Consumers' AI Agents Are Coming for the Inertia Tax — and Companies Don't Agree on How to Respond

For a certain class of business — insurers, airlines, broadband providers, gyms, subscription services — a meaningful share of margin has always come from customers who don't call to negotiate, don't shop their renewal, and don't cancel the free trial on time. That inertia is now acquiring a predator. Over the past six weeks, consumer-facing AI agents have moved from demo-stage curiosities to apps doing close to a billion dollars in annualized transaction volume, and the companies on the other end of those calls are responding in incompatible ways — some blocking the agents outright, others opening the door and hoping to out-automate them.

The volume is no longer hypothetical

Instinct, the agent startup founded by 23-year-old Noah Shinn, raised $250 million at a $2.5 billion valuation in August and then $1 billion at a $10 billion valuation barely a month later, with Shinn telling reporters the app is approaching $1 billion in annualized transaction volume — travel bookings account for roughly half. Meta's rival agent, Muse, became the top free app on the U.S. App Store within days of its September 8 launch, pulling in 2.8 million installs in its first twelve days, according to Apptopia data. These aren't pilot numbers. Individual users are reporting concrete wins: a Datadog engineer used ChatGPT to talk Verizon down from $94.99 to $69.99 a month plus a loyalty credit; a Merit Systems employee used Claude Code and a phone-calling tool to recover $73 in United Airlines travel credits across three flights, spending about five minutes on each call. None of these are large sums individually. The aggregate is what should concern a CFO.

Scale that logic to an industry. US auto and homeowners insurance premiums ran roughly $560 billion in 2025. If even 5% of policyholders used an agent to shop their renewal and achieved a 10% price cut — a scenario, not a forecast — that's on the order of $2.8 billion in annual premium at stake industry-wide. Wayve Capital Management's Rhys Williams has argued that 5–10% adoption wouldn't meaningfully move insurer economics, but 40–50% would. The market isn't waiting to find out: insurance stocks fell roughly 15% on average in the weeks after Muse's launch, and Airbnb, Booking.com, and Planet Fitness all dropped double digits in the same window — 12%, 10%, and 17% respectively — on what analysts describe as a thesis trade about inertia-dependent business models, not yet a confirmed earnings impact.

Two incompatible corporate responses, in the open

The clearest test case is Amazon versus Meta. Starting September 20, Amazon blocked Muse from shopping on Amazon.com, telling users that "access by an unauthorized AI agent violates Amazon's Conditions of Use." Amazon's stated objections: Meta never disclosed the integration, Muse doesn't identify itself as an AI agent while browsing, and it can reach customer account information in ways Amazon says it never authorized. Meta's existing Muse documentation counters that the agent can't see passwords or stored payment credentials, that a separate "Sentinel" agent reviews actions before they execute, and that purchases route through a Stripe-generated one-time card number. Both can be true — Meta's technical safeguards and Amazon's objection to being integrated without a direct relationship are not mutually exclusive, which is exactly the governance problem: there is no settled standard for what counts as authorized agent access to a commercial platform.

Shopify took the opposite position in the same window, adding Meta as an AI commerce channel and expanding Muse support through Shop Pay. So within the same month, one major retail platform blocked a leading consumer agent and another built a direct integration with it. For any enterprise relying on a browser-based or unofficial agent integration for procurement or customer acquisition, the lesson is that access is a revocable privilege set by terms of service, not a stable channel — Amazon's parallel, partially successful fight to restrict Perplexity's shopping agent (a preliminary injunction against Amazon was vacated by the Ninth Circuit on August 4) shows platforms retain real leverage here even when they lose a specific court round.

The new insurance customer is increasingly an AI agent acting on someone else's behalf.

That's Accenture's global insurance lead, Ravi Malhotra, summarizing survey results that should worry any insurer still pricing against human shopping behavior: in a poll of more than 25,000 consumers across 16 countries, 72% expect AI to influence at least half of their insurance shopping within a year, 82% already use generative AI somewhere in the process, and a third say they'd let a personal AI assistant handle claims-related tasks directly. Only 8% want a fully automated process with no human oversight at all — the opening isn't full automation, it's AI-assisted comparison shopping at scale, which is cheaper to build defenses against than a wholesale replacement of the sales channel, but still expensive.

Why insurers specifically can't just block their way out

Retailers can block an agent at the access layer because a product listing is a product listing regardless of who's reading it. Insurance is structurally harder, according to William Blair analysts who spoke with David Seider, chief commercial officer of online insurance marketplace The Zebra, in early October. A rough estimate is cheap to generate; a firm, bindable quote requires the insurer to purchase third-party data — driving records, among others — that can run up to $20 per quote. An agent that shops continuously on a consumer's behalf, rather than once at renewal, could drive customer acquisition costs up meaningfully industry-wide, which argues for insurers controlling the intake channel rather than letting any agent request firm quotes at will. Layered on top is a compliance problem with no clean precedent: insurance commissions are tightly regulated, and current rules were written assuming a licensed human handles disclosure. When an AI agent conducts the conversation, the insurer has no reliable way to verify what the consumer actually heard or understood — a liability question regulators haven't addressed and carriers can't simply engineer around with an API key allowlist.

The response taking shape across insurance and contact-center operators isn't blocking — it's counter-deployment. J.D. Power's insurance-intelligence director, Tony Soloman, frames the endgame as carriers building their own AI to proactively retain high-value customers before an outside agent gets to them, an outcome he calls, bluntly, AI-versus-AI. That reframes the adoption decision facing an insurance or telecom CIO: the question is no longer whether to deploy a retention agent, but whether to deploy one before your customers' agents start probing your renewal pricing at scale.

The metrics enterprises run on are already wrong

Contact-center operators are the ones feeling this first, because they're the ones answering the phone. TTEC and Alorica staff report fielding a rising volume of calls that behave strangely — scripted cadence, repeated callbacks, failed security questions from numbers unlinked to the account — and quality-assurance systems built to score human empathy are penalizing or mis-scoring bot interactions in ways that have required manual override. Average handle time, containment rate, and satisfaction scores all degrade in reliability once a meaningful share of calls originate from an agent rather than a person, and those are the exact metrics enterprises use to staff contact centers and negotiate BPO contracts. Regulatory tailwinds compound the pressure to get this right rather than simply block it: the FTC has already extracted more than $845 million in refunds from Amazon over Prime cancellation friction, forced Shutterstock into a $35 million settlement over subscription-cancellation practices, and pushed Equinox into a $600,000 payment to New York's attorney general for the same reason. Regulators are already penalizing the inertia tax directly — AI agents are simply automating the complaint regulators were already encouraging consumers to file.

The decision in front of enterprise leadership isn't whether personal AI agents are real — the transaction volume answers that. It's whether to compete for access control at the platform layer like Amazon, open the channel and compete on service like Shopify, or absorb the compliance risk and race to build a defensive agent before the opposing side's agent calls first. Different industries are already choosing differently, within the same month, and none of the three paths has a track record yet.