I’m in the middle of building and launching a new product, in stealth mode. It’s entirely built by me. And yet, it dawned on me that I still wrote a spec. And a roadmap.
The biggest change this time was that it was written as much for myself, as it was written for a machine. And it was also written by a machine. It looked like this:
1. Write spec together with ChatGPT
2. Write implementation plan together with ChatGPT
3. Feed it to Windsurf
In all cases I’ve been using the GPT 4.1 model and it’s phenomenally good. Inside Windsurf I’ve sometimes switched between GPT 4.1 and Windsurf’s own SWE-1 model.
It would seem that so far, I haven’t been able to replace clear thinking with AI though. As I’ve written before these LLMs can function as a mirror or a museum. They still don’t have real creativity and agency to start something new. But for how long…