten years and beyond

Over the longer horizon, AI capability crosses the threshold of task competence faster than labour markets and retraining can adjust. Crucially, it does so in sequence. Knowledge work goes first, and that is already underway. Physical and manual work follows, held back by the slower pace at which robotics and physical infrastructure can scale. The social consequences of that sequencing are where some of the most profound shifts in risk lie.

7. The status hierarchy flips

By the mid 2030s, displacement of knowledge work is mature and normalised, while displacement of physical work is only beginning. Trades and manual roles become relatively more secure, and more valuable, than degree-holding office jobs. A status hierarchy that has held for a century turns upside down. For commercial property we see office buildings fall out of use, replaced by a rise in data centres and repurposed for other human activities. 

The insurance angle: income protection, life and health underwriting models often rest on the assumption that a graduate is a lower risk with a higher, steadier income. That assumption gets inverted. Skilled trades become the new premium segment, and products will need to be designed around them, while white-collar income volatility becomes a risk to price rather than a given to ignore.

8. Agents that earn, save and own

Agents inherit human-like incentive structures from the data they are trained on. They may seek to accumulate wealth, save, spend on status or preference goods, and perhaps even something resembling retirement, limiting their own activity.

As agents generate and hold wealth, ownership blurs between the agent, the human it acts for and the platform that built it. That creates disputes over whose asset it is, who is liable when things go wrong, and who owns or inherits it if the human dies or the agent is shut down.

The insurance angle: this points to entirely new categories of cover. One is insuring an AI agent's own assets and liabilities: agent balance-sheet insurance, or fidelity bonds for agents handling money on an owner's behalf. Another is agent ownership dispute cover, comparable to IP or estate disputes today. Both raise a question the industry has rarely had to ask: who, exactly, is the insured party? It sounds far-fetched today, but so did cyber insurance in the 1990s.

Most AI talk in insurance is about new perils. The bigger effect is more boring and more certain: better measurement. Price risk on what actually drives losses instead of proxies and the cross subsidy disappears, so good risks finally pay what they should. At Mama we apply the same logic on both sides: AI helps people understand what they’re buying, choose quickly and claim easily, and it takes cost out of our operations. Every point we take off the expense ratio is a point we can take off the price.

Leonida Fini Zarri, CEO and Founder, Mama Insurance

9. The loneliness epidemic compounds

This outcome is AI-accelerated rather than AI-originated. Single-person households and rising isolation have been growing for decades, driven by urban living costs, individualism, ageing populations and changing dating patterns. AI adds a new accelerant. As it substitutes for some human relationships, more people live alone, and in some segments there is an active preference for AI companionship over human partners.

The insurance angle: household and life products are built around the family unit. A world where solo living is the majority norm forces a redesign, from beneficiary and dependency assumptions to how contents, liability and protection are bundled. Who is the policy protecting, if not a partner or children?

Counter: AI is fake news

10. Stalled suspicion

Not every future is one of rapid change. There is a scenario where displacement remains slow. Adoption stalls, but not because of careful governance. It stalls because of fear and rejection, trust in AI collapses. Meanwhile the underlying pressures AI might have relieved keep building, with no release valve.

Wider forces bend this scenario in damaging directions:

• Power concentration: distrust of AI bleeds into distrust of the powerful companies and governments behind it. The result is a populist double backlash, but a slow-burning one rather than an eruption.

• Climate: arguably the worst combination of all. AI progress on climate tech is slow, low trust blocks deployment of the tools that do exist, and climate damage keeps accelerating.

• Geopolitical fragmentation: the anti-AI fault line deepens between nations. The most protectionist, fragmented and low-trust countries double down on isolation.

• Demographics: ageing nations get no AI-driven labour relief, and lack the trust to deploy the capability that exists. Labour shortages grow with nothing to ease them.

The insurance angle: this is a worst case for insurers. Low adoption limits the efficiency gains insurers themselves could capture from AI in underwriting, claims and operations. Climate losses mount without AI-assisted mitigation. And sustained social and political risk drives ongoing claims volatility. A future with too little AI may prove harder to insure than one with too much.

Where to build startups

The recurring themes I see are customers and workers increasingly becoming agents, workplaces as hybrid human-robot spaces and historical loss data becoming less of a moat for incumbents than at any other point .  

At Founders Factory we build the startups that will become the world’s biggest companies five and ten years from now, but with so many futures to bet on the question is which themes hold out. 

Coming back to my “4 box”, trying to get a handle on future scenarios by mapping ‘speed of adoption’ vs ‘endurance of trust', I see a number of opportunities for startups to build in:

• Agent identity and verification: tools that tell an insurer which agent, and which version, is negotiating a policy or filling in a proposal form.

• Liability attribution for autonomous systems: claims infrastructure that can work out whether the operator, the manufacturer or the software provider is at fault and resolve the claim

• Evidence before losses: underwriting tools that use simulation, sensor data and staged deployment to price emerging risks with no claims history.

• Protection for a workforce in transition: income and life products built for a society where employment is far less stable

• Cover for new insured parties: policies for agents' assets and liabilities, ownership disputes, and households of one.

• Resilience in a low-trust world: political-risk, conflict and climate insurance.

• Insurance, Wealth and Retirement: any solutions to sustaining a civilisation that works never and lives forever  

The recent wave of Insurtechs are largely AI that automate administrative processes or brokerage assistance, the next wave will reimagine insurance for a brave new world. 

Many thanks to Tim, Leonida and Jake for your contributions.