Somewhere between the Mixpanel dashboard and the customer interview transcript, most product insights die a quiet death. The analytics show a drop-off. The team theorizes. Someone schedules user calls for three weeks out. By the time the research concludes, the engineers have moved on to the next sprint.
Matthew Wong and Mojmir Horvath think they've found a shortcut.
Their company, Poth Labs—a two-person operation from Y Combinator's Summer 2026 batch—wants to automate that entire investigative loop with AI agents that don't just analyze data, but actively hunt for explanations. The platform connects to whatever information a company already collects, forms theories about what's going wrong, then validates those theories on its own. No product manager required to connect the dots.
It's an ambitious premise. Whether it actually works remains an open question.
The San Francisco startup's pitch centers on what it calls "truth-seeking AI for customer feedback." Feed the system your customer tickets, employee schedules, inventory logs, supplier records—the operational detritus that most companies generate but rarely synthesize—and it will, theoretically, tell you not just that something broke, but precisely why.
The company's website demonstrates the general investigative process the AI follows, showing how it can test multiple hypotheses and trace problems through various data sources, presenting its findings with statistical confidence and rapid turnaround times.
That kind of statistical rigor, assuming it holds up under real-world conditions, would represent a meaningful shift from how most teams currently operate. Traditional analytics platforms excel at identifying patterns—users abandon the checkout flow here, engagement drops there—but they stop short of explaining causation. Research tools, meanwhile, tend to start with predetermined questions, which means they're only as smart as the person writing the survey.
Poth's approach, at least in theory, blends both. The platform begins with behavioral data to generate hypotheses, then validates them through what the company describes as "adaptive interviews" that tailor follow-up questions based on previous responses. Those conversations feed into a knowledge graph that agents can navigate to surface patterns across multiple touchpoints.
It's adaptive analytics meets autonomous research. Or that's the idea, anyway.
Crowded Territory
Timing here matters. Poth isn't entering a quiet market.
Mixpanel rolled out its own "Root Cause Analysis Agent" back in May, designed to automatically diagnose behavioral anomalies and suggest next steps. Amplitude announced "Agentic AI Analytics" in February. Qualtrics has been pushing "Experience Agents" and synthetic research capabilities since early 2026. The incumbents, in other words, are moving.
The broader AI-powered research category has attracted serious capital. Listen Labs raised $69 million in a Series B earlier this year, counting Microsoft and Sweetgreen among its clients. Pogo launched an "AI researcher" platform in June after securing $32 million. Specialized players are proliferating too: Causaly focuses on life sciences research, Hudson Labs targets financial markets, Elicit tackles academic literature reviews.
So what makes Poth different? The company's framing emphasizes the inefficiencies of the feedback industry, built on what it calls "outdated systems" that flatten rich intelligence into static themes and counts. Fair enough. But whether that positioning resonates with teams already using next-generation tools from Mixpanel or Amplitude—whether Poth offers something genuinely differentiated beyond good marketing—remains to be tested in the wild.
No customer logos appear on the site yet. No case studies. An interactive demo lets visitors watch the system work, but that's not the same as seeing it deployed at scale.
The Founders
Wong, the CEO, brings a Palantir pedigree—forward-deployed engineer and deployment strategist before Stanford—which suggests comfort with complex enterprise workflows and selling into large organizations. Horvath, the CTO, came from Tietoevry after stints at Phillips Academy and HTL-Spengergasse. The technical chops appear solid.
What's notable: they've kept the team deliberately small. Just the two of them, according to their Y Combinator listing, operating across New York and San Francisco. That's either strategic focus or very early days. Probably both.
Their YC partner is Ankit Gupta. The contact page notes a response time within 24 hours, which feels optimistic for a two-person startup but speaks to their ambition. Or perhaps to how seriously they're taking early customer conversations.
The Uncertainty Factor
Poth launched its platform in late June, though the exact contours of what "launched" means remain unclear. The company hasn't published pricing or packaging details. The primary call-to-action is booking a demo, which is standard for enterprise software but makes it difficult to gauge where the product actually stands.
No public funding beyond Y Combinator's standard batch investment has been announced as of July 2026—Demo Day would typically follow in the coming months if the company graduated from a recent batch. That timing creates some ambiguity about how fresh this venture really is, and whether what's available now is a fully-baked platform or something closer to an extended proof-of-concept.
The website includes product pages, founder bios, the interactive demo, and a contact form. The Y Combinator branding appears prominently, which is typical for recent graduates still trading on that association. The accelerator lists Poth in its market research category alongside other AI-first research platforms.
What Comes Next
The central wager here is that product teams will trust an AI agent to do the investigative work that traditionally requires human judgment: noticing the right pattern, asking the right questions, distinguishing signal from noise. That's a big ask, particularly in organizations where research and product functions are deeply territorial.
But perhaps the bar isn't perfection. Perhaps it's just speed—getting to a decent answer in 47 seconds instead of three weeks. That might be enough.
For now, Poth remains more promise than proof. The technology is real, the founders are credentialed, the market is active. Whether they can carve out space between entrenched analytics giants and well-funded AI research startups is the only question that matters. And that won't be answered on a demo page.
---
Poth Labs can be reached at [email protected] for inquiries about the platform.
