So I've been experimenting with Comet, an LLM-based AI agent by Perplexity, and decided to throw it into one of the most chaotic casual games I could think of — Skribbl.io. No scripted help, no API access to the game. Just the AI reading screenshots, parsing the DOM, and acting like a player.

Here's what happened:

• It competed against 5 real human players across 3 full rounds
• It finished 1st place with 2,165 points
• It achieved a 67% word-guessing accuracy
• It actually attempted to draw — with mixed results (turns out controlling a mouse pixel by pixel is hard)

What blew me away wasn't just that it won — it's how it reasoned. It used letter constraints to narrow down words, ranked vocabulary by frequency, and adapted its strategy round by round. It also ran into genuinely funny failure modes, like accidentally drawing with the eraser the whole time.

I ended up writing a full research paper analyzing the whole session — the methodology, results, failure modes, and what this might mean for using games like Skribbl.io as AI benchmarks going forward. It's been submitted to SSRN and published on Academia.edu.

Full paper here: https://docs.google.com/document/d/e/2PACX-1vQMPjRYBeFTF0376cjADgkiKOwlKQK9YPXnhimGlq5eKAdK0nv0hBjS-W3OOY_uIhjHvsP56hzMruJ0/pub

Hey everyone,

I'm running into a frustrating issue with the Perplexity Comet Browser and wanted to see if anyone else is experiencing this, or if there's a known workaround.

The Issue

Whenever I try to trigger the Comet Assistant for browser automation tasks, it immediately blocks me and throws an error stating that my usage limit has been hit.

The Context

• Current Activity: I haven’t run a single automation task or browser chore in the past 2 weeks.
• Past Activity: Up until last month, I used the automation features quite extensively, and it worked phenomenally well.
• Current State: Despite the 14+ days of absolute zero activity, I am completely locked out.

It feels like my quota counter is completely stuck from last month's heavy usage, or there's some weird rolling window glitch happening behind the scenes.

Has anyone else dealt with ghost usage limits on Comet? Does anyone know if the limit resets on a strict monthly calendar date or a rolling 30-day window based on peak usage? Any insights before I try opening a support ticket would be awesome.

#Perplexity #CometBrowser #PerplexityAI #AIAgents #BrowserAutomation #TechSupport #Bug

Genuinely asking because my frustration with Comet has been building up. The rate limits Perplexity (Pro user here) has been enforcing are getting tighter and tighter, and it's really starting to kill agentic workflows. Mid-task interruptions, agent loops breaking because it hit a cap .. it's not just annoying, it actively breaks the whole point of using an agentic browser.

I get that limits exist, but at some point it stops being a "pro tool" and starts feeling like a freemium product with pro pricing. The core value prop of Comet is autonomous, multi-step tasks and if the underlying assistant keeps getting throttled, that falls apart fast.

Has anyone switched over to or tested BrowserOS for this reason? Is it a real alternative here, or is it still too early?

So, for those who have tried both:
How does BrowserOS compare with Comet for actual agentic tasks?
Is it better at long browsing sessions / multi-step workflows?
Does it feel more open, less constrained, or just different?
And do you think Perplexity is damaging Comet’s momentum by tightening limits too much?
Not trying to bash Comet .. I genuinely like the idea. But if the limits keep getting in the way, people will naturally start looking elsewhere.

TIA