> AI reads Perl so Python devs don't have to, saving 30 hours per dev over 9 months.
An example of a non hand wavy, actual not hyped use of 'AI'.
iforgotpassword · 1d ago
And luckily the "plain English" the AI outputs is always 100% correct, so we don't have to worry about buggy python code down the line because those python devs got incorrect instructions. I mean how should they even verify anything, they're python devs, and perl will look like complete gibberish to them.
So they might have saved all that time, but what's gonna be the impact of incorrect reimplementation? What does that software do?
zach_miller · 1d ago
Ultimately it seems the question ought to be “is the code they wrote with AI buggier than the code they would have written without”, not “is the code they wrote with AI 100% bug free”. I doubt that any team doing a significant refactor from a language they don’t know could make bug free code on any reasonable timeline, AI or not.
If the question is the former, though, unless it’s horrendously buggy then I wonder if the speed increase offsets the “buggier code” (if the code even is buggier) because if they finish early they can bug bash for longer.
iforgotpassword · 1d ago
I guess it depends on wether the devs are able and willing to even still try to look at the old code, when they have a nice and easy to understand description in front of them what they're supposed to implement. And sure, at the end of the day management just cares about what costs less, including any accidents caused by AI giving the wrong description. Might also depend on who'll use that tech. If it's a bank, this could cost millions, if not billions. If it's a medical device (yeah I really don't think it is. I mean I really hope it isn't), it could cost lives. But at least then we can blame the AI, so nobody is at fault.
hyperhello · 1d ago
It’s only supposed to work long enough for the next guy to take over.
potato3732842 · 1d ago
Existing QA and deployment best practices should mostly answer the question of "does the new one work like the old one". The difference here is that now the devs can understand what the new one ought to do much faster.
zeroonetwothree · 1d ago
So it’s a 0.4% savings. Seems plausible at least. Unlike all the “5x faster” claims you see frequently.
hedora · 1d ago
Later, they claim 15 hours per dev per week. It’s almost like the numbers are completely made up!
Also, why rewrite perl code in python? Those two languages have basically the same set of problems, assuming the perl code is doing banking stuff and not orchestrating GPUs.
If you’re going to pay for a rewrite, at least get some sort of high level improvement, like richer static typing.
I’d guess the goal is “python devs are cheaper than perl devs”? How long is the return on investment vs. site-licensing Learning Perl?
potato3732842 · 1d ago
>Also, why rewrite perl code in python? Those two languages have basically the same set of problems, assuming the perl code is doing banking stuff and not orchestrating GPUs.
Because at some point it's no longer worthwhile to continue hacking up stuff build with the assumptions and best practices of 20yr ago and just want to write a new one that's built with today's assumptions and best practices and will be easier/cheaper for everyone who has to interact with for the next 20yr.
buttrpopcorn · 1d ago
> Though the tool can, in theory, rewrite code from an older language to a newer one, it doesn't know how to write the new code efficiently or as well as a human developer, Pizzi said.
From what I've seen, one popular use of coding agents is to refactor code. Wouldn't it be easier to create a non-ai scanner, parser, translator for the perl codebase and let AI run and refactor it? At that point human devs can come in, look out for inefficiencies and tackle bugs on a nearly-functional codebase?
GuB-42 · 1d ago
> The tool, called DevGen.AI, translates code in older languages, such as Perl (released in 1987), into plain English, which developers can then use as a basis for rewriting the code into newer languages like Python.
... (released in 1991). Yes, Perl is much older /s
Both Perl and Python are from the same generation, and both are active and both have their latest release in 2025. It is just that Python is more popular now.
The real reason is that besides losing in popularity, Perl is notoriously difficult to read, and yet has a huge existing code base, so I guess it is a good target for an AI parser.
motorest · 23h ago
> Both Perl and Python are from the same generation, and both are active and both have their latest release in 2025.
I don't think that's what they mean by "Python" or "Perl". It sounds like they have a sizeable ancient Perl codebase that they want to replace with Python due to practical reasons. No one cares if Perl's first release was a couple of years before Python's. What matters is the company's ability to maintain and extend their codebase.
cibyr · 1d ago
Also that Python has successfully transitioned its community to Python 3, while Perl 6 flamed out, changed its name, and left the Perl community slowly atrophying.
An example of a non hand wavy, actual not hyped use of 'AI'.
So they might have saved all that time, but what's gonna be the impact of incorrect reimplementation? What does that software do?
If the question is the former, though, unless it’s horrendously buggy then I wonder if the speed increase offsets the “buggier code” (if the code even is buggier) because if they finish early they can bug bash for longer.
Also, why rewrite perl code in python? Those two languages have basically the same set of problems, assuming the perl code is doing banking stuff and not orchestrating GPUs.
If you’re going to pay for a rewrite, at least get some sort of high level improvement, like richer static typing.
I’d guess the goal is “python devs are cheaper than perl devs”? How long is the return on investment vs. site-licensing Learning Perl?
Because at some point it's no longer worthwhile to continue hacking up stuff build with the assumptions and best practices of 20yr ago and just want to write a new one that's built with today's assumptions and best practices and will be easier/cheaper for everyone who has to interact with for the next 20yr.
From what I've seen, one popular use of coding agents is to refactor code. Wouldn't it be easier to create a non-ai scanner, parser, translator for the perl codebase and let AI run and refactor it? At that point human devs can come in, look out for inefficiencies and tackle bugs on a nearly-functional codebase?
... (released in 1991). Yes, Perl is much older /s
Both Perl and Python are from the same generation, and both are active and both have their latest release in 2025. It is just that Python is more popular now.
The real reason is that besides losing in popularity, Perl is notoriously difficult to read, and yet has a huge existing code base, so I guess it is a good target for an AI parser.
I don't think that's what they mean by "Python" or "Perl". It sounds like they have a sizeable ancient Perl codebase that they want to replace with Python due to practical reasons. No one cares if Perl's first release was a couple of years before Python's. What matters is the company's ability to maintain and extend their codebase.