TechTalk: Vibe-coding could leave insurers ‘crashing on the rocks’

A war hero sailing home must pass the sirens – mer-like creatures whose songs enchant sailors and draw them, unaware of the risks, towards their unfortunate fates. This is one of the most famous trials faced by Odysseus in Homer’s epic poem The Odyssey.

Much like the sirens, the enchanting call of artificial intelligence (AI) has tempted many firms in the UK insurance industry to begin implementing it, despite its unknown risks.

And, one new aspect of this AI-related risk that has emerged is vibe-coding.

The term vibe-coding was coined by OpenAI co-founder Andrej Karpathy in a statement made on the social media platform X in February 2025.

In the post, Karpathy defined it as a “new kind of coding” where “you fully give in to the vibes, embrace exponentials and forget that the code even exists”. He wrote that this is “possible because the large language models (LLMs) are getting too good”.

Later that year, vibe-coding was added to the Collins English Dictionary and defined as “the use of AI-prompted by natural language to assist with the writing of computer code”.

The official definition reflected an uptick in adoption – and this growing popularity extended to the insurance sector.

According to Gallagher Re’s most recent Global InsurTech Report, released in August 2026, there was a swing in the direction of incumbent (re)insurers venturing to build their own tools and platforms via vibe-coding.

Taking such a laissez-faire approach to software development within the highly regulated UK insurance sector, however, has raised a few eyebrows when it comes to governance, accountability and compliance.

These concerns are especially heightened when AI adoption is currently outpacing investment in governance in the UK.

Indeed, Ernst-Young’s 2025 Responsible AI Pulse Survey, based on 50 UK financial-services c-suite executives, reported that 26% of respondents said they had no or limited controls in place to ensure AI systems adhered to laws and regulations.

This is despite the survey revealing that 34% of UK financial services firms said that they have fully integrated AI solutions within their operations – with 44% having many AI systems embedded.

Speaking to Insurance Times, Douglas Dick, UK head of emerging technology risk at KPMG, said that the “biggest misconception” about vibe-coding in the market is that “AI-generated code reduces the need for governance”.

He explained that the reality is that vibe-coding only “increases the importance of effective oversight, testing and accountability”.

“AI-generated code can introduce risks that may not be immediately visible, making robust validation essential,” he continued. 

“Especially in regulated environments like insurance, firms need clear ownership of decisions and auditability of changes.”

For this reason, Tim Hardcastle, chief executive at insurtech Instanda, said that the “promise” of vibe-coding as a shortcut to software development is largely limited to “simple” and “standalone” platforms.

He warned that the “siren” call to use vibe-coding to redesign complex core systems must be resisted, as it will ultimately leave firms “crashing on the rocks”. 

The governance gap

Mythological analogies aside, Hardcastle believes that the “challenge for insurers” when it comes to vibe-coding is that their core systems must be auditable to comply with regulatory requirements.

This is especially the case, he noted, because some MGAs are using AI tools to build quote and buy platforms.

Echoing this sentiment, Andrew Johnston, global head of insurtech at Gallagher Re, said he believes that the biggest issue with vibe-coding is that firms run the risk of building systems that they may want to change in the future and discover that the “underlying model has fundamentally changed”.

He explained that because these firms have no control over the code that was written during vibe-coding, they may “not understand that the underlying system has changed” or “how to fix” what “has been created by a third party”.

As the FCA places the onus for AI governance on firms regardless of whether the technology was produced by a third party provider, vibe-coding used in this way could subject businesses to compliance risks. 

A spokesperson for the regulator added that it is expected that firms use AI-assisted coding tools with the “appropriate systems and controls to understand, test, monitor, secure and maintain their technology and to ensure that their systems and infrastructure are operationally resilient”.

Further, ensuring operational resilience is paramount for keeping in line with Consumer Duty.

Johnston told Insurance Times: “I don’t think fixing AI-generated code is as big of an issue as others may think because the underlying tools are improving themselves and the days of having to fix something code by code are possibly coming to an end [in favour of] simpler processes.

“But, if you’re wanting to develop a system that your entire business runs on where customer data, payroll and people’s livelihoods are part of the picture, you wouldn’t want to vibe-code something like that into existence in its entirety.”

Where to draw the line?

While there are plenty of technologies that can appear easy to experiment with, Martin Henley, chief executive at Mea Platform, said that turning them into enterprise-class, production-grade systems that can withstand regulatory scrutiny is “pretty difficult”.

The insurance industry has seen this pattern before time and time again, he explained, with similar tech-driven initiatives often proving tough to “stick” because of the complexity of property and casualty (P&C), commercial and specialty lines.

However, Dick noted that KPMG is seeing a growing interest from insurers and MGAs interested in AI coding tools and AI-assisted software development.

While a few firms are allowing AI to autonomously build or modify core underwriting and pricing systems, he explained that “many are successfully using AI to help developers deliver change faster where they have transformed the end-to-end development lifecycle”.

He continued: “The strongest use cases are often focused on underwriting operations, where AI can extract information from unstructured documents and provide underwriters with decision-ready insights.

“AI coding tools represent a significant opportunity for insurers and MGAs, but the value goes far beyond writing software faster. The organisations seeing the greatest benefits are using AI to transform entire processes, not just standalone issues.”