Pablo Molina and Iñaki Berenguer don’t really need to prove they can build a company. After selling their insurance startup, CoverWallet, to Aon in 2020, the pair could have easily coasted. Instead, they are back in the trenches with InfiniteWatch, a New York-based startup that just secured $4 million in pre-seed funding to solve a problem that is keeping technical officers awake at night: who is babysitting the AI?
The round was led by Base10 Partners, with participation from European funds Kfund, Kibo, and LifeX. The cap table also includes a suite of unicorn founders and scouts from heavyweights Sequoia and Andreessen Horowitz.
The pitch is straightforward. Over the last two years, enterprises have rushed to replace human support tiers with AI agents. But unlike human reps, who have supervisors and quality assurance (QA) scorecards, AI agents have largely been operating in a black box. InfiniteWatch is building the “observability layer” to fix that.
The Insurance Hangover
Molina and Berenguer aren’t approaching this as generic SaaS founders. Their time at CoverWallet was spent navigating the highly regulated insurance market, where a single misspoken phrase by a support rep could trigger a lawsuit or a compliance breach.
In 2024, they noticed a disturbing trend. Companies were handing off customer interactions to Large Language Models (LLMs) that, despite their sophistication, are prone to hallucinations and off-script behavior. The current solution for most companies is spot-checking chat logs—a method that doesn’t scale when an AI is handling thousands of conversations simultaneously.
In insurance, we learned the hard way that you can’t improve what you can’t see. With AI agents, the visibility gap is even bigger—and the stakes are even higher.
The duo argues that traditional analytics tools tell you what went wrong last week. They built InfiniteWatch to function as a real-time control plane, flagging issues as they happen rather than waiting for a quarterly review.
Session Replay for Robots
The platform isn’t trying to replace the AI agents; it sits on top of them. At launch, the company is deploying two core products:
- Session Replay Agent: This reconstructs web interactions in real-time. Instead of parsing through dense JSON logs or text transcripts, operators can watch a “replay” of the AI’s decision-making process to see exactly where it went off the rails.
- Voice Agent: This tool deploys synthetic personas to stress-test voice systems. Before an AI agent speaks to a real human, InfiniteWatch bombards it with edge cases and difficult queries to ensure it doesn’t break under pressure.
The goal is to move QA from a backward-looking administrative task to a predictive loop. If the software detects a compliance risk or a drop in sentiment, it flags it immediately, theoretically allowing companies to patch the behavior before it impacts revenue.
The Investor Thesis
Base10 Partners, a firm known for backing software that automates “real economy” workflows, views this as a necessary infrastructure play. As AI moves from experimental pilots to production-grade deployments, the lack of observability is becoming a bottleneck for adoption.
As AI agents move from pilots to production, boards are going to demand the same observability they expect from any mission-critical system. Pablo and Iñaki have already built in a compliance-heavy, high-stakes environment. They understand that missing a small anomaly today can become a huge liability tomorrow.
The involvement of scouts from Sequoia and a16z suggests that the broader venture market sees “AI oversight” as the next logical battleground in the generative AI stack.
Not Just Another Dashboard
InfiniteWatch is entering a noisy market. Incumbents like NICE and Verint, along with sales-tech unicorn Gong, have dominated the conversation intelligence space for years. However, Molina argues that those tools suffer from a fundamental legacy issue: they were designed to monitor humans.
Human-centric tools assume there is a call center manager, a shift schedule, and a human psychology at play. InfiniteWatch creates a control plane specifically for autonomous models that live inside apps and websites, independent of human schedules.
You don’t get product-market fit by slapping ‘AI’ on top of yesterday’s dashboard. You design the control plane from scratch for a world where your ‘employees’ are models, not people.
With $4 million in the bank—a substantial runway for a pre-seed team in the current market—InfiniteWatch plans to deepen its engineering bench in New York and Europe. They are betting that as CFOs and Risk Officers realize the potential liability of unmonitored AI, they will be looking for a system that does more than just record calls. They will want a system that watches the watchmen.
