i do think the code generation tools are going to replace a lot of software devs, and i say this as a person who was replaced by machine learning-based tools in my previous profession.
-
be honest: most software work is not maintaining the backend at a bank, or coding the firmware on a 737 Max. there are **tons** of low-stakes software jobs that could be eaten by automation. yeah, the results won't be as good as if a person is doing the work, but say it with me: if the margins are better and the error rate remains within acceptable parameters, they **will** automate the job.
and in turn, this loss of jobs at the bottom will put pressure on the wages at the top, because there will be more software developers chasing fewer software developer jobs.
-
and in turn, this loss of jobs at the bottom will put pressure on the wages at the top, because there will be more software developers chasing fewer software developer jobs.
now, there are a couple of wrinkles/caveats. first, bad/misbehaving software is often **very** expensive, so i think a lot of firms are going to find out that the slightly better margins actually aren't really worth the huge loses from having your payment processing system shit the bed on the busiest day of the year or whatever. the actual cost of slimming the dev team is going to be hard to gauge, and there may be some significant back-peddling.
-
now, there are a couple of wrinkles/caveats. first, bad/misbehaving software is often **very** expensive, so i think a lot of firms are going to find out that the slightly better margins actually aren't really worth the huge loses from having your payment processing system shit the bed on the busiest day of the year or whatever. the actual cost of slimming the dev team is going to be hard to gauge, and there may be some significant back-peddling.
second, we actually have no idea how much code generation truly costs. Anthropic et al. are giving everyone the teaser rate right now. they are definitely planning to take a huge chunk of whatever "savings" come from laying off the dev team, and at that point it may not even be worth it.
-
second, we actually have no idea how much code generation truly costs. Anthropic et al. are giving everyone the teaser rate right now. they are definitely planning to take a huge chunk of whatever "savings" come from laying off the dev team, and at that point it may not even be worth it.
i think it will eventually reach an equilibrium. a lot of the people who got into software development just for the money are going to move on to other things. the industry will probably over-correct and suddenly need to hire a bunch of new people who don't exist because people stopped studying CS, and wages will push back up. probably a mini-PC running an open-weight Chinese model for code generation will become a standard tool that every dev keeps on their desk.
-
i think it will eventually reach an equilibrium. a lot of the people who got into software development just for the money are going to move on to other things. the industry will probably over-correct and suddenly need to hire a bunch of new people who don't exist because people stopped studying CS, and wages will push back up. probably a mini-PC running an open-weight Chinese model for code generation will become a standard tool that every dev keeps on their desk.
but in the medium term, i think the ML automation dynamic i saw in translation is going to apply to software development. it's going to suck for a lot of people who are doing basic or low-stakes code monkey stuff. a lot of them are going to lose their jobs, and a lot more are gonna see salaries go down.
-
but in the medium term, i think the ML automation dynamic i saw in translation is going to apply to software development. it's going to suck for a lot of people who are doing basic or low-stakes code monkey stuff. a lot of them are going to lose their jobs, and a lot more are gonna see salaries go down.
@peter I think you may be right. An awful lot of people are going to be looking for jobs in the next 5 years
-
i think it will eventually reach an equilibrium. a lot of the people who got into software development just for the money are going to move on to other things. the industry will probably over-correct and suddenly need to hire a bunch of new people who don't exist because people stopped studying CS, and wages will push back up. probably a mini-PC running an open-weight Chinese model for code generation will become a standard tool that every dev keeps on their desk.
-
@kkarhan i hope you're right!!
-
@kkarhan i hope you're right!!
@peter I hope so too.
- Because if we allow the #Enshittification to continue and not flat-out ban *"#AI" code / "#VibeCoding" / #AIslop then at some point this shit will turn out to be the XKCD infrastruture meme, but with undried mud bricks that all melt together once a stray dog pisses on them!
-
second, we actually have no idea how much code generation truly costs. Anthropic et al. are giving everyone the teaser rate right now. they are definitely planning to take a huge chunk of whatever "savings" come from laying off the dev team, and at that point it may not even be worth it.
@peter It's going to be pretty damned funny if the cost of these AI tools ends up being MORE than a junior dev, though. "Oh, you want it to compile *and* pass tests? Yeah, that'll be 5 more agents, bringing your total bill to around.. $150k/yr."
-
@peter It's going to be pretty damned funny if the cost of these AI tools ends up being MORE than a junior dev, though. "Oh, you want it to compile *and* pass tests? Yeah, that'll be 5 more agents, bringing your total bill to around.. $150k/yr."
@peter Or alternatively, "oh that'll be $30k for the code and.. $50k for the agent pen testing that actually ensures the code is secure, and another $20k for.."
Maybe it'll be like the cloud where the base thing is cheap but when you actually try to make it useful, it suddenly gets very very expensive.
-
but in the medium term, i think the ML automation dynamic i saw in translation is going to apply to software development. it's going to suck for a lot of people who are doing basic or low-stakes code monkey stuff. a lot of them are going to lose their jobs, and a lot more are gonna see salaries go down.
@peter I think in our current era, robots eventually settle into tasks they're best at. When the #AI gold rush ends, there'll be a sorting out based on costs.
The end state of each era of programming are stable protocols & libraries that reduce overhead. Today's infinite code generators go in the wrong direction. Infinite monkeys will ultimately iterate back to reality where humans are still relevant or even favored. The last mile is only a problem if you're excluding humans.
That last mile is metaphorical. Yesterday it was about installers paid to lay down fiber. Tomorrow it may refer to humans implementing complex tech more efficiently because we came out of the box with our own integrated sensors.
There'll be upheaval, but that's always been the case for this industry. As a freelancer, I feel it even more. I have zero job security & my value vanishes every 10 years. On Twitter, I used to bark at young devs gushing about their paychecks to save their money for winters they couldn't see yet.
-
@peter Or alternatively, "oh that'll be $30k for the code and.. $50k for the agent pen testing that actually ensures the code is secure, and another $20k for.."
Maybe it'll be like the cloud where the base thing is cheap but when you actually try to make it useful, it suddenly gets very very expensive.
@Andres4NY that's what i'm expecting. the question is whether they can keep the balls in the air long enough to destroy the dev pipeline to the point that people **have** to pay for their robots. i don't think so, but we'll see.
-
be honest: most software work is not maintaining the backend at a bank, or coding the firmware on a 737 Max. there are **tons** of low-stakes software jobs that could be eaten by automation. yeah, the results won't be as good as if a person is doing the work, but say it with me: if the margins are better and the error rate remains within acceptable parameters, they **will** automate the job.
@peter Aside:
An atrocious or banal truth is that a lot of the people doing the work are doing a job that is not any better than the code generator
Many people I’ve worked with in this racket are “expert beginners” who are like the doctor who practices one year of medicine twenty times; they want the instant high of a quick win and an attaboy for a hello world
They get really angry when I suggest an improvement because I’m saying that they didn’t make it perfect on the first try
-
R relay@relay.infosec.exchange shared this topic
