Back in 1967, a guy named Gene Amdahl basically said: look, if you throw more computing power at a fixed job, you'll hit a wall. Some parts of the job just can't be sped up, so no matter how many extra cores you bolt on, you're stuck with diminishing returns forever. Amdahl's Law is the sound of hitting a ceiling. It's "same job, faster, until it isn't."

Then in 1988, John Gustafson looked at the same idea and said, essentially, "speak for yourself brother" His argument: give people more compute, and they don't just finish the same job quicker and clock out early. They go bigger. More power, more ambitious problem, same time on the clock. Give a scientist a beefier supercomputer and they won't run today's simulation in ten minutes and go home, they'll run a way more ambitious simulation and still be at their desk until five.

I'm firmly on Team Gustafson,  I’ll explain why.

The Amdahl version: "You raced to the airport in a Bugatti "
“But you’re still stuck at the gate”

Everyone's pitch for AI coding tools is basically the same: it makes you faster at the thing you were already doing. Same backlog, same tickets, just zoom through them quicker. Pure Amdahl. Faster, sure. But fundamentally the same job, same ambitions, same ceiling.

That's not really what happened to me.

The pile nobody wants to touch

If you've worked test automation for more than five minutes, you know the pile I'm talking about. Repos handed down from people who left years ago. Copy-pasted helper code everywhere because nobody had time to do it properly. The same fix applied in one place and forgotten in three others. Everybody agrees it's a mess, everybody knows it should be cleaned up, but it never happens, because "cleaning it up properly" is a big undertaking that competes with actual sprint work, and it always loses.

That's the thing about technical debt: it's rarely a mystery what the fix is. It's a capacity problem. You know exactly what "good" looks like. You just never have the room to go build it.

The Gustafson Version: “The house party protocol, sir?“
“Jarvis – Iron man 3”

This is basically what it felt like once I had a proper AI assistant in the loop, not as autocomplete but as an actual second set of hands. The technical debt cleanup nobody in the team ever wanted to tough, quietly turned into a “lets do it now”-project. Not because I was typing faster, but because the thing that felt too big to attempt suddenly didn't. Harmonising code across repos that had been drifting apart for years, untangling duplication, actually organising things the way you yourself would design them as if you were starting fresh, all of that moved from "a nice idea" to "just do it” Repo by repo, piece by piece, at a pace that would've been unattainable solo.

The scope grew to fill the extra capacity. That's the whole law, right there, and it's a much better story than "I closed my tickets 20% faster."

Same hours, very different ambition

If you'd asked me to measure this the boring way, how much quicker am I closing the same old tickets. Somewhat faster, nothing to write home about. But that's not the interesting metric.

The interesting metric is that the project changed shape entirely. I didn't use the extra help to blast through my usual to-do list. I used it to finally go after the stuff that had been sitting there for years, the reorganising, the harmonising, the "we all know this needed to be done" work that never survives contact with a sprint planning meeting. That's Gustafson's Law showing up in a job that has nothing to do with supercomputers: same hours in the day, wildly bigger ambitions for what fits in them.

So if you're sitting on your own pile of "we'll get to it eventually", old code, duplicated logic, a mess nobody's proud of, then maybe the right question isn't how much faster AI can get you through today's work. It's the bigger, previously-unaffordable job it finally lets you start.