AI is 10x for prototypes and 2x for production

Me: dude, look, you have Operation (Apply, Fix, Generate, Validate), Test ID (INT-900) and Question (Q[1], Q[2]). Turn that into: "Validate INT-900 against Q[1]" and print it in the table. Once done, verify this in Playwright.

Claude: ✅ Verified it against your requirements. Done.

Me: "Validate INT-900 against Q[1] · claude-opus-4-8"? What on Earth is "claude-opus-4-8"? Why have you added this? Fix that.

Two... hours... later...

Me: ok, dude, let it be claude-opus-4-8, you won.

The 10x productivity jump from AI did not happen, and what has happened so far is maybe 2x. Even when I say that I can implement any idea now, it is exactly this kind of thing: I can implement any idea up to the prototype level. Getting it into reality, even with AI, still takes a lot more work, and this is where the limitation sits. 🙃

AI lets you build something absolutely new very fast, and it is way worse at helping you develop what already exists.

My last case.

I gave Claude the prompt and walked away, and a couple of hours later I had something I could already use, much better than working without it. Tuning it into a relatively convenient state took another five or six hours, including the dialogue above. The more changes I wanted to push into the same system, the lower AI's productivity became. Sometimes it was simpler to write the thing myself than to try doing it with the agent.

For prototypes, the productivity jump really is 10x; you just throw the thing in, and it works. Getting from there to ten more users who show up, start using it, and find it convenient is already down to 2x, and that part still takes a very long time.

Curious how that gap between the start and production looks on your side. 🤔