You were able to code. A lot of what you describe was never “writing code,” it was wrestling with entropy: build tools, dependencies, stale repos, vague errors.
AI mostly speeds up the googling + trial-and-error loop. The parts that still separate “I can code” from “I can get something to run” are knowing what you actually want (edge cases, failure modes), and being able to sanity-check it when it’s wrong in a subtle way. If you can explain why it works when it works, and how it fails when it fails, you’re still coding — the tool just writes more of the boring parts.
AI mostly speeds up the googling + trial-and-error loop. The parts that still separate “I can code” from “I can get something to run” are knowing what you actually want (edge cases, failure modes), and being able to sanity-check it when it’s wrong in a subtle way. If you can explain why it works when it works, and how it fails when it fails, you’re still coding — the tool just writes more of the boring parts.