The automated voice on the other end of the line has been flawless. It understood your HVAC needs, found an available time slot, negotiated a service quote, even upsold you on a maintenance plan. Then you say those four magic words: "I'm ready to pay."
Everything stops.
The AI agent—so smooth just seconds ago—can't actually process your credit card. Not without either plunging the company into the byzantine world of PCI compliance audits, or worse, emailing you a link and hoping you'll click through later. That momentum, the entire premise that made conversational AI useful? Gone.
This is the problem Maven, a fresh-out-of-Y-Combinator startup, thinks it can solve.
Launched last week from San Francisco, the company offers what its founders call a "single-API solution" to in-call payments. Voice AI agents can now hand off callers mid-conversation to Maven's secure line, where customers punch in their card details via phone keypad, then get transferred back automatically once the transaction clears. The whole thing happens without the original company touching payment data—or inheriting the compliance headaches that come with it.
A Compliance Chokepoint Nobody Wants to Talk About
It's a technical challenge that doesn't announce itself loudly, but it's real enough. Voice AI platforms like VAPI, Retell, and LiveKit have gotten impressively good at natural language and conversational flow. They can handle objections, reschedule appointments, even detect frustration in a caller's voice.
But payment data? That's different. The moment credit card numbers enter the picture, companies hit a wall: build your own PCI-compliant infrastructure (a months-long slog through payment gateways, telephony handoffs, and compliance audits) or abandon the voice channel entirely when it's time to close the deal.
"When a customer says 'I'm ready to pay,' everything breaks," Maven's website states plainly.
The company's founders saw this pattern up close. Wasi Ahmed and Brandon Boehme—both UC Berkeley computer science grads with stints at Amazon and Meta on their résumés—watched as voice agents excelled at qualification and scheduling, yet repeatedly fumbled at the finish line. Not because the AI wasn't capable, but because the infrastructure around it wasn't designed for this moment.
Maven abstracts the messy parts. Developers create a payment session with a single API call containing the transaction amount and the caller's phone number. Maven returns a transfer number (plus an optional SMS checkout link as backup). During collection, card data routes through Maven's infrastructure, never touching the customer's servers. Real-time webhooks deliver status updates. When payment completes, the caller returns to their original agent automatically.
Three steps. That's the pitch, anyway.
Not Reinventing the Wheel, Exactly
Maven integrates with Stripe, Authorize.net, Adyen, and Braintree—meaning companies can keep their existing payment processor relationships rather than switching to a new merchant of record. That multi-gateway approach sets it apart from single-PSP integrations common in voice platforms, at least according to the company.
It also spans multiple telephony stacks: VAPI's transferCall tool, Retell's routing, LiveKit's SIP bridge, and Twilio's TwiML dial commands all work with the same Maven session API.
The concept isn't entirely new. Twilio launched its Pay product years ago for IVR and agent-assisted payments, handling DTMF capture and PCI descoping for contact centers. Companies like PCI Pal and Key IVR have built entire businesses around the "transfer to secure line, collect payment, reconnect" workflow.
Maven's bet is that AI agents require something different. Not IVR menu trees, but session-based APIs tuned to conversational state. And not just telephony abstraction, but payment gateway abstraction too—in a single integration.
Whether that's enough differentiation remains to be seen.
The Agentic Commerce Wave (Or Is It Hype?)
The timing might be right. Payment networks are preparing—at least in theory—for agent-initiated transactions. Visa announced its Intelligent Commerce initiative and Trusted Agent Protocol in 2025, claiming hundreds of secure agent transactions already completed and projecting mainstream adoption throughout 2026. Mastercard followed with its AgentPay framework. Last December, Visa and AWS announced blueprints for agentic commerce tooling through AWS Marketplace and Bedrock AgentCore.
Checkout.com CEO Guillaume Pousaz wrote in his 2025 annual letter that multiple agentic protocols would likely emerge, with processors eventually supporting Visa VIC, Mastercard AgentPay, and perhaps Google's Universal Commerce Protocol. PayOS emerged from stealth in April 2025 with partnerships across those network initiatives, branding itself as "card-native payments infrastructure for AI agents." New Gen launched AI-native storefronts with embedded payments. Oceanpayment announced its own AI Agent Payment Solution in December.
Maven operates one layer below these protocol implementations—handling the immediate, tactical problem of in-call card collection while the industry argues over authentication standards and tokenization flows. It's infrastructure for the near term, not the theoretical future.
The company's Launch YC post specifically requests introductions to home services companies (HVAC, plumbing, roofing), healthcare billing and scheduling systems, insurance agencies, collections firms, auto services, and SMB sales teams. All sectors where voice AI adoption is accelerating but where payment completion remains stubbornly manual.
What We Don't Know Yet
Y Combinator partner Ankit Gupta endorsed the launch on LinkedIn, praising the team's ability to deliver "compliant payments with just a few lines of code." The company's website includes API examples, flow diagrams, and integration tiles for supported platforms, emphasizing "minutes not months" implementation and "zero PCI burden."
But details are thin in other areas.
The corporate entity is Lambda Systems, Inc., founded in 2025 with a two-person team. Maven's site doesn't list pricing. There's no mention of specific PCI DSS certification levels or third-party audit results, though the company claims card data flows entirely through its infrastructure without touching customer servers. (How comfortable you are with that claim likely depends on your risk tolerance and legal team's disposition.)
Narrowing the Gap
What Maven represents—perhaps more than its specific technical implementation—is the infrastructure layer forming beneath conversational commerce. Not the sexy AI agent that charms customers into upgrades, but the plumbing underneath that actually completes the transaction.
Voice AI agents proved they could handle complex negotiations and qualification. The scripts got good. The natural language processing got eerily human. But they kept hitting the same wall when it came time to get paid.
Now payment rails are adapting, treating those agents as first-class transaction initiators rather than forcing them through checkout experiences designed for humans clicking buttons in browsers. Whether Maven becomes a verb or a footnote, the gap between "booking" and "billing" is narrowing.
One API call at a time.
