You know the dread. You look at a 65,000-line legacy codebase written in a language your team barely understands anymore. You calculate the human cost to rewrite it—months of engineering time, millions of dollars, endless bugs, and sleepless nights. Then, you run an AI script. A few hours later, it’s entirely in Rust. The cost? $400.
Recently, an experiment hit the feeds: someone used a tool called Fable to transpile 65,000 lines of Go into Rust. The internet’s reaction was predictable. Commenters complained the accompanying blog post was shallow, maybe even AI-written itself. They nitpicked the lack of technical depth.
They missed the point entirely. The writing didn’t matter. The prose could have been written by a toddler. The only thing that matters is the existential earthquake underneath the headline: a monumental human task was just reduced to a negligible API cost.
We’ve been obsessing over AI writing new code from scratch. We’re looking at the wrong disruption. The real earthquake is the death of language lock-in.
You’ve probably noticed how much of our industry is built on the sunk cost fallacy. “We can’t switch languages, the migration is too expensive.” “We can’t update the framework, the technical debt is too deep.” That entire economic model just collapsed. If a codebase ages out of relevance, you don’t hire a team of developers for a year-long rewrite. You pay a few hundred dollars to transpile it to the modern language du jour.
Technical debt is no longer a prison; it’s just a subscription fee.
Is the output perfect? No. The commenters rightly pointed out that direct ports have caveats, and unverified AI outputs can be shallow and buggy. But that’s a testing problem, not an engineering problem. You don’t need to verify the architecture; you just need to verify the execution. And AI is getting terrifyingly good at execution.
The era of the monolithic, immovable codebase is over. We spent decades drawing battle lines between Python, Go, Rust, and Java. We built entire identities and cult-like communities around our tech stacks. AI just poured gasoline on those lines and lit a match.
If your code can be translated for the price of a cheap office chair, your code isn’t a legacy. It’s just a draft.
We spent decades arguing about which programming language was the best. AI just made the question completely irrelevant.
FAQ
Q: What about the bugs? AI-generated code is notoriously unreliable.
A: Unverified AI code is a liability, but it's a testing problem, not an architectural one. You still need humans to verify behavior, but the months of manual translation and syntax errors are gone.
Q: Does this mean developers are obsolete?
A: No, it means developers stop being manual translators. The value shifts from writing syntax to designing systems, writing tests, and directing AI workflows.
Q: Isn't a direct port of a bad architecture just bad architecture in a new language?
A: Yes, but now you can iterate on that architecture at the speed of thought. If the new language doesn't fit your needs, transpile it again tomorrow. The cost of changing your mind is now zero.