The timing, as they say, is everything.
Just over three weeks after OpenAI quietly shuttered Atlas—its experimental standalone browser designed to automate tasks across the web—a Seattle-area startup called Polar emerged with $5.7 million in fresh funding and bold claims about outperforming the tech giants in AI-powered browsing. The seed round, led by Madrona, drew backing from a constellation of insider angels: Thomas Dohmke, who until recently ran GitHub; Erik Bernhardsson of Modal; and Robert Yang and Nico from Fundamental.
The announcement came July 29, 2026, landing in an already crowded field of startups racing to build what might be called the next evolution of the browser: tools that don't just retrieve information, but act on it. Click, type, navigate, repeat. For hours, if necessary.
Whether Polar can deliver on its ambitions remains an open question. But the company's arrival—and OpenAI's retreat—captures a peculiar moment in the AI browser wars, where the biggest labs are still figuring out what works, and smaller players sense an opening.
Big Claims, Few Referees
Polar, which operates under the name Recursive Intelligence, Inc., isn't shy about its standing. In materials published alongside the funding news, the company declared it had built "the world's most powerful browser agent," surpassing both OpenAI and Anthropic in head-to-head testing—though these performance claims are self-reported, with no independent verification available as of late July. Internal benchmarks scored Polar at 98% on agent evaluations, well ahead of YC-backed Aside (75.5%), Browser Use (70%), and proprietary models from OpenAI and Anthropic, which clocked in at roughly 60% and 44%, respectively.
Those figures draw from two assessment frameworks: Odysseys, a 200-task test suite run on live websites, and BU-Bench-V1, judged using Google's Gemini model. Polar made its evaluation outputs available on GitHub, offering a degree of transparency unusual in an industry often light on independent verification. Still, no third-party auditors had weighed in publicly as of late July, leaving the rankings largely self-reported.
It's a familiar dynamic in AI. Startups trumpet performance gains; the market waits to see if customers agree.
Automating the Drudgework of B2B
What Polar is actually selling, beneath the benchmark drama, is automation for knowledge workers—specifically those in sales, recruiting, and research. The browser is designed to run unattended, sometimes for stretches exceeding 15 hours, handling repetitive workflows while logged in as the user. Think: sales prospecting lists, candidate pipelines, competitive intelligence gathering.
The company claims one early customer saved upwards of 25 hours per week per employee and canceled a subscription to Clay, a data enrichment platform popular among sales teams, though formal case studies are still forthcoming. Polar's website offers workflow templates for market mapping and competitor analysis.
The browser itself is built on Chromium and runs on macOS, with a Windows version in beta. It's the work of a compact team—three people, by the company's count—drawn from the usual pedigree mills: MIT, Y Combinator, Jane Street, Citadel, with stints at Apple, Perplexity, and Modal. Individual founder names, curiously, weren't prominently displayed on public-facing properties as of late July.
A Window Opens
Polar's emergence traces back to Composer, a personal browser assistant the team introduced earlier in the year. By May, Composer had evolved into a full Chromium integration; by July, it had become Polar, complete with a new brand and venture backing.
The rebranding came at a fortuitous moment. OpenAI's decision to retire Atlas on July 9 created something of a vacuum. The company is redistributing Atlas functionality into its ChatGPT desktop app and a Chrome extension—an acknowledgment, perhaps, that standalone AI browsers are harder to get right than anticipated. Meanwhile, competitors like Aside, Perplexity's Comet, and The Browser Company's Dia are jockeying for position, each pitching slightly different visions of what automated browsing should look like.
Madrona, the Seattle-based venture firm that led Polar's round, has been doubling down on applied AI in recent years, particularly among Pacific Northwest startups. The firm raised new funds focused on the category between 2023 and 2025, betting that infrastructure and tooling for AI agents would become a defining investment theme.
Whether that bet pays off depends, in part, on whether tools like Polar can move beyond benchmarks and into the daily rhythms of actual companies. The demos are impressive. The challenge—as it always is—will be convincing teams to trust an AI to work unsupervised, for hours on end, with access to live accounts and real workflows.
OpenAI's retreat from the space doesn't mean the problem is unsolvable. It may just mean the solution isn't obvious yet.
