Polar

Polar is an AI browser that automates internet work: research, recruiting, sales, operations. It operates inside your existing accounts and sessions rather than logging in as a bot, handles scheduled workflows, and claims autonomous runs lasting hours with a follow-along view of what it is doing. The team publishes its own benchmark, Odysseys BU Bench V1, scoring Polar 98.0 against GPT-5.4 at 60.8 and Claude Opus 4.6 at 44.5 - self-reported, so read it as marketing with a methodology attached, not as an independent eval. Built by a team from MIT, YC, Perplexity, Modal, Jane Street, and Apple.

The insight

The insight: scheduled work is the real product

A browser that does one task when asked is a demo. Polar’s scheduled workflows - daily summaries, recurring reports, monitored queues - are what turn an agent into infrastructure, and they are the reason “operates inside your existing sessions” matters: the scheduled job inherits your logged-in context, no credential juggling. The self-reported benchmark (98.0 vs GPT-5.4’s 60.8) is worth reading as a claim to test on your own workload, not a scoreboard. Hours-long autonomy is the differentiator either way: the tool category is shifting from answer-machines to processes that run while you do something else.