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.
I have sometimes turned to Reddit to find out things about clothing and recommendations for clothes, but I find the general consensus about "if you want x, the best place to buy is y" is pretty rubbish.
The best thing you can learn is how to treat clothes, as you mention.
Looks good! Probably just a personal preference thing, but I was looking for a back button or way to swipe the “Now playing” panel down in order to see the list of radio stations again. You can use the nav at the bottom, but a back button or swipe down was my natural attempt at first.
This is exactly why I never trust 'accidental' feature rollouts in production environments. Any CDN or hosting provider that can accidentally expose private data has serious issues with their deployment process and access controls. Railway should have caught this in staging - the fact that it made it to production suggests they don't have proper testing for permissioning logic, which is absolutley terrifying for anyone hosting sensitive apps there.
Great article. I still consider myself "engineering" while exclusively using AI tools, as I'm making the same changes to the code than I would have done without AI.
There's still the same level of accountability, and the same deep understanding of what problems are being solved and the approaches being taken.
But the whole "prompt" prefix seems to give the engineering aspect less meaning. I think that's just negative connotation though.
The token usage differs day to day - that's the most frustrating part. You can't effectively plan a development session if you aren't sure how far you'll likely get into a feature.
I somewhat agree with the deploy fear point, but deploy fear can just be a consequence of one bad experience 10 years ago that ruined a weekend. At least for me :D
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.