To replace all developers, we need AGI yes. To replace many developers? No. If one developer can do the same work as 5 could previously, unless the amount of work expands then 4 developers are going to be looking for a job.
Therefore, unless you for some reason believe you will be in the shrinking portion that cannot be replaced I think the question deserves more attention than “nothing”.
Frameworks, compilers, and countless other developments in computing massively expanded the efficiency of programmers and that only expanded the field.
Short of genuine AGI I’ve yet to see a compelling argument why productivity eliminates jobs, when the opposite has been true in every modern economy.
> Frameworks, compilers, and countless other developments in computing
How would those have plausibly eliminated jobs? Neither frameworks nor compilers were the totality of the tasks a single person previously was assigned. If there was a person whose job it was to convert C code to assembly by hand, yes, a compiler would have eliminated most of those jobs.
If you need an example of automation eliminating jobs, look at automated switchboard operators. The job of human switchboard operator (mostly women btw) was eliminated in a matter of years.
Except here, instead of a low-paid industry we are talking about a relatively high-paid one, so the returns would be much higher.
A good analogy can be made to outsourcing for manufacturing. For a long time Chinese products were universally of worse quality. Then they caught up. Now, in many advanced manufacturing sectors the Chinese are unmatched. It was only hubris that drove arguments that Chinese manufacturing could never match America’s.
I have a coworker who I suspect has been using LLMs to write most of his code for him. He wrote multiple PRs with thousands of lines of code over a month.
Me and the other senior dev spent weeks reviewing these PRs. Here’s what we found:
- The feature wasn’t built to spec, so even though it worked in general the details were all wrong
- The code was sloppy and didn’t adhere to the repos guidelines
- He couldn’t explain why he did things a certain way versus another, so reviews took a long time
- The code worked for the happy path, and errored for everything else
Eventually this guy got moved to a different team and we closed his PRs and rewrote the feature in less than a week.
This was an awful experience. If you told me that this is the future of software I’d laugh you out of the room, because engineers make enough money and have enough leverage to just quit. If you force engineers to work this way, all the good ones will quit and retire. So you’re gonna be stuck with the guys who can’t write code reviewing code they don’t understand.
In the short term, I share your opinion. LLMs have automated the creation of slop that resembles code.
In the long term, we SWEs (like other industries) have to own the fact that there’s a huge target on our backs, and aside from hubris there’s no law of nature or man preventing people smarter than us from building robots that do our jobs faster than us.
That’s called class consciousness, and I agree, and I think most people have already realized companies are not their friend after the last two years of layoffs.
But like I said, I’m not worried about it in the imminent future, and I have enough leverage to turn down any jobs that want me to work in that way.
I think not only is this possible, it's likely for two reasons.
1. A lot more projects get the green light when the price is 5x less, and a many more organizations can afford custom applications.
2. LLMs unlock large amounts of new applications. A lot more of the economy is now automatable with LLMs.
I think jr devs will see the biggest hit. If you're going to teach someone how to code, might as teach a domain expert. LLMs already code much better than almost all jr devs.
As we are able to automate more and more of the creation process, we may be able to create an infinite amount of software. However, what sustains high wages for our industry is a constrained supply of people with an ability to create good software.
I think counting the number of devs might not be the best way to go considering not all teams are equally capable or skilled in each person, and in enterprises, some people are inevitably hiding in a project or team.
Comparing only the amount of forward progress in a codebase and AI's ability to participate or cover in it might be better.
I'm not sure it is as simple as that - Induced Demand might be enough to keep the pool of human-hours steady. What that does to wages though, who can say...
Well it certainly depends on how much induced demand there is, and how much automation can multiply developer productivity.
If we are talking an 80% reduction in developers needed per project, then we would need 5x the amount of software demand in the future to avoid a workforce reduction.
That's a weird way to look at it. If one developer can do what 5 could do before, that doesn't mean I will be looking for a job, it means I will be doing 5 times more work.
Therefore, unless you for some reason believe you will be in the shrinking portion that cannot be replaced I think the question deserves more attention than “nothing”.