By now, almost everyone has AI somewhere in their workday: a chat tab open, a copilot in the sidebar, the odd summary or draft. Far fewer have watched it change the shape of the work itself. That distinction, between a tool that helps you get through the work and one that does the work for you, is the thing I keep coming back to. So far, only one field has really crossed it. Engineers got there first: over the last couple of years, coding agents went from autocomplete to doing real, multi-step work, and engineers now build more, and more ambitiously, than they used to. That is what a transformative unlock looks like.
My conviction is that every sector gets its version of that “Claude Code” moment eventually. You can hear it in a question I get constantly: “can’t Claude just do that?” For some tasks, sure. We’ve all had the first wow moment with these tools, and Claude and plenty of others are great at the trivial stuff. But knowledge workers still don’t have their true AI co-worker, the kind that handles a task all the way to done. Most tools can’t manage complex, multi-step workflows across different apps and environments without getting stuck, and they can’t adapt to the specific way each person works. But, it usually comes down to two things: getting into the sites where the real work happens, and having the reasoning to understand what needs to get done. And there is exactly one interface every knowledge worker already lives in everyday: the browser.
Building an agent that can do all of that, inside the browser, is extraordinarily hard. Few founders are brave enough to leave some of the best-funded AI companies and quant funds in the world for it, in a category long ruled by Chrome. It takes a deeply technical team that can see ahead of the curve and is willing to reimagine how millions of people work. When Soma, Ted and I first met Kevin, Vishaal, and Howard, we knew they were that team to tackle this problem, and it’s why Madrona has now led Polar‘s $5.7 million seed round.
How I Polar
Every day in VC looks different: sourcing one day, evaluating deals the next, reading through a stack of new research papers, spending the day with portfolio companies and LPs. But there are really only two interfaces I use every day: my communication tools (email, text, Slack) and my browser, now Polar. I’ve shifted almost all of that workload from Chrome to Polar over the last few months, and here’s what that looks like for me.
Diligencing a new company: normally I’d log in and search across a variety of sites – Tegus, Pitchbook, Substack, Affinity, the company website, GitHub, Hacker News, and more – then aggregate all of it into one Notion doc to find the real trends. Instead, I open my browser and ask Polar to go across all those sites. It spins up a reasoning and computer-use agent and flags the key pieces of information for me, cleanly, in a doc or HTML. Claude can sometimes do this too, but the moment a site’s blocked (which is often) or a connector is disconnected or doesn’t exist (which is weekly), I’m stuck waiting on IT, or re-connecting it myself and waiting again for the output.
Hosting a happy hour in NYC for Future Founders: this is a task with hundreds of sub-tasks that would normally take me days. Instead, I hand it to Polar. It researches venues, then fills out availability forms or emails venue teams directly to nail down food, beverage, and capacity constraints. Once that’s locked, it spawns another agent to work out who to invite – scanning the web for up-and-coming AI companies and emerging talent in NYC, and cross-checking my CRM and LinkedIn for people I’m already connected to. Then it creates the event on Luma, sends each flagged person a customized invite, and logs the outreach in a shared Google Sheet.
By the end of each day or week, I ask Polar what’s still outstanding, and it puts together a to-do list for me, split by work and personal. It can do that because it has context and memory on everything I did during the day – how I engaged with sites, what I asked it to do. Doesn’t matter if it’s a long-running task that used to take days or even weeks, or a quick one where I just need an answer – Polar handles either. And the best part: I can still browse like it’s a normal browser, except now it’s starting to remember me, and what I value.
Why Polar
The browser used to be where you searched. Now it’s where we all live, and where much of our work gets done. I don’t think that changes (people still love to browse), but what they want is to hand off the busywork. Over the last couple of years, we watched a wave of browser agents show up and fall apart a few steps in, losing track of the page, never surviving a task that ran longer than a few minutes. Most of what passed for AI in your browser was still you doing the work: holding the context, remembering what you’d already tried, stitching the steps together yourself.
The frontier models that shipped over the last few months made a lot of this possible. But the real unlock is what Polar built on top of them: a harness that understands your task, and executes by breaking that work into pieces that run in parallel instead of grinding through one step at a time. Those parallel agents all write back to a shared memory, so nothing gets lost across a session that runs for hours. And it’s built directly into the browser itself, not bolted on as an extension – which is what actually lets the agent learn how you work and get smarter over time.
Where this goes next is simple: instead of connecting into a hundred different applications one by one, it’s already logged into the sites, as me. It does the aggregating so I don’t have to, handing back the parts of the job that need me – the judgment call on who to meet, the deep reflection on a prospective startup’s potential, the decision itself, rather than the ten tabs it took to get there.
Always the Team
Kevin, Vishaal, and Howard left a role that most others would be thrilled to hold onto. Kevin was previously at Perplexity where he helped build Perplexity’s browser product, Comet. Vishaal was a research engineer at Fundamental Research Labs, and Howard a quant researcher at Citadel. Building a next-generation browser agent that finishes real work takes frontier-model understanding, genuine customer obsession, and a true AI-native mentality. They grew up on this technology instead of arriving at it, understand exactly how these systems are built under the hood, and default to first principles and risk taking instead of convention.
While the demo and benchmarks initially caught my eye, what truly convinced us was spending a few days with Polar, day in and out, running the same long-horizon tasks we’d thrown at every other model and browser agent out there, and watching them break down. Polar didn’t. Once we dug under the hood to understand how they’d architected it, and their vision for the future, we were convinced this was the team to make a winning product where others had failed.
For the knowledge workers using Polar today, the shift toward a more effective, transformative workflow is already here. What’s ahead is bringing it to everyone whose work runs through a browser, which is nearly everyone. That’s the transformation this category has been missing, and we think Polar is how it arrives. Madrona couldn’t be more excited to partner with Kevin, VIshaal, Howard and the entire Polar team! Try it for yourself for free here: polarbrowser.com.