Most companies claiming to use artificial intelligence effectively are not, according to Clayton Fuchs (pictured), vice president of casualty at QuoteWell, an Austin, Texas-based, tech-enabled wholesale insurance brokerage. Loading a chatbot into an email inbox does not count, Fuchs said – and the gap between wholesalers and retail agencies that treat AI as infrastructure across their entire workflow, and those that don't, is what he expects to separate the industry's winners from its losers over the next five years.
Fuchs said QuoteWell's operating rule is to automate the transactional work – loss run requests, submission intake, drafting a first response – while keeping underwriting and placement decisions with a licensed person. An AI agent, in his framing, is software that carries a multi-step task from a submission to a finished work product, then pauses for a human to confirm the result before it moves further down the chain.
“You never want AI to make the call,” Fuchs said. “What you want is to make the call yourself – but do it faster and more efficiently.” That line matters most, he said, at the point where information moves between an underwriter and an agent and carries potential errors and omissions (E&O) exposure. “That's where we need to be locked in and have all the tools to make an informed decision. That's where it always needs to be irreplaceable.”
Fuchs's more novel concern sits beyond his own operation: what happens as agentic tools proliferate across the length of the placement chain. “If an AI agent from a client starts talking to an AI agent at a retail agency, that starts talking to an AI agent at a wholesaler, that talks to an AI agent at a carrier, that talks to an AI agent at a claims TPA that's where we have to make sure we're protecting the guardrails, that everything going in and out makes sense, and that it's not hallucinating,” he said.
That risk sits alongside a broader industry conversation about where agentic systems introduce exposure that didn't previously exist. Recent coverage of agentic AI's expansion into cyber underwriting has raised similar questions about autonomous systems operating with limited human oversight, while wholesale distributors are increasingly expected to show that AI tools improve outcomes across an account's full lifecycle, not just at a single touchpoint.
Fuchs pointed to research beyond insurance to frame the adoption problem: a widely cited MIT-affiliated study found that the large majority of enterprise generative AI pilots fail to produce a measurable business return, largely because the tools are never properly integrated into how people actually work. QuoteWell's answer, he said, borrows from the “forward-deployed engineer” model several of its founding team used at Palantir – embedding its technology staff inside day-to-day brokerage operations rather than building tools at a distance and handing them off.
“Adoption is a goal – not a nice-to-have outcome,” Fuchs said. He described the company's process as iterative and evidence-based: forming a hypothesis about a workflow problem, testing a fix, and killing or scaling it based on whether producers actually trust it enough to stop doing the task themselves. QuoteWell's self-reported team composition has shifted accordingly – from roughly 90% technical staff at founding three and a half years ago to about 30% today, even as technology still accounts for roughly 40% of spend, according to Fuchs.
Fuchs predicted the wholesale and retail sides of the distribution chain will diverge sharply on outcomes, not effort. On the carrier side, he said, the difference is already visible in faster response times and more detailed, better-targeted quote requests from underwriters using AI well. The retail side, by his account, remains further behind, weighed down by account managers handling a high volume of repetitive, mundane tasks per account – the same friction QuoteWell has spent three and a half years building against internally, and now intends to sell to retail agents directly through a new technology arm.
Whether that divide plays out as AM Best and other analysts of the surplus lines market have separately suggested – a widening gap between distributors using AI across the full submission-to-bind cycle and those treating it as a single-task fix – is likely to shape which wholesalers and agencies hold their placement volume as the market softens, Fuchs said. “We're going to see a lot of winners and losers,” he said. “Who is all in right now? Who is adopting the technology? Who is doing it most efficiently?”
