From curiosity to commercial value: what a year of Agentic AI has taught insurance (410)

Simon Torrance: Foreign.

Zoya: Hello and welcome to the instec podcast. I am Zoya, and you may recognize my voice already from the podcast intros, which briefly explained what guests we have coming up this weekend, as well as perhaps being the first face you met at the ticket desk when you came to one of our events in person. I can guarantee you if you've interacted with me either virtually or physically, or with some other INSTEC content, such as with our weekly newsletter, or listen to some of our previous podcasts, you'll know that at instec, we spend a lot of time trying to understand which technologies are genuinely changing assurance and which are simply generating headlines. And over the last 12 months, one topic has appeared rather consistently and has had a curious trajectory in the way it has appeared. And that, of course, is by no surprise, Agentic AI. Now, depending on how closely you've been following the conversation, you might be thinking one of two things. Either you've heard the phrase so many times that, uh, you're wondering whether it's just become another piece of technology jargon, or perhaps you're still trying to work out exactly what people mean when they talk about agentiki in the first place. So before we go any further, it's probably worth spending a minute explaining what it actually is. The simplest way to think about agent AI is as a new generation of autonomous digital workers. Unlike traditional automation, which follows predefined rules, or generative AI tools that respond to prompts, which we've already seen in the industry being used for over the past year or two, agentic AI systems can understand objectives, create plans, take action, adapt to changing conditions, and collaborate with both humans and other AI systems to complete complex tasks. In theory, they're not simply helping people do work. These agents are increasingly capable of doing parts of that work themselves. That's why so many people believe agentic AI could have a profound impact on industries like insurance, where so much of the day to day activity revolves around processing information, assessing risk, making decisions, handling exceptions, and coordinating workflows between different people and different systems. And that's also why we've seen such an explosion of interest over the last year. But as mentioned by myself briefly earlier, uh, what I've found most interesting isn't the technology itself, it's how quickly the conversation around it has evolved. Every week, as part of putting together the instec newsletter, uh, I'm reading product launches, deployment announcements, research reports, investment stories, and industry commentary. Also, as part of marketing at instatech, I'm trying to work out how broader industry interests in specific new deployments in the market resonate with the interests of our Instec community. In speaking with insurers, brokers, MGAs, technology providers and consultants all seem to be trying to answer roughly the same question, what does Hntki actually mean for insurance? What's striking is that if I compare the conversations we're having internally today with the conversations we were having this time last year, they're almost completely different. So in this podcast episode, I will take us back to the conversations we were having last year, share, uh, some of the developments we've seen emerge across the market since then, and explore why the industry's focus has shifted from experimentation to something much more challenging. Along the way, you'll hear a mixture of perspectives. I've included audio excerpts from our first Agent ki event in November 2025, when many of these conversations were just beginning to take shape, alongside written responses submitted by speakers ahead of our upcoming event, the Age of Agenda From Strategy to commercial value, on the 7th of July. Together, they provide an interesting snapshot of how the industry thinking has evolved over the last 12 months. So, looking back in November 2025, we hosted what was, to our knowledge, one of the first major insurance events dedicated entirely to Agentic AI, in collaboration with AI Risk, run by Simon Torrance. More than 350 people joined us in London. The event sold out. But the mood in the room wasn't one of certainty. Perhaps the best way to put it was curiosity. People were trying to understand whether agentic AI represented another incremental step in automation, or whether it signaled, perhaps something much bigger. There was excitement, but there was also plenty of unanswered questions. Could these systems be trusted? How would they be governed? Would insurers actually deploy them? And, uh, perhaps most importantly, where would the value come from? One of the most memorable sessions came from Simon Torrance, who argued that we shouldn't think about agentic AI as another software tool. Instead, he challenged the audience to think about it as a completely new form of workforce. In his keynote, he explained as you

Simon Torrance: know, generative AI generates content. You ask it a question, it gives you an answer, but then you do the work. Now, agentic AI is rather different. It does the work for you. So AI agents are a bit like human workers, like you and me. They undertake tasks, they have agency, they have job descriptions, they have behavioral traits, they have access to data.

