AI does change the equation. It frees a solo developer to focus more on the big picture. BUT... with the current generation of AI agents, I think you are still right. You still need a second (or ideally more; I'd say 5) developer(s) to get enough perspective to have solid plans and roadmaps.

So, while I currently [mostly] agree, I think the / a next generation of agents might take that over a threshold and make solo development close enough to the equivalent of current 2-dev work to meaningfully change the equation. Furthermore I think that does not even need new models; I think current models with better "harness / tooling / system prompting / skills / etc." (whatever you may call the text files describing important procedures), might be able to fill most of that gap.

Obviously work that needs more than 2 devs planning might take even longer to fully solve with 1 dev + multiple agents, if ever.

My current mental model is that humans can very well think about and walk the boundary of problems, while [current] AI agents can fill the inside to some extent. If a problem has inherent "multi-dimensional boundaries", it might be hard for a person to imagine and walk it well to guide the agents. And I think most of the interesting problems fall in this category.