Will AI Replace Most Tech Jobs

[7Wannabe5]

I notice quite a bit of animated body activation here.

What?? Waaay too boring to watch the whole thing, but I skimmed fairly thoroughly, and all I saw was jerky right-hand motion, the occasional neck twirl, and some overall tensing. I've seen a couple of my over age 70 male partners exhibit 10X more muscular animation during a sleepy afternoon nap session with me. I mean, one of my early 60s partners bounced and yelled a bit playing PubG all the time, but I still had to send him out to chop and haul wood on the weekend or he would have been annoying me too much while I attempted to read.

That said, I do agree that engagement likely varies at the level of the individual based on prior experience, wiring, druthers, etc. For example, I've recently been indulging in live musical theater, and I could watch a show like "Moulin Rouge' or "Hadestown" all day long in a state of enthrallment, whereas I'm not nearly as engaged watching movie musicals on small screen. And this is likely in part due to the fact that my parents frequently took me and my sisters to musical theater, dance performances, and big-screen musicals when we were young, and we very frequently played at putting on shows ourselves. Also, the bright colors and lights of live theater tend towards pinging me into a low-key manic enchantment, which is not dissimilar to that which I experience gazing upon all the contrasting colors and forms and other sensory inputs to be found in a garden or a woodland setting. So, I would suggest that you are probably much more wired for the testosterone mediated happy-feeling caused by simply viewing/interacting with moving objects hitting goals like a spear thrown while hunting, and I am much more wired for feeling joy when a brightly colored sweet fruit is spotted standing out from the lush green background of the new-to-me-path in the jungle which I am exploring in the company of a singing, dancing bear and toucan.

I don't like alcoholic beverages (dulling and headache inducing) or sports (dull and utterly pointless) or sitting in bleachers (uncomfortable and often too hot or cold and/or stinky with old, rank sweat), so the only way anybody ever gets me to willingly attend a sporting event is by bribing me with something else that is more interesting or stimulating. For example, funds for gambling on the sporting event or dinner at a truly interesting new restaurant afterwards. Although, I suppose if I somehow wind up with a grandchild who is a jock, I would try to be supportive and not even read a book or spend most of my attention on people-watching or savoring my snack bar purchase during the game.

[Revan]

AI will replace some tech jobs and others new ones will be made with a focus on improving AI. New jobs are always being created and others destroyed.

I'd say AI will hugely affect how future generations write. It'll affect how we can communicate with each other, as younger ones will depend on AI for paragraphs, essays, etc.

[alex123711]

Seems like AI has got a lot more attention recently and apparently improved a lot, anyone think it will take their job soon?

[jacob]

Seems like AI has got a lot more attention recently and apparently improved a lot, anyone think it will take their job soon?

I think it has to a large extent taken my job as an author. Not because AI has started writing good [non-fiction] books, but because fewer and fewer people are interested in reading entire books when asking chatGPT provides a quick&easy answer. The irony is that the rate of publications keeps increasing. There's clearly a race to the bottom for getting the attention of falling reader base.

[Scott 2]

As someone who came from tech and is not working, I think AI has dramatically reduced half-life of my existing skill set. While I'm using AI for personal needs, I'm not learning how to do AI augmented technical work. I can't see how I'd compete with someone who's been practicing the past 2 years.

I'd agree on chatGPT substituting for reading. Just yesterday, I used it to justify removal of both Kegan and Plotkin from my reading lists. I only made it through the first 2 hours of Wild Mind before giving up. Plotkin's verbose. He was no match for the AI alternative. At least with Kegan, I got through Immunity to Change. ChatGPT confirmed the essence of his earlier works though, and I decided that was good enough.

[Slevin]

Yes, in my workplace we are shifting to AI augmentation in execution. I.e. it is 10x better to point me to documentation for a subject and give me working examples of how to implement an algorithm / tool / solution to a weird problem.

AI still isn’t great / amazing and lies to me all the time, but so does stackoverflow and google, so no big surprises there. Not having to dig through the metalayer of garbage saves me a lot of overhead even if I’m just asking for a link to a certain documentation.

Of course, I’m an SRE / platform engineer. My job is gluing tons of different technologies together, so this use case is very useful to me. For people doing high level optimization and performance testing / optimizing kernel bits, it is radically less useful.

Overall, I see my job shifting to more of a managerial / architectural one in the mid term, as there is still drastic patchwork and shenanigans from the AIs. I’m not sure how you would pull the humans out of oversight of all these systems even if the AI is writing the code.

[zbigi]

I can't see how I'd compete with someone who's been practicing the past 2 years.

I'm an opposite example. I was out of job market for 2.5 years and just recently got a short-term coding contract. I'm doing my job and meet expectations without ever touching AI. FWIW, I haven't yet seen anyone else at the company use AI either.

[Scott 2]

My direct technical experience with it is limited. Getting up to speed on Linux, primarily. There was an issue with responses equally weighting deprecated tech from years ago. But the scripting snippets often needed minimal modification. And it was great for orientating myself in the overall context.

My impression from LinkedIn has been agent augmented IDEs are changing how people code. Automatic setup of test harnesses, generation of classes etc. There's claims of higher level architecture as well, though I'm more skeptical there. The vibe coding mindset looks dangerous long term.

