There's a certain logic to testing conversational AI the same way you'd test a self-driving car. Both involve complex systems making split-second decisions. Both need to handle edge cases—the unexpected swerve, the garbled phrase. And both, ideally, should fail in simulation rather than in front of actual humans.
That's the thesis behind Coval, a San Francisco startup that just closed a $28 million Series A led by Norwest. The round, which was announced June 24, 2026, included Base10 Partners, Twilio Ventures, and Y Combinator, lifting the company's total funding to approximately $31.3 million since it emerged from YC's Summer 2024 batch.
Founder Brooke Hopkins knows the playbook well. She spent years at Waymo, Alphabet's autonomous vehicle unit, building the infrastructure that let engineers evaluate self-driving systems at scale—running scenario after scenario before a single car touched pavement. Now she's applying that methodology to voice agents, the AI systems increasingly answering customer service calls, booking appointments, and handling the kind of phone interactions that used to require an actual person on the other end.
The problem Coval targets is deceptively simple: How do you know your voice agent won't buckle when a customer calls from a subway platform, speaks with a heavy accent, or interrupts halfway through a sentence? Manual testing doesn't scale. Neither does deploying an agent and hoping for the best.
So Coval simulates. Tens of millions of scenarios, Hopkins says, involving background noise, regional dialects, conversational detours, and the sort of edge cases that make engineers nervous. The platform monitors production calls, flags issues for human review, and provides audio-specific metrics—the kind of systematic validation regime that autonomous vehicle teams take for granted but voice AI developers have largely cobbled together on their own.
"Every company is going to have a voice agent just like they have a mobile app or a web app," Hopkins said in the funding announcement. It's a bold claim, though not an implausible one given how quickly enterprises are adopting conversational interfaces.
The company counts over 60 customers, including Zoom and Deepgram, plus what it describes as Fortune 500 enterprises it can't name. According to Tech Funding News, Coval's revenue has grown 10x year-over-year based on company-provided data, though the startup hasn't disclosed actual ARR figures. Company-provided metrics suggest the platform cuts manual QA processes by up to 30x and accelerates deployments by 10x—numbers that, like most startup benchmarks, carry the usual caveats about sample size and customer mix.
Capital for Scale, and Competition
The fresh funding will go toward expanding sales and solutions engineering teams, a typical Series A move for an enterprise software company gaining traction. Coval also plans to deepen its simulation capabilities and add integrations with more voice AI platforms, building out the connective tissue that makes enterprise software sticky.
The competitive landscape is still forming. Hamming focuses on regulatory edge cases for voice agents, while Roark—a YC Winter 2025 company—emphasizes production call replay at scale. Coval positions itself as broader: a full-stack platform handling pre-deployment testing, production monitoring, and labeling workflows, all optimized for audio rather than text-based chatbots.
Norwest partner Scott Beechuk pointed to Hopkins' Waymo background as a key factor in the investment decision. "Voice is becoming the number one interface for AI," he noted. Twilio Ventures VP Andy O'Dower called testing and observability "the foundation for enterprise voice AI adoption"—the kind of infrastructure-layer bet venture firms often chase.
Market projections vary wildly, as they tend to with emerging categories. Grand View Research estimates the AI voice agents market will balloon from $3.5 billion in 2026 to $35.2 billion by 2033, though such forecasts depend on assumptions about adoption curves that haven't fully materialized yet. What seems clearer is that enterprises deploying these systems face real pressure to ensure reliability. Zoom's Head of Product for CX AI, Ram Rajagopalan, called reliability and observability "a top priority" in Coval's announcement—the kind of public validation that matters when selling to cautious IT departments.
Coval launched publicly in October 2024, then raised a $3.3 million seed round in January 2025 led by MaC Venture Capital with Y Combinator and General Catalyst participating. The company has since announced partnerships including Zoom's ISV Exchange program and integrations with voice platforms like Pipecat, Retell, and Rime—building the ecosystem connections that enterprise buyers expect.
Whether Hopkins' Waymo-inspired approach becomes the industry standard for voice AI testing remains to be seen. But the funding signals investor confidence that systematic simulation beats ad hoc testing, and that the lessons learned from autonomous vehicles might transfer to conversational AI more smoothly than anyone expected. Sometimes the best way forward is to borrow a roadmap from another industry entirely.
