If AI can run a virtual insurance company, what else can it run?
We handed the controls of a virtual insurer to student-designed AI agents and watched what happened. The findings say less about insurance than about the future of every organisation.
Can agentic AI realistically run a virtual insurance company with minimal human intervention? That was the question we set out to explore at the Hack-to-the-Future Thinkathon, a first-of-its-kind initiative by Trusted Hub in partnership with Temasek Polytechnic's School of Informatics & IT.
At first glance it might seem like a niche challenge. But beneath the surface lies a question every organisation will soon have to confront: what happens when AI stops being a tool and starts becoming a decision-maker? To find out, we handed the controls of a virtual insurer to student-designed AI agents and watched what happened.
01Why insurance?
We chose the insurance industry deliberately. Insurance is one of the most complex decision-making environments imaginable, where each choice affects multiple stakeholders. Premiums influence customer acquisition, coverage affects profitability, claims impact trust, service quality shapes retention, and risk determines survival.
An insurer must constantly balance growth, customer satisfaction, financial sustainability, fairness and risk. In other words, it is a miniature model of almost every modern organisation — if AI can navigate an insurer, then in principle it can navigate the fundamentals of logistics networks, healthcare systems, financial institutions and public services.
02Putting agentic AI in the driver's seat
Student teams were tasked with designing AI CEOs capable of running a virtual insurer over a simulated multi-year period. Their agents had to make decisions across a range of competing priorities:
- Customer growth and pricing strategy
- Claims management and risk mitigation
- Customer satisfaction and retention
- Long-term profitability
Each team experimented with different prompt architectures, governance mechanisms, operating philosophies and AI models. Some built highly structured systems with strict decision frameworks; others allowed the AI greater freedom to interpret business objectives.
What follows are the three findings that mattered most.
AI doesn't replicate business models. It reinvents them.
One of the most striking observations was that AI often approached problems differently from humans. The agents were not merely following instructions — they developed strategies, trade-offs and operating behaviours that nobody had explicitly programmed. Some pursued aggressive growth that expanded customer bases but created unsustainable risk; others became overly conservative, protecting capital at the expense of growth.
The most successful systems were rarely the ones with the most detailed instructions. They were the ones that provided clear objectives, appropriate constraints, and room for adaptation. Agentic AI does not simply automate existing processes — it has the potential to redesign them entirely.
Autonomous AI behaves in ways nobody predicts.
Many teams entered believing better prompts would naturally produce better outcomes. The reality proved more complicated. Small changes in instructions sometimes produced dramatically different business behaviours, and two models operating under similar objectives could arrive at entirely different strategies.
This does not mean AI should not be trusted. It means organisations must learn to design systems that can safely accommodate unexpected behaviour while remaining aligned with strategic goals. As AI systems become more autonomous, predictability becomes harder.
Governance is not optional.
Perhaps the most important lesson had nothing to do with AI performance — it had to do with governance. Without guardrails, optimisation often produced unintended consequences. An AI focused purely on profitability might become unfair; one focused purely on satisfaction might become unsustainable; one focused purely on growth might expose itself to catastrophic risk.
The strongest teams understood that success was not about building the smartest AI. It was about building the most trustworthy system. Governance can no longer be treated as a compliance exercise or an afterthought — ethical guardrails are not a checkbox, they are a first-principles design decision.
"The biggest question in business is no longer how to use AI. It is how organisations will adapt when AI becomes capable of participating in decisions once reserved for humans."
04Beyond the Thinkathon
The Hack-to-the-Future Thinkathon was never intended to answer whether AI can run a business. It was designed to help us ask better questions:
- What kinds of decisions should AI make — and which should remain human?
- How do we govern autonomous systems responsibly?
- How do we build AI that is not only intelligent, but trusted?
These questions will shape the future of every industry. The Thinkathon offered a glimpse of that future — not one where AI replaces organisations, but one where organisations learn to collaborate with increasingly capable agents that can analyse information, execute workflows and support decision-making at unprecedented scale.
// Takeaways
- Given objectives and constraints, AI reinvents business models rather than copying them.
- Autonomy brings unpredictability — systems must safely absorb surprising behaviour.
- Governance is a first-principles design decision, not a compliance checkbox.
- The goal isn't the smartest AI — it's the most trustworthy system.
And if AI can run a virtual insurance company, perhaps the more interesting question is the one we're left holding: what else can it run?