The company I'd contracted with last year, was working to license every seat with an agent. Last I heard, I think it was copilot. Realistically, people were probably already using stuff against corporate policy. From the chatter I've seen on linked in, Claude might be preferable. The playing field seems like it's constantly shifting.

That's part of the reason I've not tried to keep tactically current. The optimal meta is evolving rapidly. It's likely much of today's necessary skills (ie multi shot prompt engineering) is transient. Without a day to day payoff, I haven't wanted to make the time investment.

Were I still in the field, augmenting myself would be my top technical priority. It's clear there's going to be a divide, with AI adopters taking all. Maybe late comers can't even find work in a few years.

[brainstorm]

@zbigi - Is it a US firm? Sorry if you've mentioned it before, I scanned older pages but didn't see. I'm curious if the primary location influences that particular aspect of the culture.

@Scott - Definitely agree about people already using stuff against corporate policy, and keeping up with the SOTA being a constant effort. But I don't agree that you can't compete with someone who's been practicing for two years!

Personally, I'm a hater. I don't think using AI tools has a significant impact on one's ability to deliver good software - I think they're more of a band-aid that helps with documentation or initial drafts of code, and produce marginal gains. In my experience, top performers contribute enormously whether or not they use an AI tool. My favorite coworkers are a mixture of users and non-users, and I wouldn't know which ones are which based on their work alone.

I still see teams and organizations struggle with people and business problems: picking the right problems to work on (can be solved by software AND is high-leverage - including solving problems that affect multiple teams and reducing duplicated work), planning long-term and not bouncing around constantly, and prioritizing. And I still see underperformers doing the same things that lower quality and slow down the team. They lack the debugging and incident response skills, the understanding of the systems they own, the knowledge of how to check their work (and to actually do it), and how to clearly communicate all the above. These are not really within the realm of LLMs. When there's so much more to being good at your job, I don't see how the revolutionary (and job-killing) claims can hold up. Readers and writers, well, another story.

I think many more jobs are at risk due to offshoring, like @AnalyticalEngine said a couple pages ago - I see it at mine and my friends' companies all the time. Offshore developers are much, much cheaper than US ones. If LLMs tank the job market, I think it's moreso due to the pure craziness and sensitivity of investments and stock prices to progress on these [highly specific] benchmarks. If there are mass layoffs, it's because organizations are bloated and/or they can survive and make tons of money in the short term by gutting staff, Twitter-style.

Anyways, that's my rant. I'm not saying it's not helpful to save a couple hours by having copilot draft some code or tests for you - I'm glad you find it helpful. I just don't think it's usually the bottleneck, and it can introduce some really sneaky bugs because it looks convincing but is not correct, which is kind of their whole MO to begin with. I'll spend my time upskilling in other areas - keeping up with the meta seems like working against ERE principles.

[zbigi]

@zbigi - Is it a US firm? Sorry if you've mentioned it before, I scanned older pages but didn't see. I'm curious if the primary location influences that particular aspect of the culture.

It's a German financial institution so, in all fairness, likely one of the more conservative companies in the world. I don't think they're using cloud yet, and least not in my department (data).

Anyways, that's my rant. I'm not saying it's not helpful to save a couple hours by having copilot draft some code or tests for you - I'm glad you find it helpful. I just don't think it's usually the bottleneck, and it can introduce some really sneaky bugs because it looks convincing but is not correct, which is kind of their whole MO to begin with.

That was always my take when looking at what AI can do for coding right now. When I look for some advanced or obscure information (everything else related to my field is already in my head, or I know exactly where to find correct information), most stackoverflow answers are wrong/inapplicable. Often, all of them are. I prefer to use my own judgment when sifting through the crap on stackoverflow than relying on AI to do it.

[Scott 2]

Good to hear from those with hands still on the keyboard. @zbigi your role sounds especially well insulated. High expertise, conservative organization.

AI could facilitate globalization of the knowledge workforce. Several ESL devs I worked with were psyched about the value for written communication. They'd feed their message through and fix the grammar.

Bridging that communication gap was always a tough part of off shoring. Someone could theoretically even use an AI voice, though there could be some lag. And it doesn't solve the cultural or time gaps. But real time accent reduction would be a viable benefit.

[just]

In my experience, LLM are good for getting a new project off the ground, since it's free to code as it sees fit - and it's good at generating the boilerplate that's required to get started. However, in a large established codebase it's less helpful, since it cannot hold all relevant source in the context window (or you have to manually point it to the relevant source files). What "large" is relative to an LLM, will of course change as they change, and something like Cursor could be more viable.

I do some algorithmic work and for that I find copilot lacking. Sure, if you just have to implement a vanilla algorithm (e.g. transpose a matrix, search a graph/tree) it's fine, but when I do more than that I often switch the auto-complete off, since it suggests nonsense that looks somewhat right.

[Western Red Cedar]

I just wanted to chime in and thank those for sharing their experiences. I don't work in tech, but have started looking at how AI may disrupt other white collar industries. Appreciate hearing from those on the front lines.