Newsletter · · Ashutosh Agarwal

Insurance AI Moves From Pilots to Five-Minute Quotes and an Agent-Native MGA - Vertical Spotlight: Insurance - Week of September 29, 2026

Startups and venture newsletter for the week of September 29, 2026. The Insurance vertical spotlight, where an agency using Kara quotes and binds a 150 dollar policy in under five minutes, Gaya cuts a home quote from an hour to 20 minutes across 600 carriers, Convex launched Kinetic as an AI-native MGA on Quotech, and underwriters debated whether AI makes pricing more disciplined or looser.

Vertical Spotlight: Insurance

Week of September 29, 2026: The Five-Minute Quote and the Agent-Native MGA

The rotation comes back to Insurance this week. The podcasts kept landing on the same idea: AI in insurance has moved past the pilot stage. It now shows up in the numbers: a $150 policy quoted and bound in under five minutes, a home quote cut from an hour to 20 minutes, a new MGA built from day one around AI agents. Not everyone thinks that is good news for underwriting discipline, though.

The Landscape

For two years, insurance people on podcasts talked about AI as something they were "experimenting with." This week the question changed. On The Voice of Insurance, host Mark Geoghegan put it to Everest's Jason Keen directly:

"If we'd been talking two years ago, we'd be saying, what are you experimenting with? And these days, well, you've experimented. Now, what have you implemented and where are you getting the best results?"

The answers came from every layer of the industry, and they were strikingly alike. Nobody described AI making the big judgment calls. Everyone described it getting rid of the paperwork around those calls:

  • At the agency level (the local shops that sell policies to households and small businesses), AI now answers the phone, collects the details an insurer needs to price a policy, fills in carrier websites, and puts together side-by-side quote comparisons. Florida Risk Partners says it now runs quoting "just like a factory."
  • At the carrier level (the insurers who actually take on the risk), Everest and Nationwide both described AI reading incoming submissions (the bundle of documents a broker sends when asking for a quote) and pulling out the data. At Nationwide, a single farm or agribusiness application can involve "50 documents, 24 data sources."
  • At the startup level, Convex's new MGA, Kinetic, is being built from scratch around an "agentic native factory." An MGA, or managing general agent, is a specialist firm that underwrites on an insurer's behalf. "Agentic" means AI that carries out multi-step tasks on its own instead of just answering questions.

Keen said it most clearly. The point of AI, he said, is to give "our underwriters and our claims experts more time to concentrate on judgment and not data assembly, which is not valuable." He remembers when the industry employed "data scrubbers who would scrub submissions that would come in for hours and there was hundreds of them." That job is the one going away.

The second theme was quieter but just as important: the software is becoming a layer that sits on top of the systems insurers already use, not a replacement for them. Gaya is a browser extension that works inside whatever agency system you already have. Kara writes straight into HubSpot. Nearmap feeds its aerial imagery into Guidewire so adjusters "don't have to go into another screen." Quotech's CEO predicts underwriters will soon reach his platform mainly through their company's AI chat assistant, not by logging in. The founders winning deals are the ones who fit into the existing workflow.

The third theme is the one to watch. As AI speeds everything up, some people are worried about what it does to pricing discipline in a market where prices are already falling. That argument is the Debate at the end of this issue.

Companies to Know

Kara (GetKara.ai): an AI receptionist that feeds the sales pipeline

Kara makes AI phone agents and workflow tools for insurance agencies. Its three co-founders ran their own digital brokerage for about three years, built the software for themselves, sold the brokerage to the McGowan Companies, and then turned the internal tool into a product. They have been selling it for about 18 months. Co-founder and CTO Nikhil Kansal joined Power Producers host David Carothers, who uses Kara at his own agency, Florida Risk Partners. Carothers gave the most detailed return-on-investment account of the week:

