Lesson 07 · ChatGPT Mastery Pro ~12 min Browser + actions

ChatGPT Agent: real autonomy in your browser.

ChatGPT Agent is the agentic mode that merged Operator and Deep Research into one system. It drives a virtual browser — navigating sites, filling forms, running research, and delivering finished work. This isn't chatting. You hand off a task, it works for minutes, and it comes back with a result — pausing for you at the moments that matter.

The mental model

Agent ≠ Chat. You're delegating, not conversing.

💬 Chat
works in turns

You ask, it answers, you refine. You're present for every step of the thinking.

vs
🤖 Agent
works in tasks

You describe an outcome; it works on it for 5–30 minutes in a real browser and reports back. Like a remote worker with internet access.

It can log into sites you authorize, comparison-shop, scrape data into spreadsheets, book things. What it won't do: act without authorization, pay without confirming, or push through a high-stakes step on its own. Watch exactly where it stops for you below.

Watch it work · then decide

Run an agent task end to end.

Here's a real delegation. Hit run and watch the agent work autonomously — until it reaches the one step it won't take without you.

Delegated task
"Find the best conference-room speakerphone under $500. Compare Best Buy, Amazon, B&H, and the maker sites, then give me a ranked table."
⏸ Agent paused — needs your call
Top pick found: Jabra Speak2 75 — $412, in stock at Best Buy. The agent can place the order, but it won't spend money without you.
Delivered · comparison.csv
#ModelPriceWhy
1Jabra Speak2 75$412Best mics + battery; top pro reviews
2Poly Sync 40$329Cheaper, great range, fewer features
3Logitech P710e$289Budget pick; dated but reliable
Your call · delegate or just chat?

When is it worth handing off?

Agent runs take minutes and need supervision. Some tasks earn that; some are faster in plain chat.

Keep yourself in the loop

The agent is powerful, not infallible. Build these into every run:

  • Say "don't purchase / don't submit" explicitly when you only want research — it asks by default, but be explicit.
  • Expect CAPTCHA pauses — they're designed to stop agents; it'll hand those back to you.
  • Fact-check the output — it can still be confidently wrong. Treat results as a strong first draft, not finished work.
About ChatGPT Agent: released July 2025 by OpenAI as the successor that merged Operator (browser-action agent) and Deep Research (long-running research agent). Available to Plus, Pro, and Team users; free-tier access is limited.

Delegate a multi-hour task

Pick something that would take you 3+ hours by hand — comparison research, data entry, a researched draft — and hand it to ChatGPT Agent. Don't multitask; check back every 10 minutes in case it needs you. Pay attention to the shift from "I'm doing this" to "I'm reviewing this."

What you can do now

  • Tell when to reach for Agent vs. regular ChatGPT
  • Specify "don't purchase or commit" when you only want research
  • Use Agent for tasks that span multiple sites or sources
  • Stay in the loop on high-stakes steps — and always review the output
  • Expect CAPTCHAs to pause the run, by design
Pro
Up next in ChatGPT Mastery

Lesson 8 · Deep Research — long-running multi-source synthesis

When you need depth over breadth: multi-source synthesis with citations, output that reads like an analyst's report, now connectable to your own tools and data via MCP. Start lesson 8 →

🎓
AI Coach
Ask anything about this lesson
Hey! I'm your AI Coach for this lesson. Ask me anything about what you just read — concepts, examples, how to apply it to your work. What's on your mind?
Free lesson coaching is limited to 3 questions. Upgrade to Pro for unlimited coaching on every lesson.