I think the "middle" isn't empty, it just requires more discipline than either extreme. AI works well when the problem is clearly framed and constrained, and fails when we skip steps and expect it to fill in the gaps correctly.
Your pipelne is basically a way to force good engineering habits. Thinking through the problem, design, and spec before coding, which reduces hallucinations and keeps understanding intact. So a big part of the improvement comes not just from the AI itself, but from the fact that it pushes you to be more structured and explicit in how you think about the problem.
Thanks for the write-up! Comprehension debt definitely feels like one of the biggest issues when I'm leaning heavily on coding agents.
It might fall into the too-heavy prior art category you mention at the end, but I quite like Superpowers (https://claude.com/plugins/superpowers and wrote a short review of why I like it: https://emschwartz.me/a-rave-review-of-superpowers-for-claud...).
I think the "middle" isn't empty, it just requires more discipline than either extreme. AI works well when the problem is clearly framed and constrained, and fails when we skip steps and expect it to fill in the gaps correctly.
Your pipelne is basically a way to force good engineering habits. Thinking through the problem, design, and spec before coding, which reduces hallucinations and keeps understanding intact. So a big part of the improvement comes not just from the AI itself, but from the fact that it pushes you to be more structured and explicit in how you think about the problem.
Great read, thanks.