Vibe Check: OpenAI Enters the Browser Wars With ChatGPT Agent

Was this newsletter forwarded to you? Sign up to get it in your inbox.

Surprise! Operator and deep research had a baby, and it’s called ChatGPT Agent.

You can imagine the meet-cute happening in a dark, humming server room in Texas:

Deep research—highly verbal, long-winded, and bolted to a server rack—completing sentence 97 of a 10,000-word report on the hiring landscape for AI engineers circa 2024 to 2025. It stares at the LinkedIn login screen, sweating, thinking… thinking… thinking… in circles. It can think, but, alas, it can’t do.

Operator streams in, equipped with the high personality and low IQ of a golden retriever, but also—miraculously—with hands. It comes up behind deep research, leans in close, and presses “Login.”

“Done!” says Operator with a twinkle.

The rest, as they say, is artificial history. (Reader, they interfaced.)

Their progeny, ChatGPT Agent, is like deep research in that it can do long-running research tasks. And it’s like Operator in that it can use a computer, including a browser, a terminal, and LibreOffice. So it can do a host of things that neither of its parents could do on their own:

1. Navigate complex multi-step workflows: Read all of your technical support emails, identify product promoters, search for them on LinkedIn, and synthesize customer archetypes
2. Transform raw business data into executive presentations: Analyze P&L spreadsheets and performance metrics, and generate PowerPoint decks with insights
3. Conduct comprehensive UX audits: Browse through multiple websites, document user flows, and compile detailed usability reports
4. Create intelligence briefings from real-time data: Scan news sites, research papers, and forums to produce daily executive summaries on specific topics
5. Handle authentication and dynamic content: Log into password-protected sites, navigate JavaScript-heavy pages, and extract data from behind paywalls

After a few days of using it, here’s my Vibe Check. We’ll go through how it works, the Reach Test (is it a habit?), and what it means for the competitive landscape. Let’s dive in.

AI that gets how you work

AI is better with context. Notion’s new AI tools take your stored notes and knowledge across apps to create exactly what you need, whether it’s the right document to share before a big meeting or a stroke of genius that unlocks a complex project pulled from a years-old email. Don’t start from scratch with a new AI tool.

Try it today

Want to sponsor Every? Click here.

A tour of ChatGPT Agent

We’ll start our tour of ChatGPT Agent with an example. Our email management product, Cora, has been growing quickly, and I wanted to understand more about the current state of the customer base. I wanted to know:

1. Why people love us (what the job to be done is)
2. Who loves us (customer archetype)
3. The biggest complaints

So I started a new ChatGPT chat and added “ChatGPT Agent” as a tool, similar to how you invoke deep research:

Then, I prompted ChatGPT Agent to read two months’ worth of support emails and feedback forum posts. I instructed it to look at our support email account and all of the support forum posts, and asked it to tell me who uses the product, who loves it, and what the complaints are.

In about 15 minutes, it went through 1,300 emails and a ton of posts to create a 2,000-word report on our most common complaints.

It also found all of the people who love the product, searched for them on LinkedIn, and created a set of customer archetypes to help me understand who loves Cora and why:

Neither deep research nor Operator could do this on its own, but ChatGPT Agent can do it easily.

Here are a few more tasks I had it perform:

1. Find me a place I can spend a week as a writing retreat somewhere in the Berkshires or Upstate New York.
2. Given Every’s P&L and weekly performance tracker spreadsheets, do an in-depth analysis of our business, and compile the results into a PowerPoint presentation.
3. Review the UX of Every’s main websites (Every, Cora, Sparkle, Spiral) and present a detailed report on the findings.
4. Read the internet and create a daily executive briefing on the last 24 hours of AI news, specifically focusing on AI and games, alignment, and synthetic data.

It’s an impressive piece of technology, and unlike other frontier agent experiences like Claude Code, it’s not intimidating to use if you’re not a developer. You don’t have to open up a terminal, and it doesn’t run on your computer like Claude Code does. Instead, for each chat for which you invoke ChatGPT Agent, it starts a virtual machine in the cloud and runs everything right from inside of ChatGPT.

As always with a Vibe Check, here’s your Reach Test:

Become a paid subscriber to Every to unlock this piece and learn:

1. The results of the Reach Test
2. How ChatGPT Agent compares to Claude Code
3. The battle for browser-layer dominance