For years, the primary obstacle to my ideas was resources. I had the design instinct. I had the taste. I didn't have the hands. Every project past a certain complexity would stall at the same point: too much implementation for one person, not enough budget to hire, no team to delegate to.
AI coding assistance closed that gap. The result wasn't what I expected.
I expected to ship more of the same, faster. What actually happened is that my job changed. The implementation loop that used to eat 80% of my week is now closer to 20%. The remaining 80% is architecture, DSL design, release sequencing, client conversations, product direction. In effect, I stopped being an engineer with an architect's ambition and became an architect with engineering leverage.
Three specific shifts I've noticed:
Idea exploration got cheap. Prototypes that used to be "someday" work are weekend work now. This is not a small change. It rewrites which ideas are worth having.
Cross-domain confidence went up. Not being a specialist in a particular stack no longer blocks entry. I can start a project in a language I haven't written production code in, because the assistance covers the surface layer while I focus on the shape.
The failure mode inverted. My old failure was not enough hours. My new failure is not enough clarity of objective. When implementation is cheap, muddled goals get built anyway, and you find out only after they're running.
Important caveat: this only works if you already have architectural taste. AI amplifies existing judgment. It doesn't create it. Someone without a strong sense of what "good" looks like will produce more code faster and end up with a bigger mess. The bottleneck genuinely does move upstream, but if there's nothing solid upstream, the leverage works against you.
The people who benefit most from this shift aren't the strongest engineers. They're the people who were already thinking like architects but were stuck at the keyboard.
Drafted in July 2023
Updated for site in 2026