On November 5, a carrier did something no carrier had done before: it let independent agents quote a Business Owner's Policy without ever visiting its website. MGT Insurance listed Aimee, its AI quoting assistant, in the ChatGPT app directories. An appointed agent opens ChatGPT, describes a business in plain language or uploads an ACORD application, and gets a priced quote. Aimee classifies the business, checks MGT's appetite, asks only for what is missing, and — this is the detail that matters — tells the agent what she assumed and why. The quote then opens in MGT's portal, where the agent reviews it and binds coverage.
The scale behind it is not a demo. The platform has handled over $83 billion of quoted total insured value for nearly 15,000 appointed agents, and Aimee has been answering appetite questions on MGT's own site since 2025. This is a production system moving to where the agent already works, not a pilot looking for a user. "Agents shouldn't have to come to us," MGT's co-founder Graham Topol said. "We went to them."
Two weeks ago in this series, the shopper became an agent: consumer AI assistants calling carriers to request quotes and check claim status. Now the producer's assistant has arrived. The quote no longer happens on the carrier's portal, on the carrier's form, in the carrier's interface. It happens inside someone else's assistant, in a conversation the carrier does not control.
That is a distribution decision, and it is also a governance decision, and the two cannot be separated anymore.
What it means for carriers
First, the moat moved. For a decade, "easy to do business with" meant a better portal, a faster form, a cleaner API. MGT just showed that the next version is none of those things: it is being quotable where the work already happens. Every carrier that measures distribution friction in portal logins is measuring the wrong thing. The question is now whether your appetite, your pricing, and your authority can travel — intact — into an assistant you do not own.
Second, the design choices inside this announcement are the whole story, and they are worth copying. Three of them: the quote states its assumptions out loud instead of burying them; the system asks when it does not know instead of guessing; and the bind still happens in the carrier's portal, in front of a licensed agent who reviews it. That is agents pre-qualify, a licensed person confirms, the ledger keeps the tape — applied to distribution. It is also the direct answer to the warning Lockton issued three days earlier, that in agent-to-agent transactions "there is less opportunity for the human in the loop to one, exist in a meaningful way, or two, mitigate risk." MGT kept the human in the loop by keeping the bind in the loop. The quote can travel. The decision cannot.
Third, the accountability question is now open in a way no carrier can dodge. Zywave's new broker study notes that no insurer has yet emerged to underwrite an AI system's mistake at the point of bind. Read that alongside MGT: the industry is simultaneously putting quotes inside third-party assistants and admitting nobody will insure what the assistant gets wrong. That gap is the product opportunity and the governance obligation at once. The carrier that quotes through someone else's assistant owns every assumption that assistant makes on its behalf.
From the operator's chair
If you run distribution, underwriting, or operations, do three things this week.
One, map where your quote actually happens today. Not where your portal is — where the agent's work happens. If the answer is "in our system," ask how many of your agents' quotes start as a conversation in an assistant you have never seen. That is your real distribution surface now.
Two, write the four rules MGT got right into your own agent policy before your first quoting integration ships: the assistant states its assumptions; it asks instead of guessing; appetite is enforced by your rules, not the assistant's judgment; and no bind happens outside your system, in front of a licensed person, with a record. If a vendor or a platform cannot meet all four, it is a demo, not a channel.
Three, price the risk you are creating. Every quote your appetite generates inside someone else's assistant is a decision with your name on it and their interface around it. Your governance program, your vendor management, and your market-conduct file all need to say so explicitly — before an examiner asks, and before a wrong assumption becomes a bound policy with no one to sue.
The quote left the building. The accountability has to leave with it, on purpose, or it will arrive later as a surprise.
— Krishna