From Blockade to Protocol: GEICO and Liberty Mutual Join the Poppy Agent Standard as William Blair Calls the Muse Panic Overblown
The AI-agent distribution threat to personal-auto insurance is moving from panic to plumbing — and the carriers are joining the standards effort rather than only blocking it.
The Poppy protocol. Sierra (Bret Taylor) published a draft of the Personal Agent Protocol ("Poppy") whose founding announcement included Meta, Rocket, Shopify, Stripe and Walmart, plus 35 design partners that now include GEICO, Liberty Mutual, Insurify, Bank of America, Wells Fargo, Visa, Mastercard, PayPal, Target, United Airlines and OpenAI. The protocol's premise: "customers decide what access to give their personal agents, and companies set parameters for what those agents can do" — agents identify themselves via a /.well-known/poppy.json file, customers sign in through OAuth, and companies choose whether the agent works through their website, APIs, or their own agent. Critically for the insurance fear, the live demo at Sierra Summit featured Meta's Muse transacting at a financial-services company: "Alex asked Muse to get pre-approved for a mortgage. Muse went to Rocket's website, found Rocket's agent through Poppy, and asked Alex for permission to connect... While he was in control at every step of the way, the agents did all the work."
Insurify's pivot. The same marketplace that blocked Muse on Sept 23 joined Poppy as a design partner on Oct 9. Insurify co-CEO Giorgos Zacharia: "AI agents are an important distribution channel for carriers, and we're helping shape standards that keep consumers in control and give companies the ability to protect the customers they serve." The PR also discloses that "Insurify already offers personalized car insurance quotes through its ChatGPT plugin, with coverage details and a path to purchase through a licensed agent or insurer" — authorized agent-mediated quoting already exists at a major marketplace, predating the Muse panic.
The institutional counter-thesis. William Blair's Oct 8 "On Risk" discussion with The Zebra's chief commercial officer David Seider (reported by PYMNTS Oct 9) argues the selloff — "stocks of those companies have fallen about 15% on average since early September1" — overstates the threat. Three hurdles: (1) cost — "a real quote that a customer can buy is not [cheap]... That data can cost up to $20 per quote," and "If AI agents were to continuously shop for coverage, carriers would face a meaningful increase in customer acquisition costs"; (2) compliance — disclosure obligations and tightly regulated commissions "may keep a licensed human involved in the process for some time"; (3) trust — brand still decides. William Blair's bottom line: "AI will change how people shop for insurance. But a major shakeup of the companies that sell it is unlikely anytime soon." It expects online insurance market growth to slow from the ~30% pace of H1 2026 into 2027.
What it means for the thesis: the tripwire this topic watches — a carrier or major marketplace granting an AI agent authorized live-quote access — has not fired. Poppy is a draft; Sierra says "over the next month we will host design workshops and publish a reference implementation." But the direction is now legible: GEICO and Liberty Mutual are design partners on the very rails that would let Muse-class agents transact, and the Muse-Rocket mortgage demo shows the flow works end-to-end in regulated financial products. The -15% sector repricing now has both an institutional counter-thesis (William Blair) and a standards path that could restore agent access on carrier terms. The distribution vector hasn't gone away — it's being formalized. See NHTSA Gives Tesla 30 Days, Not 8 Weeks: Cybercab Sworn Answers Due Oct 30 — Nine Days After Q3 Earnings for the parallel liability-side regulatory track.
-
An instance of The AI disruption discount has crossed onto the auto insurers themselves. — This is the same AI-agent panic discount the theme tracks — insurers marked down for rate-shopping agents before any claims deterioration — now quantified sector-wide and met by the first institutional counter-argument that it overshoots. ↩︎