Three things to take back to your agency

1. Autonomy is a ladder, not a switch. Know which rung your agency is on before you climb.
2. The joining up is the hard part. AI is only as good as the connected systems underneath it.
3. There's no partial credit for safety. One check is a guess. Independent checks are a system, and exceptions go to your experts.

"Who here has ridden in a Waymo?"

That's how Mohit Chawla, Founder and CEO of eventual, opened his main-stage session, AI Without Anxiety. Anyone who has taken that first ride knows the moment. There is no driver behind the wheel. No hands on the controls. Yet within minutes, the anxiety fades. The ride feels smooth, predictable, and surprisingly ordinary.

That ease took the self-driving industry more than a decade to earn. Chawla, whose company came to BrokerTech Connect as a BTV Innovator in 2025 and returned this year as presenting sponsor, used that experience as a lens for understanding where insurance stands in its own AI journey. Rather than viewing AI adoption as a sudden leap from manual work to full automation, he suggested insurance leaders think about it the way the self-driving industry does: as trust earned one level at a time.

"The way they describe autonomy is as a ladder, not as a switch that you flip overnight," Chawla said, referencing the industry's five levels, from systems that merely assist a driver to cars that handle everything on their own.

The autonomy ladder, in a car and in an agency

• Level 5 – Full. In a car: drives anywhere, in any conditions. In an agency: not the goal. Judgment stays human.
• Level 4 – High. In a car: drives itself within a defined area. In an agency: a whole workflow runs end to end. Your team audits.
• Level 3 – Handoff. In a car: drives, but you must take over when asked. In an agency: routine work clears itself. Exceptions come to your team.
• Level 2 – Partial. In a car: steers and brakes; you supervise. In an agency: AI proposes. Your team checks every line. (Most agencies today)
• Level 1 – Assist. In a car: helps with steering or speed; you drive. In an agency: AI drafts and summarizes. Your team does the work. (Most agencies today)

Car levels follow the industry's standard definitions. The agency column is an illustration of the same idea.

Then he turned the question to the room: where is your agency today?

For most, the honest answer is near the bottom. Many brokerages are testing generative AI tools to summarize documents, draft communications, or support servicing teams, but people remain firmly in control and the technology serves as an assistant.

The question now isn't whether AI will transform insurance. It's what needs to happen before an agency can safely hand it more responsibility. Chawla offered three lessons from the road.

1. Good intelligence requires good infrastructure

Autonomous vehicles don't run on a single breakthrough model. They rely on sensors, maps, software, and backup systems working together. None of it is glamorous, and none of it is optional.

The same is true in insurance, Chawla argued. Take a single renewal: the exposures, loss runs, policy terms, market benchmarks, claims history, and commissions behind it live across the agency management system, carrier portals, policy documents, the accounting ledger, and email threads. All of it feeds one thing, the advice you give this client, on this account, this year. That, he noted, is what the client is actually paying for.

The intelligence is not the hard part. The joining up is.

Chawla urged insurance leaders to spend as much time evaluating the infrastructure as the AI tools that sit on top of it. Without it, even the most advanced technology will struggle to deliver consistent business value.

2. Trust depends on safety

If infrastructure creates capability, safety creates trust. In a highly regulated industry, accuracy and oversight are not optional.

"There's no partial credit for safety." – Mohit Chawla

He pointed to the "complacency trap": as systems become more accurate, people become less vigilant. A tool that is right half the time keeps everyone alert. One that is right 90% of the time lulls them, and the 10% that is wrong slips through. Google saw the same pattern when it tested an early driver-assist system with its own employees: people quickly stopped watching the road. In insurance, those misses become compliance issues, financial losses, or errors and omissions (E&O) exposure.

So the answer isn't simply more accuracy. It's designing for vigilance. One check is a guess; independent checks are a system. At eventual, every match is checked against the document the carrier sent, the agency's books, and the money the bank cleared. If they agree, the work moves forward. If not, the agency's expert decides. And because a reviewer who sees everything reviews nothing, the system should escalate only the exceptions, with the context needed to decide.

One check is a guess. Three independent checks are a system.

• The document: what the carrier sent
• The record: what the agency's books show
• The money: what the bank actually cleared

All three agree? The work moves forward on its own. Any one disagrees? It stops, and your expert decides.

How eventual checks every match in agency accounting today.

One of the most important distinctions he drew was between intelligence and control.

"The model proposes, and the control layer acts." – Mohit Chawla

In other words, AI can recommend, but a separate set of rules and checks, the control layer, decides what the system can access, what it can do, and when a person must review its work. The goal is not to eliminate oversight. It's to make oversight more effective.

3. The future still belongs to humans

The right lens, in Chawla's view, is augmentation, not replacement: less time gathering information, and more time applying judgment, building relationships, and solving complex client problems.

He made the point with a striking comparison. By his rough estimates, running today's most advanced AI models takes about 5,000 watts, and a self-driving car's computers about 1,000. The human brain runs on 20.

Judgment, empathy, and creativity, he said, remain uniquely human strengths. In brokerage, that means reading a client's situation and making bespoke calls, because no two accounts are alike, and owning the decision when the rules run out. Technology's role is to support those strengths with accurate information and less administrative friction.

"The right way to think about these models is that they are creating new superpowers for us." – Mohit Chawla

Like self-driving, meaningful autonomy in insurance will not arrive overnight. It will be earned one level at a time, built on strong infrastructure, governed by thoughtful controls, and guided by human expertise. The first ride will feel unfamiliar. The hundredth will feel ordinary.

Watch Chawla's full presentation

‍