You are reading Molekyl, finally unfinished thinking on strategy, creativity and technology. Subscribe here to get new posts in your inbox.
The words in the above video1 are from this post I wrote in February. In it I tried to describe a nagging sensation that had been building up through the fall of 2025. That something had changed with AI.
Several times in the period from mid fall 2025 to January 2026, I had tried to write a post that captured what I felt was going on with AI. But every attempt ended in the drawer. I just couldn’t get it right.
In February, the feeling got stronger, and I tried again. With a different strategy. Instead of trying to describe what I felt with a coherent argument or smart narrative, I picked up my phone and spat out the words that felt natural at the time. After some quick polish, I hit publish.
The response to the post was mostly flat. A few people nodded and told me that they saw things in similar ways, but I got a sense that most who read it just wrote the post off as slightly weird. Like a friend of mine who gave me a call to ask if I was all right.
Since then, the post has not only felt increasingly accurate, it has also evolved alongside the same AI developments it sought to describe. A few months after writing it, I used its text for a song to test a new Suno model, and I later used that song in a video I created to test a new workflow with Seedance 2.0 and Claude.
This fall, the pace of new releases, events and creative use cases has once again sped up (if you don’t know what I am referring to, you have some catch-up work to do). Among the many things that have changed in just the last weeks is the release of Claude Opus 5.5. A remarkably capable and efficient model that overnight allowed anyone to do so much more with the same token budget. And as a bonus, it turned out that the model was also incredibly good at motion design and animation. That is, building sophisticated videos end to end, with code.
To put these capabilities to the test, I decided to see if I could once again breathe new life into the words of my post from February.
To arrive at the final edit above I spent a couple of hours of my own time, and maybe 2–2.5 hours of Claude compute. Spread over 12 iterations and a day and a half.
While it took some effort to ”get it right”, the end result left me kind of stunned. For one, a video like this would not have been possible for an amateur like me to make only a four weeks back. It also felt almost illegal to think about the hours that would have gone into this if humans were to do it all without AI. When I asked Claude to make an estimate, it came back with 600–1,200 hours (song included) by a team of highly capable specialist creators. Whether this estimate is accurate or not doesn’t matter. It would in any case be a pretty hefty project to pull off before AI. Now it’s something I could do in between other work.
So while something had changed both in the fall of 2025 and in the early spring of 2026, something has for sure changed this last month. It has truly been a wild one, and the capabilities of Opus 5.5 make up only a fraction of what creates this feeling.
Compared to when I published that original post back in February, I therefore sense that more of us now share a nagging sensation that something has changed.
In other words: I would be more surprised if anyone called to check if I’m all right this time around.
You can also find the video on Vimeo via this link.
Positive uses of AI are under considered for the creative trades at the moment mostly masked by the anger from stolen copywrite from businesses like Suno. I think you’re right the quality of the models has shifted, there is a lot left unexplored. I felt the same shift when web and digital technology hit the advertising and broadcasting trade 20 years ago.