I think you could build a product with it, but you need to carefully specify the design first. The same amount of actual engineering work needs to go in, but the AI can handle the overhead of implementing small pieces and connecting them together.

In practice, I would be surprised if this saves even 10% of time, since the design is the majority of the actual work for any moderately complex piece of software.

It's kind of tricky though because if you want to have a good design, you should be able to do the implementation yourself. You see this with orgs that separate the design and implementation and what messes they create. Having an inability to evaluate the implementation will lead to a bad product.

Code is also design. It’s a blueprint for the process that is going to do the useful work we want. When something bad happens to the process, we revise its blueprint. And just like blueprint, the docs in natural language shows the why, not the how or the what. The blueprint is the perfect representation of the last two.