PSA: The line between "Chat" and "Computer" mode on Perplexity is dangerously blurry.
I'm on the Pro plan. Last month, I thought I was just using regular Search to work on my texts. Turns out, I had unknowingly switched to "Computer" mode — and it spawned autonomous agents without me realizing it.
What happened:
· One agent turned my project into a recurring "university lecture" task, running 8 times in a single day.
· Another agent built a full AI-news website structure with daily updates, tables, and sections.
· When I tried to cancel a recurring task, it entered an error-retry loop that burned ~$40 in credits within minutes.
The core issue: the UI doesn't clearly signal when you move from "asking a question" to "deploying an autonomous agent." No red banner, no cost estimate, no confirmation for recurring tasks.
The capability is incredible, but we need guardrails: per-task spend limits, explicit confirmation for scheduled agents, and better visibility into what's running in the background.
If you use Computer, double-check which mode you're in — and set your own limits.