12 comments

[ 23.9 ms ] story [ 2950 ms ] thread
If there are 4500 years of developer work (not including "real work") it sounds like there are bigger problems.

The whole point of having developers do upgrades isn't just the manual work but so they learn and understand it. Skipping that isn't helpful.

And why would you wait until there's 4500 years of work before upgrading? Maybe the problem is that company policies need to be fixed rather than introducing AI.

At a guess, it's about drudge with relating to changed APis, and 4500 is a period dictated by code review procedures.

I wrote an AI, oops, a shell script to handle such an upgrade decades ago.

Yeah let's see if their code lasts 4500 years unchanged.

Also

Imagine having to debug something that has 4500 years of lines of code.

Ancient Egyptian comes to mind. And we couldn’t figure it out until we found the Rosetta Stone.
Saved our developers working from 2476 B.C. to 2024 A.D. (4,500 Years of Developer Work)
Slightly off topic, but my mind went there.

Bezos could spend $41 million dollars a year since 2476BC or $112k a day (all without earning interest) , and he'll still have change.

What is $112k worth in 2476BC, adjusted for inflation? ;)
I've counted exactly 659 times when a statistic was used to sell something. Exactly 659.
I think he has to say 4500 years because the one who is only willing to say 4000 years gets fired?
If you have a backlog of 4500 “developer years” worth of upgrades, lack of AI wasn’t the problem, and making some attention nodes go brr cannot solve the root problem.

Java is a super verbose language, yet keeps getting written because “everyone knows java.” Has the cost of maintaining Java finally reached such critical mass that only an LLM can affordably maintain it?

I save a lot of work code generating the non-biz-logic parts of applications from simple declarative specs. I typically end up having to only hand code about 10% of the application.

No AI required.