When OpenAI released a browser called Atlas in October 2025, Samantha McLean wanted to test it on something real. Not organise a holiday – a real estate task.
The brief: “I’ve got a potential listing in Broadbeach Waters. It’s a four-bed, two-bath, freestanding home, renovated kitchen, good school catchment. Create me a Google Sheet of recent sales in the area so that I can provide effective price guidance.”
Sam puts ChatGPT into what OpenAI calls agent mode — where the AI takes over the user’s screen and operates it the way a human would. In the video, ChatGPT navigates between tabs, clicks links, and enters data while visual indicators track what the AI is doing on screen.
It searches the portals for recent comparable sales in the Broadbeach Waters area. It opens a new Google Sheet, titles it “Broadbeach Waters recent sales,” and adds columns for address, sold date, and sold price. It enters the comparable sales data it has found and writes a formula.
Sam does not touch the keyboard at any point.
“It’s actually acting like a human,” Sam says. “Can you see that I’m not touching. Here are my hands. No hands.”
The process mirrors what a junior team member might do when given the same brief: search the portals, collate the results, organise them in a spreadsheet, and present them for review. Sam’s term for this is “trusted delegate.”
“That’s using AI as a trusted delegate,” she says. “Here is a job. Go and find me comparables. I’m going to grab a cup of coffee. Come back, and there it is.”
What this means for agents – and their vendors
The demo was recorded when Atlas first launched. The capability has advanced since, which makes the underlying question more pressing, not less.
Sam extends the scenario: how far off is giving AI a single listing brief and getting back the comparables, listing copy, and agency agreements?
But the implication cuts both ways. If an agent can give AI a brief and get a comparable sales spreadsheet back in minutes, so can a vendor. The same tools are available to both sides of the appraisal table. A buyer preparing to make an offer can run the same comparable sales research that used to require an agent’s access and expertise.
That shifts the value proposition. The agent’s edge is no longer having access to the data — it’s knowing what to do with it. Interpretation, local knowledge, negotiation strategy, and the relationships that come from years in a market. The CMA becomes table stakes. What the agent brings to the conversation after the CMA is what matters.
Key takeaways:
- OpenAI’s Atlas browser puts ChatGPT in agent mode, allowing it to take over a user’s screen and complete multi-step tasks — searching property portals, opening Google Sheets, and entering comparable sales data autonomously.
- The demo was recorded when Atlas launched in October 2025. The technology has only improved since.
- The shift from “AI as assistant” to “AI as trusted delegate” changes the agent’s role: from writing prompts to writing briefs, from reviewing text to reviewing completed work.
- The same tools are available to both vendors and buyers. If AI can pull comparables for an agent, it can pull comparables for anyone — which means the agent’s value moves from data access to data interpretation.
- Sam’s forward-looking question: how far off is the fully delegated listing campaign – comparables, copy, agency agreements, and marketing – from a single brief?
The AI Edit is a weekly series of short clips from the AI First Agent Accelerator.
New episodes weekly.
Want Samantha to present to your team or at your next event? Get in touch. Want the full system? Visit aiagentcourse.com.