Zapier and n8n both lean users into building an agent by requiring them to define exactly what it should do; n8n is rising in popularity because it does this better, but neither is a general-purpose agent—they are useful only for specific agentic tasks like extracting text from a document into a spreadsheet.

Comet is a true general-purpose agent assistant, with the closest prior comparisons being OpenAI's Operator and Gemini's Project Mariner, but Operator's vision (ChatGPT decisioning across the web) was doomed by poor execution and UI—a slow, awkward toy-sized browser inside a chat window.

All major AI players are converging toward owning the browser because we live so much on the web that becoming the dominant browser of choice effectively makes you the operating system for AI—the strategic prize Perplexity is pursuing with Comet.

Whether Comet is worth $200/month requires disciplined measurement: adding up the 5–15 minute increments it saves throughout the day; if it saves 10–15 hours a month it is clearly worth it, and at a $35/hour valuation 10 hours saved yields close to 2x ROI monthly—representing a new valuation paradigm for software driven by baked-in cognitive intelligence.

Comet can run multiple tasks simultaneously across integrated services—LinkedIn, calendar, Gmail—and a running task continues even when the user navigates elsewhere in the address bar.

Many current AI agents bury the user in the cost of building or directly controlling the assistant, forcing heavy upfront investment in defining and configuring behavior, which undermines their value as general-purpose tools.

Comet works as an AI agent not because of superior AI but because of superior user interface—the team understood that the user's goal is to let the assistant do the work and disappear, rather than to watch or micromanage everything the assistant does.

The reviewer suspects Operator misreports its task time, claiming an eight-minute job actually took around 20 minutes of clock time, possibly because it starts the clock late.

When prompted with the thesis that TikTok Shop's slowdown affects social commerce broadly and may impact Amazon's outlook, Comet's analysis concluded Amazon is pursuing an agnostic social commerce approach—explicitly responding to TikTok with integrations across Instagram, Snap, Pinterest, and YouTube Shorts—and is hedging risk and investing further rather than ignoring the slowdown.

Comet's browser-takeover UI—glowing the main browser blue while it acts, then removing the glow—is an intuitive signal that the agent is driving while preserving the user's ability to stop it at any time.

The widespread design assumption that users must see what the agent is doing is a legacy of treating agents as untrustworthy; once an agent can hook into data and just works, close supervision becomes unnecessary and undesirable—users want it to be fast, trustworthy, and dependable.

Comet's response on Amazon's social commerce strategy was more coherent, actionable, and on-point than typical AI output, which tends to be bland.

Comet completed a real meeting-rescheduling workflow end-to-end from the sidebar: it found the meeting, suggested a better time fitting the user's preference for clean working blocks, drafted a calendar change, then proactively recommended sending a notification email which it also drafted—all without the user opening Gmail or Google Calendar.

Comet feels three or four months ahead of schedule—the reviewer expected general-purpose agents were coming but did not expect one this fast, efficient, effective, and connected across flight reservations, restaurants, LinkedIn, Gmail, and calendar this soon.

Before recommending Comet, the reviewer subjected it to eight different workflow tests across diverse tasks (flight reservations, restaurants, LinkedIn, Gmail, calendar), recommending others sign up for one month and rigorously test it in their own environment, honestly measuring time savings before committing to the investment.

Because Comet operates natively within the browser, it can help users make better decisions in a way screenshotting cannot—there is something about native assistant-browser collaboration that cannot be replicated cleanly elsewhere, representing additional hard-to-measure value Perplexity is positioned to capture.