Actuaries will only use AI they can audit: Akur8 on trust at ITC Vegas 2026

Why did a five-day AI feature never ship?

Capparelli opened with a story from Akur8’s early days with AI. One of its top engineers used AI tools to build a feature in five days. The same kind of work had taken a team of engineers three to six months.

It never went out to customers. An enterprise product has to pass reviews, validation and quality checks, and the team could not validate the new feature the way it was built. So nobody trusted it. The review process had been designed for the speed and scale of work before AI. Until the company changed how it checks work, the gains stayed on paper. In his words, “it wasn’t as much a technical challenge.” It was a people and culture problem.

How did Akur8 change its culture around AI?

Akur8 set up an AI transformation team. It is made of leaders from across the business, not only product engineers but sales executives too. These are people who know where the old rules sit and can reshape them. The slide listed the rest: honest talks about the risks, small team coaching, set time for everyone to explore AI, and champions who sit down with the people who adopt later.

Holmes shared a lesson of his own. When Slack messages started arriving that were clearly written with AI, his first thought was, “Did you do any work?” Then he saw that people were saving time. In a company that works in many languages, it was often just a way to clean up the English. Now he assumes the message is checked and complete, and he gives feedback as usual. Leadership, he said, is now even more about vision. Teams get more freedom, and ideas get pruned early.

Why is actuarial work different from other work?

Holmes was clear that actuaries cannot move the way a software team can. His slide said hallucinations can lead directly to insolvency and regulatory action. Actuaries set reserves. Regulators act on pricing. Only some actuarial judgment can be copied by a model, and the profession runs on strict credentials.

He made a prediction: in about three years, someone will lose their actuarial credential over improper use of AI. He also warned that a regulator can ask, years later, why a number was set. If you cannot replay the logic, you are exposed. He pointed the room to Precept 1 of the actuarial Code of Professional Conduct, which asks actuaries to act with integrity and competence and to uphold the reputation of the profession. He called that last part the most important.

“Your real bottleneck with actuaries is trust.”

Thomas Holmes, Akur8

What is constraint engineering?

Capparelli explained how Akur8 now builds with AI. If you let AI run free across a complex system, you end up with a tangle that no person can understand. For a company that sells software to insurers, losing control of the system is not an option.

So the system is split into smaller parts. People design each part and set its limits, which are human and software checks that always give the same answer. The AI then works inside those limits. Holmes gave an actuarial example. The rules for how data is grouped and totaled should be written down as fixed logic first, and the AI should use those rules rather than make up its own. Otherwise, he said, you will not know if your data has gone wrong.

What four controls keep AI trustworthy in actuarial work?

On context, Capparelli said good documentation, even going back thirty years, is now a real advantage. “The context you can build as an organization becomes your source of truth,” he said. Holmes then asked him a question they had not rehearsed: what if a mistake gets into that context? Capparelli’s answer was that data quality now matters more than ever, because a mistake in the context repeats every time.

On pricing, Capparelli was firm. Even if you give an AI model every rating rule in its prompt, it will not always give the same answer, because these models are not deterministic. The premium should come from an execution layer you already trust. The AI calls it and uses the result.

Where does AI stop and the actuary take over?

With seconds left, I asked from the floor where the line sits between what AI can decide and what needs an actuary’s sign-off. Holmes answered in one breath:

“AI will create the numbers. Actuaries will decide what that number means and if that number is reasonable.”

Thomas Holmes, Akur8

He added that AI might one day produce a draft statement of actuarial opinion, but the line is drawn at credibility and auditability. He does not see AI wiping out junior actuaries either. Nobody needs a new actuary to copy and paste cells anymore, he said, but “I need you to be able to look at a lift curve and see if it actually makes sense.” That, he said, is the heart of actuarial judgment.

Topics