  • 24/7 intake. Kara answers every call, "24-7-365," works out whether the caller needs personal or business insurance, collects the underwriting details, and creates the contact, company, and deal record in HubSpot automatically. Someone calling at 9 PM on a Saturday gets the same automated follow-up as someone who fills in the website form.
  • The pressure-washer niche. Florida has a lot of pressure-washing businesses, and the insurer Coterie will cover them. Each policy earns only about $150 in commission, which used to make it "unprofitable for me every time I have to touch it." Now the time from the customer's call to "a bindable quote in their inbox or on their phone via text is under five minutes." The agency writes five to seven of these a day, which Carothers calls "a $300,000 plus revenue stream on one single vertical."
  • Payback. "I told you we've been with Kara for two and a half months. I have written enough premium in the commission derived from that that I have paid for Kara for over two years already."
  • Quote comparison. For home insurance, the agency runs a comparative rater (QuoteRush), cleans up the results, and then uses Kara to produce a branded PDF. It rates each option red, yellow, or green, flags errors-and-omissions risk (the risk of the agency being sued for a mistake), and recommends one. A typical example: a policy that costs $300 more a year but covers water damage in full, against cheaper ones that cap water damage at $10,000. Carothers says including the comparison "the close rate has gone up."
  • Customers don't mind talking to AI. Callers often ask partway through, "wait, am I talking to AI?" Kara says yes, and "they just finished the conversation with it."

A smaller detail worth noting: asking new leads how they would like to be contacted (email, text, or phone) "literally probably tripled the engagement." Carothers credits his oldest son, who runs personal lines at the agency.

(Power Producers Podcast, "Building an AI-Powered Insurance Agency with Nikhil Kansal," September 23, 2026)

Gaya: autofill for carrier websites, heading toward "agentic book rolls"

Gaya is a browser extension that fills in insurer quoting websites automatically. Co-founder Carl Ziadé says it now works with 600 carriers (up from roughly 300 about six months ago, according to host Jason Cass). The pitch is simple. Cass says a typical auto-and-home quote takes "an hour without the phone ringing"; with Gaya, "you can do it in 20 minutes." Customers report "an ROI of 2 to 3x" on quoting speed. What to know:

  • Why it exists. Comparative raters are tools that pull prices from many insurers at once. They return a "rate call one" (RC1) price, based only on what the agent typed in. The real, bindable "rate call three" (RC3) price comes after the insurer checks driving records and claims history. Some carriers, Ziadé says, deliberately show an aggressive RC1 to land at the top of the list. So more producers now skip the rater and go straight to each carrier's website, which is exactly the work Gaya automates. It is especially popular for coastal carriers and high-net-worth insurers (Pure, Chubb, Cincinnati and others) that raters handle poorly.
  • Integrations. Live with HawkSoft; Momentum going live; an Applied Epic SDK integration built with the regional broker Heffernan Insurance; and a native "Send to Gaya" button in AMS360.
  • What's next. A proposal generator where agencies can "bring your own key" to Gemini or Claude, a "submission OS," partnerships that would let Gaya act as a rater, and, in about six months, tools that let carriers run "book rolls" (moving an agency's whole book of policies from one insurer to another). For example, "Travelers might be using Gaya to book roll your book with them in a day."
  • The carrier angle. Ziadé says carriers are "a bit scared" of agentic quoting. It reminds them of the old screen-scraping bots from before insurers built proper data connections, because "it's actually sitting on your machine and Claude doing the work." AI lets an agent send a quote request to every carrier at once, which "reduc[es] intentionality." Gaya's answer is to build the guardrails with the carriers, and Ziadé says "a lot of the biggest referrals we're receiving is from carrier reps."
  • Funding stance. Ziadé stressed that Gaya is profitable and "not addicted to VC money." He thinks that helps with enterprise buyers who expect an AI bubble to burst: "knowing that we will not be swimming naked, people feel very good about it."

