Everyone has work they’re uniquely great at: the judgment calls, the relationships, the craft.
And then there’s the rest of the day: opening 30 tabs to answer a single question, filling out hundreds of forms, and updating spreadsheets, emails, and docs over and over.
The first wave of AI browsers didn’t change that. They were good at chatting with your tabs and doing a couple of clicks for you, but the work still landed on you.
Last year, some of us helped build Comet at Perplexity, and we saw how the best agents couldn’t reliably add a calendar event. But with the newest AI models, we built prototypes that worked for hours. That convinced us there was a bigger step to take.
A browser where you get 100 hours of work from 30 seconds of typing. And then you can run ten of those at the same time.
The Product
So we created Polar. It clicks, types, and navigates the web the way you would, logged in as you. It completes complex tasks on any website for as long as it takes, whether that’s a couple minutes or several hours.
For seven months, people have come back week after week, asking Polar to take more than 4.5 million actions on the internet to do research, recruiting, sales, ops, marketing, and other grunt work.
The product has become a daily driver for many individuals and teams, rapidly growing through word of mouth and saving one company 25+ hours per week per person (and they also cancelled Clay). We’ll be launching some use case studies in the coming weeks to showcase the benefits!
Polar’s Performance
Polar is the world’s most powerful browser agent, leapfrogging both OpenAI and Anthropic.
Although the leading browser agent benchmarks don’t represent knowledge work tasks on domains people actually use, we evaluated Polar on them:
Polar is especially good at large scale, long running tasks due to how we designed the browser agent (we’ve seen Polar complete tasks for more than 15 hours):
- Building many tools to allow Polar to navigate and interact with dynamic and difficult web UIs in a generalizable way (e.g. spreadsheets, maps, docs, social media)
- Giving ways for Polar to orchestrate a team of agents, delegate properly, and ensure smooth agent to agent communication
- Mixing and matching different models across the agent to design the most powerful and capable knowledge work agent
- Designing a memory system to learn more about you over time, and therefore, do your tasks better
Our Mission
We want to automate all knowledge work. Not to replace humans, but to multiply everyone so you can focus on the things you actually care about.
A vast proportion of knowledge work runs through the browser. So if we build an excellent general browser agent that can use any website, then we can automate a significant percentage of knowledge work.
Eventually, we want to completely redefine how knowledge workers do work every day, just like how coding agents did that for programmers. So you’ll know we succeeded when everyone who does work (spoiler alert, that’s basically everyone in the world!) uses Polar as their daily driver.
The Company
Our company is Recursive Intelligence, an applied AI lab building end-to-end browser and computer use agents.
Our founding team of three comes from MIT, YC, Jane Street, Citadel, Apple, Perplexity, and Modal. We also have olympiad medals, Team USA international medals, and a chess national master between us.
We’ve raised a $5.7 million seed round led by Madrona to realize our mission.
And we’re grateful for the support of our angels, including Thomas Dohmke (former CEO of GitHub), Erik Bernhardsson (Modal), Robert Yang (Fundamental), Nico (Fundamental), and many more.
And, of course, we would literally not exist without our users. So a special thank you to Oliver Pour, the Andera team, Nam Dao, Neelan Veloo, Jason Wexler, Shervin Dehmoubed, and countless others for helping us on our early journey.
We’re incredibly excited about the future of work. If you are too, then consider joining us!