Zoya: That's quite a powerful idea, because if true, then we're not simply talking about technology adoption, but about a fundamentally different way of organizing work. And that's where the story starts. To become really interesting because over the months that followed, the industry did indeed begin moving from discussion conversations to actual, tangible deployment. One of the advantages of sitting where we do at, uh, Instech is that we get to see a lot of these developments emerge in real time. Every week as I'm putting together the newsletter, certain themes start appearing again and again. Sometimes they're isolated announcements, sometimes they're small product launches, but occasionally, when you get to step back and look at everything together, a pattern begins to emerge. And that's exactly what I see happening with agentic AI. Take cfc, for example. Earlier this year, the insurer launched Lane Assist, a live agentic underwriting platform capable of taking submissions from email through to, quote, recommendation in seconds. For years, discussions around AI and underwriting have focused on assistance, so helping underwriters find information faster, helping them assess risk more efficiently, and overall, helping them make better decisions. This was slightly different. This is an AI system participating directly in the underwriting workflow itself. Then there was McGill and Partners, which became the first London market broker to deploy AgentIC AI using Salesforce Agentforce. Not long after that came news of a collaboration between McGill and AIG using AI driven underwriting and agentic portfolio management. Elsewhere, Federato launched what it described as an agentic underwriting platform. Banyan Risk became the first MGA deploying Hyper Exponential's AI native underwriting suite. Duck Creek announced an insurance native agentic AI platform spanning underwriting, claims, billing and policy administration. That's a whole load of stories and examples. Individually, these stories are, of course, interesting, but collectively they're really significant and, um, paint a very interesting picture because they suggest that, uh, agentic AI is beginning to move beyond presentations and pilot projects and into real operational environments. But this doesn't mean that deployment is suddenly really easy. Actually, far from it. In fact, one of the strongest themes emerging from our conversation over the last year is that the biggest challenges aren't necessarily technical. They're organizational, cultural, and of course, commercial. When we asked speakers ahead of this year's agentic AI one day event what gets in the way of moving from pilots to production? The answers were remarkably consist. Erdal Atikan, COO and CTO at Insure Indigo, pointed to what he described as the human elements of conviction, trust and literacy that enable adoption, end quote. He went on to explain that pilots often stall because organizations lack the business leadership, understanding, or even excitement required to support them. Gina Gill from Apollo was equally direct. She told us, quote, using AI to show you're using AI rather than solve real problems. If a pilot exists to demonstrate AI adoption rather than address a specific operational pain, production is never going to be end game, end quote. What struck me about these responses is that they're not really talking about AI. They're touching on a broader macrocosm such as business transformation strategy and changes pertaining to leadership. Now they're talking about something we've all seen happen before. A new technology arrives. Organizations rush to experiment. Pilots appear everywhere, but only a handful ever make it to day to day operations. Simon Torrance captured this challenge perfectly when he said they're building pilots that can't compound. A pilot proves a task can be automated. It doesn't capture how your best underwriter actually reasons. So every pilot starts from zero and then the learning evaporates the moment it ends. Production at scale needs somewhere for that learning to accumulate. Most insurers haven't built it. That's the gap, not the technology, end quote. And that idea of compounding is important because it leads directly to what I think has become the defining question of 2026 economics. When we were planning this year's event, we asked speakers which of our five themes felt most urgent. One, the economic questions, two, real implementation, three, operating at scale in the London market, four, lessons from outside insurance and five, the agentic insurer of 2030. What surprised me was how many people independently chose the same answer. The economics question. Simon Torrance described it as the economics question. By distance, boards are signing AI budgets against a baseline nobody can measure. Most insurers can't tell you what their processes actually cost. So ROI gets reconstructed backwards and the number that wins funding rarely compounds. Get the economics wrong this year and you don't lose a pilot, you compound the wrong bet for five. Max Richter, uh, from MIA Platform said open quote, boars are no longer funding AI Curiosity. They want to know where the return comes from, how quickly it lands and whether it compounds across underwriting, claims and operations, end quote. Gina Gill from Apollo echoed the very same. In the softening market, the cost benefit case for AI needs to be honest and rigorous rather than aspirational. And Elena Moran from Alphesis AI highlighted another challenge. The absence of a clear economic case, as most teams cannot quantify benefits against costs, stems coming from risk materialization and