Test AI browser agents on real workday tasks
The author ran Comet and another agent through real workday tasks like booking a flight and handling calendar chaos; one took 16 minutes to book while the other did the calendar work in about 5.
Done withPerplexity Comet
- What they did
- The author ran eight everyday workday tests on Comet and on OpenAI's Operator, giving both the same tasks. They included rescheduling a meeting and drafting the follow-up email from a sidebar, finding a restaurant, sifting LinkedIn contacts affected by layoffs, and pulling insights from Amazon filings. The detailed per-test results are paywalled.
- How it went
- The free portion says Comet delivered, handling calendar chaos in minutes, while Operator was slow and clumsy. The 16-minute flight booking and the full side-by-side are in the paid section, which I couldn't read.
- Worth knowing
- The post says the tool costs $200/month, so the author recommends testing it on your own real work and checking the time saved before paying.
Try it yourself with Perplexity Comet
Help me test you on real work tasks: first, find and compare flights from [origin] to [destination] on [date] and get as far as the checkout page without paying; second, look at my calendar for [week] and propose a fix for any conflicts or double-bookings. Tell me how long each task took and where you got stuck. Pause and ask me before submitting any form, sending any invite, or changing any calendar event.
Want help getting this working? Join the Vibe Coding Academy ↗ Run by Alex Finn, who also runs this site.
Source: natesnewsletter.substack.com · Undated
Related use cases
Find coffee makers under $100 and add options to the cart
Perplexity Comet searched Amazon for coffee makers with a grinder, built a comparison table and added two options to the cart, stopping before purchase.
Shortlist destinations and compare stays
Comet shortlisted sunny destinations by flight time from Amsterdam, then pulled Airbnb and Booking.com listings into a comparison table.
Open recent protein-ligand structures in PDB
A researcher had Comet find the three latest GSK3B protein-ligand complexes in the PDB and open each in Ligand Interaction view.
Run a solo company with eight agent departments
A founder set up CEO, CFO, COO, lawyer, accountant, marketing, CTO, and improver agents that share memory and consult each other.
Surface unanswered emails from the past week
Author had Devin build a Gmail API assistant that lists unanswered emails from the past week, since Attio lacks that tab; he wrote the auth part himself in Cursor.
Source and screen candidates and reply over WhatsApp
Recruiters run OpenClaw to monitor GitHub for new contributors overnight, score resume folders against a job description, and draft WhatsApp replies held for human approval.
Get 5 like this every morning
The best things people got an AI agent to do, each with the prompt to try it.