We don’t need to rebuild the ladder, and we don’t need juniors. By the time the seniors leave the job market, the software engineer profession as we know it simply won’t exist. I very much doubt that there will be the higher level architect position either. If Fable 5 is anything to go by, we’re all replaceable within 12 to 24 months. The rest is social inertia.

▲

QuiEgo 2 hours ago | parent | next [-]

I’m always struck by a bit of wonder at comments like this. It seems everyone’s experience is all over the place. Curious, what types of things are you working on where you see these results?

I’m at 90%+ code AI generated by stats. I work in embedded systems. It still goes off the rails all of the time and needs a heavy hand to guide it. It does not currently feel like it will ever be truly able to operate independently. It’s a very useful tool, but it’s just not there yet in my day-to-day.

Obviously, YMMV.

▲

MattPalmer1086 7 hours ago | parent | prev | next [-]

How will AIs train on new tech without new data to train on?

On the bright side, maybe that means the end of new javascript frameworks every 6 months :)

▲

Chance-Device 6 hours ago | parent | next [-]

I think they’ll be writing that new tech, and will be able to read the source code, and reason from first principles. I doubt they will need human generated training data.

▲

MattPalmer1086 6 hours ago | parent [-]

I thought training new models on AI generated output leads to model collapse?

▲

aytigra 8 minutes ago | parent | next [-]

I think it case of coding it may not be as bad, because new training data (AI generated code) is always empirically validated by tooling and by consumer. It may not be good but it mostly works, otherwise it is discarded or patched, so it has a bottom bar of "it works".

▲

Chance-Device 6 hours ago | parent | prev [-]

My point is more that they won’t need new training data to extrapolate to new problems, especially if that new problem is just a new syntax or API. Put the whole shebang into the context. Done. Yes, you need super long context for this, or just pretty good search over it, but I think this is likely to become a solved problem.

	▲	MattPalmer1086 5 hours ago \| parent [-]
		Agreed if its just a new API using a current language using common patterns, that is mostly just having a big context. However, the ability to reason generally on novel problems is AGI and we aren't there yet. Eventually in the absence of AGI, we will have to train models on them, and that will require data.

▲

charcircuit 5 hours ago | parent | prev | next [-]

If needed it can synthetically create the datasets. There isn't a need for software engineers to make a dataset when AI could even more easily put one together.

▲

mikert89 5 hours ago | parent | prev [-]

Do you understand reinforcement learning environments? This already a solved problem

▲

lurking_swe 6 hours ago | parent | prev | next [-]

just the other week I asked Fable 5 to diagnose the cause of some intermittent latency spikes on an API that queries an OpenSearch cluster at work. I encouraged it to look at the datadog metrics, splunk, the whole works. I let it loose to look at whatever it wanted.

End result - 2 hours later it produced a convincing theory with lots of references, and burned a bunch of tokens too of course. just for fun we tried its suggestions and deployed them to prod. Guess what? Didn’t fix the issue. Alas, a human was needed after all.

either everyone’s working on toy problems, or they’re working on very cookie-cutter code. I’m really not sure. I DO remain impressed with Fable 5 but the idea that we’ll all be unemployed in 2 years is hilarious delusion. we’re already at the point where many organizations are scaling back some of their AI spend.

▲

Chance-Device 6 hours ago | parent [-]

By any chance, is the following true: you had no empirics in the loop. It couldn’t validate any of its theories by experimentation. If the solution requires being able to make actual changes in order to gather more information and it is not allowed to, and this is the only way to solve the problem, by definition it couldn’t do it, nor could you. Or was it something that could be worked out entirely on an a priori basis from the available data?

And while we’re talking about hilarious delusions, perhaps you should look at the current capability curve of AI and weigh it against the constant stream of arguments for why it couldn’t have continued at every point and yet has.

▲

lurking_swe 4 hours ago | parent | next [-]

i’m not an anti-AI believer. And our career is changing dramatically no doubt. i just think people overestimate short term gains and underestimate long term gains. (applicable to your comment)

In this case i had no empirics in the loop. The scenario was only reproducible under high api load. I could load test, but management isn’t eager to spend prod-like costs in staging (requires scaling opensearch a lot in stage). What can i say.

▲

surgical_fire 5 hours ago | parent | prev [-]

> in the loop

Yes, the solution is to just burn more tokens.

	▲	Chance-Device 5 hours ago \| parent [-]
		Would you care to say something substantive, or is that just not a thing you do?

▲

whazor 7 hours ago | parent | prev | next [-]

Actually I believe these agentic models will teach you the value of software engineering faster. You can vibe entire code bases in days and learn more quickly.

In my experience with all agents, including Fable, is that they work great when there is automated validation. But as soon as it needs to design something, it just keeps adding so much slop.

	▲	Chance-Device 6 hours ago \| parent [-]
		That’s just the latest position that the goalposts have moved to. They haven’t stopped improving yet, I have no reason to think they’ll stop at the ideation phase. Or any of them, really.

▲

mikert89 5 hours ago | parent | prev | next [-]

I think it’s 24 to 36 until businesses really trust an autonomous developer. But I agree otherwise

	▲	Chance-Device 5 hours ago \| parent [-]
		Probably, might take even longer. That’s the “social inertia” part. I think it’s less trust and more status quo bias.

▲

pixel_popping 7 hours ago | parent | prev | next [-]

Fable 5 will be genuinely weak compared to what's coming, I mean, we need to remember this is kinda the beginning still, we will genuinely reach a point where all benchmarks will score 99.9%. Think Opus 10, GPT-10... :)

Also Fable 5 isn't "that impressive" as a lot of people have that kind of intelligence since 6 months+ by using combo of models and loops (I scored better on HLE than gpt-5.5 xhigh last January with some good tooling and 6x the cost), but for a lambda Claude Code user, I can see why it looks that good.

▲

Chance-Device 7 hours ago | parent | next [-]

Are you using it or are you just going off benchmarks?

▲

PestoDiRucola 6 hours ago | parent | prev [-]

What makes you think that models will improve with the same pace that they have been improving in the past few years?

	▲	petra 5 hours ago \| parent [-]
		A few reasons: -2.5 years is a pretty short time for a new tech development, even if it fails eventually - usually when a new tech is introduced, the biggest gains happen when the environment is changed to fit it. That takes time: libraries, api's, verification tooling, rl environments, skilling users, etc. - possibility of orders of magnitude hardware cost reduction - Optical. Analog. Rram. Many others. Something will work. And internal improvements in the model architecture. And there's scaling in reasoning time.

▲

boshalfoshal 6 hours ago | parent | prev [-]

wow an actual ai-pilled comment on here for once, I agree with your sentiment. People opining about "rebuilding the ladder" have no idea whats coming for the software industry, and the general populous of white collar work.

"Models can code well now but they cant do high level architecture" is just a logical fallacy. Its literally only true in this particular moment in time. But if they can code well, whose to say they wont architect well? And at that point, what do SWEs do? If anything, SWEs are in the critical path of automation for these AI labs anyway, so theres a very strong incentive to automate us out vs other professions, and it'll happen soon. All these random 1-off datapoints of "Fable 5 can't do X very idiosyncratic thing" are completely missing the point. 6 months ago, even attempting that problem with any "tool" would be totally intractable, and now it _just_ writes a slightly subpar solution. You can do some basic extrapolation here, its not that complicated.

Your best bet is to just chose a different career, or, if you still want to be in the software industry, be more enterprising.