resistance to reinventing processes. Rather than retrofitting AI into existing ones, pilots stall because the business case never matures. Beyond efficiency claims. What's fascinating is that this isn't really a, uh, technology problem anymore. It's a measurement and value problem. One that makes you ask yourself, do you really have the business case for this. Most organizations can see what's possible. Now the question has moved into whether they can prove it's worthwhile. And that's where another theme begins to emerge. Competitive advantage. Because even if agentic AI works, and even if organizations can deploy it successfully, there's still an uncomfortable question sitting underneath everything. What happens when everyone has access to the same technology? This is where Simon Torrance's concept of intelligence capital becomes particularly interesting. His argument is that most AI spend buys parity, not advantage. The tools you buy today, your competitors buy tomorrow. That isn't advantage, it's parity at a higher price. The only AI that compounds is the reasoning your business has already built, how your people assess risk and handle exceptions. That's intelligence capital. Most insurers are about to miss this value, end quote. Whether you agree with that argument or not is difficult to ignore. Because if every insurer can buy similar models, similar platforms, and similar tooling, then differentiation has to come from somewhere. Perhaps it's from your data expertise, operating models, or perhaps it comes from how effectively organizations combine humans and AI together. Which brings us to another major shift I've noticed over the last year. The conversation is no longer just about singular agents. It's increasingly about orchestration. Earlier this year, we hosted an event in New York with ServiceNow called the Future of Insurance Will be orchestrated, not built. The central idea was, uh, that once organizations begin deploying multiple AI systems across underwriting, claims servicing and operations, a, uh, new challenge emerges. How do you coordinate them, how do you govern them? And how do you ensure they're accountable? And how do you maintain trust? Elena Moran addressed this directly when she told us governance has to move from the system level to the decision level. With oversight applied at runtime. This means each agentic decision has to be evaluated for risk as it happens and rooted to the component human authority. And when the level of risk warrants it, end quote. She then added something that feels particularly relevant as adoption accelerates. Something certified once cannot be assumed safe under life conditions, end quote. These aren't future concerns. They're currently happening. And they're concerns that, uh, insurers, brokers and MGA's are already wrestling with today. And perhaps that's why this year's agentic AI event feels so different from the one we hosted in November. Last year, the industry was still exploring possibilities, but this year the conversation has become much more practical, more commercial, and arguably more important. We're talking about implementation, governance, operating models, workforce transformations, competitive advantage, and ultimately, we're Talking about value. As Ian Thompson told us, real change will only be embedded through reshaping strategic workforce plans. End quote. And as Urdel, uh, Atakan reminded us, focus, focus, focus. There is so much noise in this sphere. Pick one major idea to form a team around to implement, then reflect on if the whole team understands the objective and has the context and knowledge required to achieve your aim. Start simple with the most practical part and increase complexity as you demonstrate success in achieving the main idea. End quote. That's perhaps the best advice of all because despite all the excitement around agentic AI, despite the announcements, the pilots, the predictions, the organizations that succeed are, uh, um, unlikely to be the ones chasing every single new development. They're more likely to be the organizations that understand where they can create value, focusing on the right problems and execute this consistently. So as we prepare for the age of agenda KI from strategy to commercial value on the 7th of July, those are the questions we'll be exploring, not whether Agent Aki is real. That conversation has largely moved on and we have with it. The questions are now where the value sits, what successful implementation actually looks like, and how insurance organizations can build capabilities that endure long after the hype cycle has passed. Thank you for listening to our brief summary of what we've been seeing regarding agentic AI here at Instech. Now, if you've made it this far, then I do actually have a little surprise for you. If you email in at hellonstech co wanting to come to our event on the 7th of July, we can offer you a, ah, slight discount on ticketing. Just please let us know what ticket bracket you fit into and we'll continue the conversation on email. But thank you very much for listening once again and I hope that little surprise reaches a few of you keen listeners and hopefully see a lot of you in London in July. Thank you. Well, if you've made it this far, then I'm pretty sure you found that as interesting as I did. The Instec podcast comes out every Sunday morning where we spotlight the latest news, leading voices and freshest updates across insurance that you need to know about. If you would like to take part in these conversations, head to www.instec.co to find out how you can join our network and um, be a part of the insurance intelligence for the curious.