Are there any sources of success stories for AI-driven hiring reductions from a long-term perspective?

Hello, I'm Munou.

I've been thinking about this quite a bit lately.

What I thought in the past

What I thought when I wrote an article about Stable Diffusion about three years ago is recorded here.
【Stable Diffusion】AI Image Generation is a Psychedelic Space You Can View Sober [With Various Images] - SOULMINIGRIG

Eventually, since Amazon and Google provide cloud GPUs (which are quite expensive), I thought that a next-generation search engine using AI and cloud GPU power might be possible. After all, unless you're a geek, as long as you can get text or images that contain the information you're looking for, it doesn't really matter who wrote it.

If everyone were "geeky," they would seek out humanity and things with character; otherwise, the future result might make us wonder, "Is there really a need for humans?" In a way, it's like a vibe of "Humans, try harder!" It reminded me of The Clash's "White Riot," which was used in the context of racial issues at the time. (The author does not hold discriminatory views toward any group.)

I'm a supporter of traditional things like photographers and artists, but I thought that by actively engaging with AI like this, we might be able to find the inherent value of humans.

Since it feels like it's becoming a substitute for search engines nowadays, I suppose you could say it was somewhat correct in some respects...

Articles that caught my eye

If You Stop Hiring Juniors, Your Senior Engineers Own You · eval ( code )

If you stop hiring junior engineers, senior engineers will rule your company

So where do they come from? They don't just appear as fully formed talent from the start. They start their careers as juniors and spend years growing into those roles. If you cut off the talent pipeline, you cut off the supply of the very people you need. Just ask anyone trying to hire a COBOL engineer today. The talent pipeline dried up decades ago, and the few remaining individuals command high fees themselves.

This is something that has actually happened in the short history since computers were born. Perhaps, instead of going out of your way to hire a COBOL engineer, you hire an engineer with no COBOL experience and have them perform modifications using AI. However, if you were actually in the position of the hirer, and you were to replace an entire critical and massive financial system, would you choose someone who has never touched COBOL at all?
Some might say yes, others might say no. Since most people are not in a situation where they are operating a COBOL environment anyway, it's highly likely that this is just talk.

There are currently no actual success stories or sources from a long-term perspective

It has been about a year since coding agents like Codex, Claude, etc., began to permeate the mainstream, and there are currently no success stories or sources spanning several years or decades.
In this case, the risk from an organizational standpoint is significant, as it deviates from the principle of learning from history.

As for volatility risk, although it was likely rarely used at the corporate level, Sora suddenly shut down its service. Furthermore, costs related to API call limits and fees can no longer be quantified due to repeated limit increases and resets.
And currently, the exponential growth of LLMs continues unabated amidst the competition between OpenAI and Anthropic. No one knows the convergence point at this stage, but since hardware growth and software growth are correlated, we can only hope for the sustained growth of hardware—meaning semiconductor companies like NVIDIA and others.

While AI companies operate while bleeding red ink, supported by companies paying the price for those costs, the timing for capital recovery—or the risk of bankruptcy if recovery is impossible—remains opaque. Because the speculative aspect is too large, we should prepare for the stage where things eventually converge.

Thinking about what we can do now and what will happen next

For those who have something they want to create personally, being able to use current specs for just 3,000 yen a month with a ChatGPT Plus plan—even just for a hobby—undoubtedly feels cheap. I think an annual cost of ¥36,000 for a hobby offers very high cost-performance.
This is because you can implement scripts or apps you've made yourself as they are, replace them with apps designed for container operation, or quickly create things you wish existed. Considering your own hourly wage for the work or outsourcing costs, it is significantly cheaper.
In fact, looking at the applications listed below as if they were being "called out," it becomes clear that the systems and applications we use daily are being maintained or fixed with the help of LLMs.
small-hack/open-slopware: Free/Open Source Software choosing to use and/or support LLM usage/AI. - Codeberg.org

Ultimately,

• The cost of human labor
• The cost of having LLMs do it

I think things will change depending on where this convergence point lands.
Furthermore, as security risks increase, I believe the demand for security expertise at the engineer level will likely grow. In other words, Alphabet's massive acquisition of Mandiant might have been with this future in mind.
If it cannot be maintained, it fails as a service.
I recall the words of a former senior colleague: "Once it's gone, everyone forgets about it."

Don't stop doing what you love

There are many unsettling topics in recent news, but people who love working with computers should continue without giving up.
In a sense, I believe it is the duty of people who love the internet and computers to work in engineering, and services are meant to be maintained. Even if some companies scale back hiring, new services, including those from startups, should emerge more frequently now that development has become easier. Also, due to recent security incidents, there will likely be more than a few companies moving away from outsourcing and toward in-house development.

It's just an impression, but I have the arbitrary feeling that most people in engineering have experienced a battle with lonely hours, including during their student days. Since working with PCs inevitably involves a lot of solitary time, I'm sure I'm not the only one who wants companions.
Without becoming toxic, enjoy what you can enjoy pleasantly, and you may eventually find a place where you can share that joy.