(Agency Intelligence, "Carl Ziadé On What's Next For Gaya, From Proposals To Agentic Quoting," September 22, 2026)

Kinetic Insurance Services (Convex) + Quotech: an MGA built around AI agents from day one

Convex, the Bermuda and London specialty insurer, launched Kinetic in June 2026. It is a hybrid: part insurer, part MGA platform. Entrepreneurial underwriters join as separate "cells," get Convex's capacity plus outside capacity, and get a ready-made technology stack. CEO Theo Butt, CTO Andy Roberts, and Quotech CEO Guillaume Bonnissent explained how it works on InsTech:

  • The build. Roberts spent nine months setting up the core systems. Kinetic worked with AWS and the systems integrator Prefectus "to help us to build an agentic native factory," covering "the first email coming through to generating our quotes with the underwriter very much in control." The goal: "almost all of our non-revenue generating work is automated." On top of that, a small team builds "superchargers," meaning automations for "all those annoying little repeatable tasks" the underwriters ask for.
  • Why Quotech. Kinetic wanted a policy administration system (the core record-keeping software) that runs "from underwriting through to the billing collections piece, and not just a shiny underwriting set of screens." It also needed to be API-enabled and "moving towards MCP enablement." MCP, or Model Context Protocol, is the standard that lets AI assistants connect to outside software. Bonnissent says Quotech has cut the time to set up the platform for a new underwriting team from the usual "three to six months" to "a matter of days or weeks."
  • Milestones. The first underwriter joined in Bermuda in early August to set up an emerging-industries cell. Kinetic is aiming for "three or four cells" live by the January 1, 2027 renewals, and "at least two or three more cells" during 2027. Convex itself is only about 480 people.
  • Bonnissent's forecast. Underwriters will start their day in a company-approved AI chat window that, through MCP, shows them overnight submissions, pending referrals, and account queries. "Is it in three months? Is it in six months? No, it's actually doable today." The real obstacle, he says, "is adoption rather than where the technology is at," along with controlling costs so it doesn't become a big bill "because every single user is vibe underwriting, vibe coding."

He also looked back 20 years to a project at a previous employer called "Underwriters to Underwrite," which "never really worked because the technology wasn't there." Kinetic is essentially the same idea, attempted again now that the technology exists.

(InsTech, "Theo Butt & Andy Roberts: Kinetic Insurance Services and Guillaume Bonnissent: Quotech: Building an AI-native MGA from scratch: inside Kinetic Insurance Services (419)," September 27, 2026)

Everest (Global Wholesale & Specialty): AI in four business segments at a ~$4B book

Jason Keen is CEO of Everest's non-treaty specialty business: close to $4 billion in premium, 26 product lines, about 1,000 staff, and offices in Warren (New Jersey), London, and Singapore, with premium split about 60% US and 40% international. It is the largest operation in this issue that described specific AI deployments:

  • Underwriting: AI is "embedded in our underwriting workflow. Submission, ingestion and turnaround to quotation time. We've seen significant acceleration." Keen says time is the main measure, but "accuracy is another one because of the consistency." It is now live in "four segments of our business."
  • Contracts and claims: "We are using it for our reinsurance contracts, extracting data and first notice of loss." (First notice of loss is the first report of a claim.)
  • Headcount: He avoided a direct answer. "Whether headcount goes up or down, ultimately what we want our core talent... to be doing is spending their time on value-added work." He did note that AI "also manages expense, which in any cycle should be something that's atop people's minds."

Keen also spoke about AI as a risk Everest insures. He called it "silent AI," meaning AI exposure buried inside existing policies without being named, much like "silent cyber" before cyber became its own product. His view: "It's either got to be excluded or included."

(The Voice of Insurance, "Ep317 Jason Keen CEO Wholesale & Specialty Everest Insurance: No time to be stuck in the middle," September 22, 2026)

Nationwide: thousands of agents and 20% of a $1.5B budget

On Technovation, Nationwide CTO Michael Carrel described AI at the scale of a large insurer. About 20% of a $1.5 billion technology investment goes to AI, split into three buckets:

  • "Everyday AI": search, summarizing, content creation, and the ability to build personal agents, rolled out to every employee.
  • "Flagship AI": larger projects aimed at underwriting, claims, and post-issue policy changes, mostly about making sense of unstructured documents. In large farm and agribusiness underwriting, one application can involve "50 documents, 24 data sources, you know, hundreds of data fields." The new approach combines traditional predictive models with generative and agentic AI to cut the time needed.
  • AI-assisted software development for every technology employee.

