GenAI tools are Quitters

Building out an enterprise-grade ML/AI platform on Google Cloud has taught me many things, but here’s one I didn’t expect: GenAI tools are quitters.

Case in point: I was developing a custom admin dashboard for system monitoring when I hit repeated access issues (turned out to be a subtle access scope misconfiguration). But long before I figured that out, ChatGPT was ready to throw in the towel: “Try starting over with a brand new VM image.”

Meanwhile, Claude took a different approach when I couldn’t figure out how to create a centralized logging mechanism:  “Yeah, that’s not working. Let’s ignore it and build a half-baked, half measure and maybe we’ll figure it out later.”

It struck me: these tools, for all their brilliance, lack one essential ingredient, perseverance. They don’t feel that itch to make something work. There’s no sense of “Yes … I did it!”

That’s still our job. And maybe that’s the point: Humans are still the ones who debug because deep down we find satisfaction in our accomplishments; it’s a human instinct to solve.

Still, I’ll take the speed boost … as long as I keep the steering wheel.