OpenAI just launched Atlas, their AI-powered browser. Let me be direct: it's a worse version of Perplexity's Comet right now. Like ordering a ribeye and getting served the Costco food court hot dog. Sure, it’s still food, but come on.

What Atlas does (and doesn't):

We tested Atlas against Comet on actual shopping tasks. Comet wins on reliability and speed. (If you tried Comet back when we covered it, you know what I'm talking about. If you didn't, check that newsletter.)

So why are we even talking about Atlas? At this point, you’re probably thinking I wasted your time reading about a worse browser. Fair, but hear me out.

It doesn't matter if Atlas is worse right now. What matters is what it signals.

Perplexity Released an AI Browser. Microsoft has an AI browser extension. Google has an AI browser extension. OpenAI now released an AI browser.

The real story isn't the browser. It's that OpenAI building a browser means they know what's coming.

Klaviyo just released research on how shoppers are already using AI agents for shopping. The data is wild. Consumers are actively using AI to find deals, compare prices, and discover products. You know it’s real when even your email platform is telling you to worry about something other than your welcome series.

78% of consumers used AI for shopping or product research in the past three months. McKinsey is predicting up to $5 trillion in agentic commerce by 2030. That's not a typo. I checked three times because even I don't trust numbers that look like my kid made them up

But most of this isn't happening in AI browsers... yet. It's happening in ChatGPT, Perplexity, Gemini, and other chat interfaces. Your customer tells an agent what they need, and the agent finds it. No AI browser required.

Should you try Atlas?

Yes. But not because it's better than Comet. Try it because you need to experience what the future of discovery looks like. Even if we're not there yet, you need to feel how frictionless this could be. Open it. Use it. Watch how an AI can browse your site, understand your products, and complete a transaction. If you haven’t done this with ChatGPT on Etsy, you should try it there, too.

That experience… that's the future of commerce. And it's coming faster than you think.

Here's what you should actually do with Atlas this weekend:

Test 1: The New Customer Discovery Open Atlas and pretend you've never heard of a competitor. Ask it: "I'm looking for [your product category]. Find me the best options under $[your price point]."

Watch what it recommends. Watch if it finds you or recommends someone else. This is how customers will actually discover you in an agentic world.

Test 2: The Seasonal Shopping Intent Ask Atlas: "I need [seasonal use case for your product] and I want it within [timeframe]. Find it and add it to the cart for me."

See if agent mode can navigate to the site, understand urgency, and recommend products. Does it? Or does it get stuck?

Test 3: The Comparison Shop Tell Atlas: "Compare [your product] vs [main competitor]. Which is better for [specific use case]? Set me up to purchase the best one."

Does it pull accurate information about your product? Can it find your product page easily? Does it compare fairly?

Test 4: The Deal Hunter Ask Atlas: "Find me the best deal on [your product type] this week. Then add it to my cart."

This tells you how agents hunt for your category. Where are they looking? Are they finding you?

Why this matters: You'll see exactly how agents navigate your site, how they understand your product, and where they get confused. It’s going to show you how important it’s going to be to show up in AI search, and how important it’s going to be for your site to be optimized for agents in the future.

Here's what happened in the last two weeks: OpenAI dropped Sora 2. Google dropped Veo 3.1. Everyone started losing their minds about "studio quality AI video" and "replace your whole creative team."

Look, I spent about 12 hours testing Sora 2, Veo 3.1, and the landscape around them.

Here's the reality: We're at a moment where the marketing is running about 12 months ahead of what actually works.

The hype narrative:

The actual truth: Both models are genuinely better than they were. The physics are legitimately improved. The audio sync is real. But here's what people aren't telling you: they're still hitting a ceiling for DTC content creation.

I’ve been playing around with a brand idea, so I used Sora 2 and Veo 3.1 to see if I could create some great TOF videos for it. I spent about 4 hours between the two of them and $400 seeing what I could get.

Sora 2:

It’s ChatGPT-3 for video. It’s better than Sora, makes a lot of mistakes still, but impressive when it gets it right. Sam Altman’s quote on this was brilliant and tells you all you need to know:

More specifically:

Veo 3.1:

This is a definite improvement over Veo3. It is so much better than Sora 2. The controls they’ve added make a huge difference, and scene builder is way better.

What’s worth using: If you’re looking for quick b-roll, or a scroll-stopping 8s opener Veo 3.1 is great. If you’re willing to spend $10 trying to connect two scenes together, Veo 3.1 is still great. Just understand, you'll need 3-5 iterations, per scene, to get close to something worth using. That kills the "just prompt it and done" narrative.

When it comes to Sora 2… honestly, the most interesting part of Sora 2 is what an entirely AI content geared social network would be like (like the Meta Vibes test). I’ll say, it actually wasn’t as bad as I thought it would be.

With the release of Haiku 4.5 last week, I did a deep dive comparing 7 different AI models to determine which one was the best for creative marketing. The results were actually not at all what I expected (hint: It’s not ChatGPT-5, surprising no one).

And that's it for this week's edition.

The real story isn't the video models or Atlas. Everyone's watching Sora 2 and missing what matters: the front door to commerce is shifting from Google and Amazon to AI agents, and I’m not sure if any of us are ready.