Nationwide has built "multiple thousands" of custom agents. Its summer hackathon drew 172 teams and nearly 900 employees, 128 of them from outside the tech department. Employee sentiment on AI rose 9% in its twice-yearly survey. Every AI use case goes through the same risk-committee review Nationwide has used for "15 years since we've been doing AI models." Carrel says the process is meant "to get quick yeses, quick nos."

(Technovation with Peter High, "How Nationwide CTO Michael Carrel Is Building an AI-Ready Workforce," September 24, 2026)

Mantas: parametric insurance for when the cloud goes down

Basil Mimi is a software engineer turned insurer. He came up with Mantas after a food-delivery app crashed while his office was ordering lunch. His thought was that if an outage like that hit his own company, "we would also lose a lot of money." You cannot control Amazon or Google, so he looked at insurance. That led him to parametric cover, which pays out automatically when a measurable event happens instead of after a loss adjuster assesses the damage. Mantas is based in the Dubai International Financial Centre. It waited a year for its license, received it in January, and has now started selling.

  • Underwriting approach: Mantas built two tools, one that monitors cloud and data-center risks in real time, and one that analyzes "all the historical events that happened to data centers... for over 15 years." These feed its expected-loss and catastrophe modeling.
  • Market: Mimi says 67% of UAE businesses depend critically on the cloud. Mantas targets mid-size and large businesses through retail brokers, with a minimum premium of about $10,000. To show the stakes, he pointed to a 15-hour AWS outage "back in November" and a delivery app processing "20,000 orders per minute" at $30–40 each.
  • Positioning: It is designed to sit alongside cyber insurance, not replace it. Cyber cover requires a malicious attack; Mantas pays out whatever caused the outage. More digital-risk products are in development.

Host Joey Kaplan's aside says a lot about the market: "every other interview I do is an AI founder that claims to be using AI to revolutionize insurance. And it's getting very hard for me to tell them apart." Mantas stood out to him because it is new coverage, not new software.

(Profiles in Risk, "Basil Mimi, Co-Founder & CEO at Mantas - Ep. 865," September 23, 2026)

Comotion AI: "Nobody buys AI. They buy a migration that finishes."

Tim Vieyra is an actuary by training. He co-founded Comotion 13 years ago as a data and actuarial company. Its new AI platform lets actuaries and analysts ask questions of insurance, finance, and mortgage data in plain English, but the product it actually sells is narrower: data migration. That means moving a book of business from an old system to a new one, or into an acquirer's systems after a deal (for example, pension risk transfer, where an insurer takes over a company's pension obligations).

  • The problem: On traditional migrations, teams "discover things about the source data quite late," usually only once reports run in the new system. Dan Griffith, who is advising Comotion on entering the US market, told a story about a client project called "On Time" that had to be renamed "About Time."
  • The fix: Comotion analyzes the source data as soon as it arrives, so that "on, say, week two of receiving data or week one... you can say, hey, guys, here's your data issues." Vieyra says this turns data "from the longest pole in the tent to the shortest pole in the tent." The pricing is fixed-cost, and it can run inside the client's own systems with "forward-deployed engineers" (Palantir's term for engineers placed on-site with the customer).
  • A lesson for founders: Griffith said the difference between AI vendors that succeed and those that fail is being "tied to some very specific use cases." Vieyra's version: "while the unit cost of software is going down... the value of the data is going up."

(Insurtech Leadership Podcast, "Nobody Buys AI. They Buy a Migration That Finishes," September 24, 2026)

Send, now part of The Creek: buying working AI rather than piloting it

Celine Thierry runs the reinsurance, compliance, payments, and loss-control products at the insurance-software company she calls "The Creek," a vendor that has been around "almost 30 years." Speaking at the Monte Carlo reinsurance meetings, she explained why it bought Send, an AI-first underwriting platform. Clients have seen "lots of POCs, lots of experience. And they have the feeling that it's not really there." (A POC, or proof of concept, is a small trial project.) So the company acquired something already working "to bring tangible things directly into place." The plan is to connect underwriting, reinsurance, loss control, and exposure data so insurers can see "the one that are showing some difficult trend" instead of drowning in data. Her read on where things stand: "the US market is really ready to adopt AI. I have a feeling that on Europe, it's a bit early."

