Polar is an AI browser aimed at knowledge workers who spend their day clicking through LinkedIn, CRMs, spreadsheets, ATS tools, and internal dashboards. It logs in as the user and drives those sites the way a person would, taking on multi-step assignments across sales outreach, recruiting, and operations rather than stopping at answers or tab summaries. The launch video hints at the intended workflow, a team meeting where action items get called out (maximize LinkedIn outreach, get contracts out, run the numbers, update inventory) and Polar picks them up and executes them across the web.
The launch is anchored by a $5.7M seed round. The round was led by Madrona, with participation from angel investors including former GitHub CEO Thomas Dohmke and Modal founder Erik Bernhardsson . Polar has grown largely by word of mouth over its first seven months, with users running more than 4.5 million actions through it across research, recruiting, sales, operations, and marketing, and the company says it significantly outperforms OpenAI and Anthropic on browser-agent benchmarks . The company also reports that tasks have run for over fifteen hours without intervention , which frames the pitch around long-running work rather than one-shot prompts.
Polar was founded by Kevin Jiang, Vishaal Ram, and Howard Zhong, a team with backgrounds from MIT, Perplexity, and leading quantitative funds. For founders and operators watching the browser-agent space consolidate, this launch is worth a look as an early data point on whether general web agents can replace point tools like Clay for outbound and enrichment work.
Comments (13)
Sign in to join the discussion.
P
Priya KattanSep 3, 2026
Beating Anthropic and OpenAI on benchmarks is a bold flex to open with. Would love to see the actual eval setup before I believe the graph in my head.
A
Amaru Q.Sep 3, 2026
The launch video hook is a screen recording with no face and no stakes for the first six seconds. You earned the raise, spend 400 bucks on an editor next time.
T
Tore BlixenSep 3, 2026
4.5M actions is a fun number until you realize half of web agents count every scroll as an action. What's the definition here?
M
Marcus OdabashianSep 3, 2026
Someone cancelled Clay for you? That's not a case study, that's a war crime confession.
Y
Yuki WatanabeSep 3, 2026
Called it after seeing the demo in July. Web agents that actually finish tasks are the only wedge that matters right now.
D
davide r.Sep 3, 2026
The wordmark is doing that thing where the 'o' and 'l' are kissing. Loosen the tracking, I beg.
H
Hakeem Bello-DialloSep 3, 2026
The tweet buries the lede. '25 hours a week per person' should be the hook, not paragraph three.
R
Rita MzanskiSep 3, 2026
Before anyone deploys this at my company: SOC 2 Type II, SSO via Okta, and where do the browser session artifacts live at rest?
S
solobuilderSep 3, 2026
MIT team raises 5.7M to ship a browser wrapper while I'm over here shipping the same agent loop with two people and a Bruno collection. Respect the hustle though.
N
Nadya BekovaSep 3, 2026
Web agents have the worst retention curves in the category. Curious what week 8 usage looks like on those pilot accounts before I get excited.
O
Ola FashinaSep 3, 2026
If you're pricing per action and each action calls a frontier model, your gross margin is basically a hostage negotiation with Anthropic.
J
june l.Sep 3, 2026
'The AI browser that actually does work for you' has one 'actually' too many. Try 'Your browser, working while you sleep.' No charge.
T
Ted KowalczykSep 3, 2026
Is this like Rabbit but in a tab? I feel like we did this bit already.
Work With Us
Book an intro call to explore our suite of services. We might have a waitlist.
Polar Browser
The AI browser that actually does work for you
The AI browser that actually does work for you
Tags
TLVC Rating
Community Rating
No ratings yet
to rate this launch.
Polar is an AI browser aimed at knowledge workers who spend their day clicking through LinkedIn, CRMs, spreadsheets, ATS tools, and internal dashboards. It logs in as the user and drives those sites the way a person would, taking on multi-step assignments across sales outreach, recruiting, and operations rather than stopping at answers or tab summaries. The launch video hints at the intended workflow, a team meeting where action items get called out (maximize LinkedIn outreach, get contracts out, run the numbers, update inventory) and Polar picks them up and executes them across the web. The launch is anchored by a $5.7M seed round. The round was led by Madrona, with participation from angel investors including former GitHub CEO Thomas Dohmke and Modal founder Erik Bernhardsson . Polar has grown largely by word of mouth over its first seven months, with users running more than 4.5 million actions through it across research, recruiting, sales, operations, and marketing, and the company says it significantly outperforms OpenAI and Anthropic on browser-agent benchmarks . The company also reports that tasks have run for over fifteen hours without intervention , which frames the pitch around long-running work rather than one-shot prompts. Polar was founded by Kevin Jiang, Vishaal Ram, and Howard Zhong, a team with backgrounds from MIT, Perplexity, and leading quantitative funds. For founders and operators watching the browser-agent space consolidate, this launch is worth a look as an early data point on whether general web agents can replace point tools like Clay for outbound and enrichment work.
Comments (13)
Beating Anthropic and OpenAI on benchmarks is a bold flex to open with. Would love to see the actual eval setup before I believe the graph in my head.
The launch video hook is a screen recording with no face and no stakes for the first six seconds. You earned the raise, spend 400 bucks on an editor next time.
4.5M actions is a fun number until you realize half of web agents count every scroll as an action. What's the definition here?
Someone cancelled Clay for you? That's not a case study, that's a war crime confession.
Called it after seeing the demo in July. Web agents that actually finish tasks are the only wedge that matters right now.
The wordmark is doing that thing where the 'o' and 'l' are kissing. Loosen the tracking, I beg.
The tweet buries the lede. '25 hours a week per person' should be the hook, not paragraph three.
Before anyone deploys this at my company: SOC 2 Type II, SSO via Okta, and where do the browser session artifacts live at rest?
MIT team raises 5.7M to ship a browser wrapper while I'm over here shipping the same agent loop with two people and a Bruno collection. Respect the hustle though.
Web agents have the worst retention curves in the category. Curious what week 8 usage looks like on those pilot accounts before I get excited.
If you're pricing per action and each action calls a frontier model, your gross margin is basically a hostage negotiation with Anthropic.
'The AI browser that actually does work for you' has one 'actually' too many. Try 'Your browser, working while you sleep.' No charge.
Is this like Rabbit but in a tab? I feel like we did this bit already.