A new generation of artificial intelligence (AI) is beginning to move beyond analysis and into action within the insurance industry.
Bringing together experts from both the insurance and technology landscapes, a recent virtual expert panel, hosted by Tinubu, explored where such technology actually stands today and how it’s set to shape the sector.
Moderated by Mark Abrams, Head of Trade and Receivables Finance at TFG, the discussion centred around agentic AI, a cluster of systems designed to operate with far greater autonomy than ever before, carrying out tasks with minimal human input.
Unlike earlier forms of digitalisation, which often relied on rigid systems or relatively basic machine learning models, agentic AI is designed to coordinate processes across far more complex workflows, leaving its predecessors somewhat in the dust.
Rather than merely analysing information, these new programmes can retrieve data, interact with other platforms, and execute multi-step processes. In doing so, they behave less like passive tools and more like digital co-workers, seemingly supporting faster and more efficient decision-making.
As Abrams explained in his opening remarks, the purpose of the panel was, in fact, to evaluate “what’s real, what’s challenging, and what feels achievable in 2026” regarding AI’s implementation.
Operational applications of agentic AI
For most insurers, the real hurdle of the job is the sheer volume of information underwriting requires. Assessing a single risk can mean pulling data from internal databases to financial records and market reports, often scattered across different platforms. On an operational level, this means underwriting is rarely the tidy, linear process it might appear to be on paper.
This is where agentic AI begins to take shape, and where it starts to look particularly appealing for insurers. That said, if the phrase ‘agentic AI’ conjures images of fully autonomous systems running insurance desks, the reality is a little more down to earth.
Well, at least for now.
Taking a closer look at its more tangible impact, organisations are already beginning to experiment with agentic AI that could take over some of the industry’s most time-consuming legwork. In fact, according to Artūrs Karlsons, Director at Berne Union, the majority of Berne Union members have already begun preparing pilot projects, with some moving into the practical deployment stage for specific use cases.
Adding to this, Scott Ross, Emerging Technology Specialist at AWS, pointed out that the shift isn’t limited to the integration of AI within companies; it involves the implementation of the entire technological ecosystem surrounding automation itself.
In previous phases of digital adoption, organisations were more tied down by the familiar ‘build versus buy dilemma’ when selecting new tools. But now, with agentic AI, that binary is beginning to dissolve, and firms are increasingly looking for more holistic solutions.
Rather than managing multiple AI tools internally, companies are now choosing instead to purchase all-encompassing infrastructure, capable of handling processes on a more foundational level. This enables firms to automate the majority of the workflow that is mandatory and non-differentiating, allowing them to then focus on the remaining portion of work that actually differentiates them competitively.
This is reflected in the move toward a new ‘buy-partner-build’ model, where partnerships allow organisations to stay adaptable, while concentrating internal resources on that higher-value work.
AI for augmentation - not replacement
However, the panellists made it clear that cost and efficiency are only part of the motivation behind AI acquisition. For many insurers, the technology is framed less as a tool for reducing headcount and more as a way to reshape an organisation.
Jon Holvoet, Chief Technology Officer at Credendo, emphasised that the firm’s goal is not to replace its workforce, noting that in their niche sector, “the human relationship is still of the utmost importance.”
Instead, he highlighted how the ambition is to simply become an “AI-enabled company,” where automation handles lower-priority tasks while underwriters focus on judgment.
To support this transition, Holvoet suggests five principles he considers essential for enterprise-ready AI:
- A governed architecture ensuring security and auditability
- Modular orchestration of agents and tools
- Built-in compliance and regulatory guardrails
- Deep integration with core business systems and data
- A workforce that is AI-literate enough to understand both the risks and potential of the technology
However, this framing of AI as a tool solely used for augmentation rather than replacement is not entirely straightforward.
If a significant share of routine underwriting processes can be automated, the distinction between augmenting workers and gradually reducing the need for them may become increasingly blurred. This inevitably raises broader questions about what happens to the human labour currently responsible for much of that standardised work.
Emerging risks and regulatory questions
While much of the discussion focused on the positive operational potential of agentic AI, the panel also emphasised that the new technology definitely has its grey areas, introducing a new set of risks and challenges for insurers.
One of the most immediate concerns by far is liability. “Customers are looking for those agents to be, quote, ‘certified’,” said Ross. “If I have an agent that does a regulatory check and the partner guarantees that it will be accurate, and if I buy this agent, then who's liable?” he asked.
The issue amplifies a central obstacle, as the delegation of administrative tasks to automated systems complicates traditional frameworks of responsibility and oversight.
Beyond liability, a closely related concern is data governance. The effectiveness of agentic AI within the sector ultimately depends on the quality and organisation of the data feeding these systems. The transfer toward automated workflows is exposing long-standing weaknesses in how companies have historically managed their information.
“In the past, it was not too bad to have not the best labelled data.” said Tim Hilbig, Head of Digital Concepts and Innovation at Euler Hermes. “We really have not managed data correctly until now.”
Such gaps not only make it difficult to trace where errors originate but also raise concerns about accountability and whether flawed data could undermine the reliability of the wider AI infrastructure.
An ongoing experiment
Last but not least, the panellists stressed that the real test for agentic AI will be whether it delivers genuine, measurable improvements in practice. For Hilbig, this means treating its implementation as an ongoing experiment, rather than one-off technological integration.
AI-driven processes, he argued, must be continuously evaluated and compared with traditional workflows to determine whether the technology truly saves time and meaningfully supports existing human roles.
For Holvoet, this means, “always start with the problem and don’t focus on the technology.” In other words, businesses must first identify the issue they are trying to solve, and only then determine whether AI is the appropriate tool for the job.
Systems that appear effective today might not remain so tomorrow, making it essential for companies to approach AI-adoption pragmatically and apply it within the right operational context.
—
While agentic AI is beginning to reshape the insurance industry, its long-term impact depends on how firms manage governance, data, and, perhaps most importantly, the human role surrounding it.
The panel ultimately revealed a subtle tension: while the potential benefits of automation are widely recognised, the practical realities of implementing it responsibly remain complex.
“Looking ahead, there’s a need for us to collaborate to define the protocols, the best practice, and the best standard,” said Yvan Saule, Chief Technology Officer at Tinubu. “That is the way we can all win together.”
Watch the full discussion
This article highlights some of the key perspectives shared during the panel, but the full conversation explores the operational realities of agentic AI in much greater depth.
You can access the full webinar replay here