(The Reinsurance Podcast, "Monte Carlo #65 - Celine Thierry: What Actually Works With AI in Insurance," September 22, 2026)

Nearmap: aerial imagery wired into AI claims workflows

David Tobias is Chief Product Officer at Nearmap, which merged with the insurance analytics firm Betterview. Nearmap now pairs its "hyper-up-to-date, hyper-accurate imagery" with AI that classifies damage, and delivers it inside Guidewire and other core systems. The use case Tobias is most excited about: when a customer calls in a roof claim and talks to an AI voice agent, Nearmap's property data can flow into that conversation in real time. That could "shorten the time from FNOL to payment." His summary: "the technology is kind of caught up with the data."

(FNO: InsureTech, "Ep 315: David Tobias, Chief Product Officer, Nearmap," September 25, 2026)

Also worth a line

On The Reinsurance Podcast, reinsurance brokers Carla Moffett and Sam Sweeney said their firm uses AI to build renewal submissions, process incoming data, and review contracts, and uses its Sage analytics platform to advise clients. "I would argue we're already AI-enabled." On AI liability, Moffett says reinsurers are not rushing to add blanket exclusions, because "silent AI" is already inside professional liability, D&O, and general liability books.

(The Reinsurance Podcast, "Monte Carlo #70 - Carla Moffett & Sam Sweeney: Casualty Isn't Softening Like Property," September 25, 2026)

One Debate: Will AI make underwriters more disciplined, or give them a better excuse to cut prices?

The backdrop is a softening market: prices are falling. This week's roundup episode from Monte Carlo on The Voice of Insurance was full of underwriters calling property price cuts "disappointing," and Everest's reinsurance CEO Jill Beggs preferring the word "normalization." The obvious question is whether AI helps the industry hold its nerve or helps it slide.

The optimistic case came from Jason Keen. His argument: underwriters now have "30 years more of data" than their predecessors, and "if people are using data the way they should be, it's less of a subjective view, it's much more data driven, which the numbers do the talking." That produces benchmarks that are "undeniable, really." On this view, AI and better data make it harder to fool yourself about whether a price is actually adequate.

The skeptical case came from James Slaughter of Apollo, in the same roundup:

"Everyone will talk about AI data and all that. That's great. But actually how it's being used, what we're seeing in some of the markets suggests people are either ignoring the facts and the data that's presented to them, or that they found some magical AI tool that allows them to write below adequacy, which is possible. I mean, if you ask Claude, Copilot, and ChatGPT the same question, you'll get three different answers."

In plain terms: if the model can be persuaded to justify a lower price, a lower price will get written. Host Mark Geoghegan tied it to a bigger puzzle, suggesting this "perhaps explains why the market softened so much faster than many expected."

Slaughter does see a positive version of the same story. With all this data, "it's difficult for a CEO to sit there with all of that data in front of them, and not reverse trend." His prediction is that the cycles will not go away but will speed up: "we'll come off quicker, but we should spot the trend when it gets sub-adequate, and we should react more quickly."

The same tension showed up at the agency level. Carl Ziadé of Gaya said carriers are uneasy that agentic quoting lets agents send a request to every insurer at once. Before, the effort involved forced an agent to decide where each piece of business should go. So one side of the market worries AI makes pricing too loose, and the other worries it makes shopping too easy. In both cases, AI removes friction that used to act as a natural brake.

Geoghegan's conclusion is the right place to stop: "The only thing that is certain is that now that AI is with us in our underwriting decision-making processes, this is a question that's now going to run and run." If you are building or investing here, there is a real opportunity in whichever company can show its AI makes pricing decisions more consistent and auditable, and not only faster. Keen used exactly that word when he listed "accuracy... because of the consistency" as Everest's second metric.

(The Voice of Insurance, "SpEp The State of (Re)insurance 2026," September 25, 2026; The Voice of Insurance, "Ep317 Jason Keen CEO Wholesale & Specialty Everest Insurance: No time to be stuck in the middle," September 22, 2026; Agency Intelligence, "Carl Ziadé On What's Next For Gaya, From Proposals To Agentic Quoting," September 22, 2026)

Next week's Vertical Spotlight: Fintech.