Is Insurance About to Repeat Its Biggest Technology Mistake with AI?

Today’s legacy technology estates were never intended to become barriers to innovation. Most began life solving genuine business problems. Over time, however, new products, changing regulation, rising customer expectations, and successive waves of technology added new systems, integrations, and workarounds until complexity became embedded in the operating model itself. 

The result was technology estates carrying increasing technical debt and operational complexity, making meaningful transformation progressively harder.  There is now a growing risk the industry is laying the foundations for its next generation of legacy, this time with AI.  

Across the industry, organisations are deploying copilots, chatbots, productivity tools, and isolated AI use cases that promise faster execution and measurable efficiency gains. Many are delivering genuine value. The danger is not that these projects fail. It is that they succeed just enough to convince organisations they are transforming when they are actually reinforcing the structural constraints that created today’s legacy challenge in the first place. 

The New Layering Problem 

For years, insurers have dealt with technology limitations by adding another layer. A new integration. A new workflow. A new application. A new vendor. A new workaround. Each decision solved an immediate problem. Collectively, they created an environment that became increasingly difficult to change. 

AI creates a similar temptation. When an insurer faces a process bottleneck, it is often easier to deploy an AI tool than to redesign the process itself. When customer servicing is slow, an AI assistant can help. When employees spend too much time reviewing documents, AI can summarise them. When claims teams are overloaded, machine learning models can support decision-making. 

None of these are bad outcomes. But they do not necessarily address the underlying issue. If AI is simply layered onto fragmented processes, disconnected systems, and outdated operating models, insurers may find themselves automating complexity rather than eliminating it. 

Why Agentic AI Changes the Equation 

Previous waves of technology modernisation focused primarily on digitising existing activities. Agentic AI is different because it moves AI from assistant to operator. Rather than supporting isolated tasks, autonomous agents can increasingly monitor conditions, coordinate actions, trigger workflows, escalate exceptions, and execute processes across multiple systems simultaneously. 

That shift makes the operating environment far more important. A chatbot can answer questions in isolation. A machine learning model can identify patterns. A copilot can help an employee complete a task. But agentic systems require connected data, interoperable systems, accessible workflows, and operational environments capable of supporting continuous coordination across the enterprise. 

Without those foundations, autonomous systems quickly encounter the same barriers employees face today: disconnected information, fragmented processes, manual handoffs, and organisational silos. The technology may be more sophisticated, but the operating constraints remain the same. 

The False Confidence of Early Success 

This creates one of the biggest strategic risks facing insurers today. Many organisations will achieve meaningful AI wins over the next few years. Productivity will improve. Tasks will become faster, operating costs may fall, and customer interactions will become more efficient. Those successes are important but they can also be misleading. 

An insurer may believe it has successfully embraced AI because individual initiatives are producing measurable results. Yet the organisation may still be relying on technology foundations that prevent AI from scaling across the business. The challenge is not whether AI can generate value in isolated use cases. The challenge is whether the organisation can support what comes next.  

As AI becomes increasingly autonomous, competitive advantage will depend less on access to models and more on the ability to orchestrate them effectively across the enterprise. That is not primarily an AI problem. It is an operational and architectural problem.  

When AI Starts Knocking on Your Door 

There is an external dimension insurers need to consider. As agentic ecosystems mature, insurers will not only deploy their own agents internally. Increasingly, external agents will come looking for policy information, pricing, eligibility, claims status, decisions, and service actions. 

That raises a very different readiness question. Can the organisation make the right information available in machine-readable form across backend systems, with the necessary permissions, governance, and security controls in place? If not, the insurer may become harder for agent-led ecosystems to interact with. 

In a market where more interactions are initiated, filtered, or mediated by autonomous systems, being difficult for agents to understand could become a new form of invisibility. 

Why Architecture Matters More Than Ever 

Research from the MACH Alliance highlights a growing divide between organisations able to operationalise AI successfully and those struggling to move beyond experimentation. The difference is rarely access to technology. More often, it is the architecture sitting beneath it. 

Organisations operating on modern, composable architectures are generally better positioned to deploy AI quickly, connect data effectively, integrate new capabilities, and achieve measurable business outcomes. Those operating on fragmented legacy estates face a different reality. Integration complexity, disconnected data, and operational silos create friction that slows progress and limits what AI can realistically achieve. 

This becomes even more important as agentic systems mature. Autonomous systems cannot effectively coordinate workflows across an organisation if the underlying systems cannot coordinate with one another. Composable architectures help address this challenge by allowing insurers to modernise incrementally while creating the connected environment autonomous systems require. 

More importantly, they create the flexibility to adapt as AI continues to evolve because nobody knows exactly what the next decade of AI innovation will look like. The organisations that succeed will not be those that predict the future most accurately. They will be those that build operating models capable of adapting to it. 

Avoiding Another Decade of Constraint 

The greatest risk facing insurers is not that they fall behind on AI adoption. It is that they adopt AI in ways that reinforce the structural limitations already holding them back.  

The industry has spent years attempting to overcome the consequences of technology decisions that prioritised short-term fixes over long-term adaptability. AI offers an opportunity to break that cycle. But only if insurers resist the temptation to treat it as another layer. 

The winners will not be the organisations that deploy the most AI. They will be the ones that use AI as a catalyst to rethink processes, modernise operating models, and build foundations capable of supporting continuous change. The real question is not whether AI can transform insurance. It is whether insurers can avoid repeating the same perfectly rational decisions that created today’s legacy constraints, only this time with AI.