[p1esk]

I don’t know anyone who’s looking for junior devs today. That was not the case 3 years ago.

Re LLMs: most people responding here seem to assume there will be zero progress in ML after today. I don’t understand why. It’s likely we will get gpt-5 later this year, followed by Opus 3.5 and Ultra 1.5. I expect all three to be significantly better at coding than current models. Again, all three are expected within the next 6 months. Next gen after that (gpt-6 level): 2025-2026. Again, I don’t see any reasons not to expect further improvements. At that point (2026 at the latest) it will be strange to pay humans to write code, at least in the typical way we view SWE role today.

[b20000]

define “better at coding” and also estimate cost of fixing generated solutions

[p1esk]

“Better at coding” as defined by seniority levels: GPT4 - an intern; GPT5 - a junior developer; GPT6 - a senior developer.

Yes, GPT5 will do a lot of damage, as it will be the first model replacing actual human coders. But GPT6 should be able to fix most of the issues. During this transition period I expect human senior developers mostly doing code reviews.

[meiraleal]

By this logic, Photoshop 6 or 7 should have replaced designers.

[p1esk]

It will, as soon as it becomes “powered by GPT6”.

[b20000]

i have seen no proof of actual reasoning or intelligence

[trod123]

I think this is not actually a reasonable marker to compare against in this case.

Neither of these are actually needed, all that is needed is to be able to disrupt the pipeline of jobs that interns/junior developers would normally build their initial skills on before moving on to more senior positions. Cutting these entry positions out means continually fewer replacements, and less talent (a deflationary trend, the limited number of positions acts as a filter, and its not necessarily merit based).

Without replacement, eventually older more senior developers will age out, retire, and then everyone will panic but by then it will be far too late (its already cascaded).

If it is not economical to develop or maintain a skill set, the knowledge involved becomes lost knowledge when that population dies out (as we all do given sufficient time).

A perfect corollary is the repair of electronics in the US. There might be 1 person per population of 100-200,000 that goes into business for this sector. If they get hit by a bus, its no longer locally available, and mortality happens to us all.

This is one of the main structural problems with burning bridges in blind marches towards some idealized progress.

How can you go back when all your competitors do the same thing and its no longer economic to run the business except by following the same doom loop as your competitors.

[cultofthecow]

Have you seen at leas one tool that is capable to replace a junior dev? At least one please. I would love to read anything about it.

[p1esk]

No, GPT5 is not out yet. As I said, it should be out within 6 months. Please be patient.

[cultofthecow]

Do you think it will be capable to replace a junior dev?

[p1esk]

Yes. It will go like that: “should we hire a senior dev and a junior dev? Nah, a senior dev plus GPT5 subscription should be enough to do the workload”. And it will be enough.