Wow. Last year, everyone got excited because the 4.6 models changed the game. Suddenly Claude Code was good. Leave OpenAI! We have found the winner!
But actually no.
It's not the models that made Claude Code worth using.
It's the product idea from March 2025.
In 2024, Anthropic quietly scored a coup when they brought on board Instagram co-founder, Mike Krieger. He had been working on Artifact, a nice way to read the news. He joined as Chief Product Officer (and now runs their Labs team, who put out Claude Design.)
ChatGPT popped everyone's socks off because of that crazy feeling when you typed a few words and WOW it generated so much more than the effort put in.
But as the use cases have become more advanced, the required thinking to operate this technology has gone up. Yes you get "prompt engineering" for free now thanks to enrichment in the thinking loop. But for great results, most of the real management and thinking and creativity must be sourced from your mental model, not the language model.
I've been saying for years now: chat is a horrible horrible interface.
For most people, writing emails, going to meetings are the worst parts of their day. Using language to make things is as difficult as it is unpleasant. And it's crystal clear from LinkedIn that they still can't tell what bad writing looks like even with arguably the most powerful technology ever created.
But you know what, we didn't ask them to write the code for the software they've been using for 20 years either. We used our understanding of what they are trying to achieve to make it easy for them.
This is essentially what a product is. My favourite product man, and sometime colleague, Mikkel Bo Schmidt once told me that great design is about thinking through and solving most options for the user — and only leaving them the vital ones you cannot.
Where Anthropic has pulled ahead is in offering more product-shaped experiences beyond pure chat.
The best product for coding was one that could live in your terminal, IDE, apps or anywhere else. Because devs may work anywhere…
CoWork is an effort to translate that, with product thinking to knowledge workers — for whom you can make more assumptions and design a more fixed environment.
Claude Design is what a product might look like that captures the crucial ideation step in a space equipped for a user's needs in that moment.
At the start of this year, I predicted we would start to see chat disappear behind interfaces. We aren't quite there yet on the INPUT — but we ARE seeing less and less of the agent's behind-the-scenes OUTPUT. Instead, Anthropic's products show us… the product of your work: the work doc CoWork put together, the Design you created. And whatever might be next.
Now.
The problem here is, Anthropic isn't magic. They got there first. But OpenAI has a phenomenal team too, and has learned from the last year.
Perhaps the most important lesson of Claude Code is how quickly a product can change the direction of travel.
And even more so: how unfair is the advantage of being super close to the model to building the next great product experience with it? Maybe you get a better result from it at the edges, and you certainly pay less to run your product.
But there's no reason why the next big innovations in PRODUCT and INTERFACE couldn't come from anywhere.
In fact, it confuses me constantly why we haven't seen more of this. A thousand flowers blooming.
At those moments I try to remind myself what early days these are. Working at Beyond Work in 2023, I still think that team had better ideas than a lot of what I see in the market now. But very few have ever seen them.
I think the ideas are bubbling beneath the surface.
For what pops next, I think it will be about better input interfaces. Not chat. Not letter by letter. Probably generated almost automatically and naturally through context, and agentically in a way that's so good you can trust it. You give consent more than commission.
The question is just how soon that moment comes…