GPT-5 launched yesterday. Most people are still waiting in line. We used our API to run our full DTC benchmark over the last 24 hours. I expected to be underwhelmed… I’m shocked. 23 real DTC scenarios. Results: - GPT-5: 0.91 (tied for highest score ever) - GPT-5 Mini: second place, way faster - It crushed product copy, email sequences, strategy, analysis One big drawback: speed. ChatGPT-5 is SLOW. The “is it broken?” kind of slow. Hot take: GPT-5 Mini is the real winner right now. Same intelligence. Better speed. Detailed analysis below 👇

Biggest surprise from the GPT-5 announcement? It wasn't they're trying to compete with Lovable. Or that they hopefully improved the coding capabilities. It's that it actually costs less! If GPT-5 produces, it's actually a big undercut to Claude's $3 price tag.

Kicking off the internal GPT-5 viewing party.

A prospect put our AI email generation to the test on a call this morning. They wanted to see how well our email generation created crazy variants. I honestly wasn't sure how it would go🤣 Our email agent crushed it. Asked our agent: Put it underwater and in a scuba gear.

Ideogram Character is actually nuts. They nailed the human/model side. This is actually a big upgrade for design. We're sooo close. Take a few shots from the photoshoot, create a character around it, and then you're set. Here's a test I did using Jones Road Beauty.

@nicgrasset saw the awesome update from tobi yesterday on agents. While it's awesome for the customer... we need some love for store admins! You know, so we can let agents do a little automation for us vs. getting blocked by cloudflare (just verifying doesn't work)?💙

I've used Chrome since it launched 16 years ago. But now... Comet has won. It's my default browser. Not just because it somehow operates more efficiently (same number of tabs open on Chrome & Comet). But it's how AI browsing should be. It's my non-dev IDE.

Insights from this: - Most people are not good at getting results from LLMs based on how "underwhelming" the response to o3 was - 4.5 doesn't have the same coding/math which explains why it's so good at creative - GPT5 lift difficulty tells me focused models coming soon

Raleon demos always make me smile. "We only care about X" "Wait, you can do Y, too?" "Can you just do this end-to-end for me and I'll review?" ✅

As AI eats software, every incumbent is going to be facing the innovators dilemma over the next 2 years. If you're not re-imaging what your software looks like in an AI-first way, you should be looking at who to buy, or who to be bought buy. MASSIVE moats being the exception