Transcript
Title text: This is how you all fucking sound
[A smug tech bro wearing a sideways cap, watch, chain around his neck stands in front of a data center by a lake with dead fish. A smoke stack blows pollution into the air]
Tech bro: AI is already here, there’s no going back.
[A smug man in a suit with cigarette in hand stands in a restaurant while two disgruntled diners cough from the smoke]
Suit: Smoking indoors is already here, there’s no going back.
[A smug man in a top hat and suit stands in a factory with two sad and dirty children]
Hat: Child labor is already here, there’s no going back.
[A smug plantation owner stands in front of a field with with two angry slaves]
Plantation owner: The Atlantic Slave trade is already here, there’s no going back.
Reminds me of when the stupid "Web 3.0" made up by blockchain freaks was supposed to be the future. Not every technology will be as widespread as the internet. The internet facilitates communication across the entire world and offers many advantages over phone, mail, and other forms of communication.
The use cases and advantages are clear, even if there was an overly eager hype cycle in the 90s. AI might have some uses, but a clear advantage has not actually been established yet, nor have the legal challenges been ironed out. Remember that the current iteration of AI would not have been possible without breaking tons of IP law, slurping up as much data as possible.
Distributed web is a great idea. With a shitty vibecoded reference implementation (IPFS). And then also buzzworded as "Web 3" together with blockchain DNS and everything else.
What do you mean clear advantage has not actually been established?? Im getting weeks worth of stuff done in hours, like it's self evident how powerful AI is. If you use it to make deepfake porn instead of making you better at your job, that's a human choice. If you make a half decent effort, AI makes you an order of magnitude more productive. It's pretty freaking amazing, like how the automobile was a massive improvement over a horse and buggy.
I get your point. There's real benefit here. The question is if the benefit really outweighs the real costs, once they start charging for the actual resources used.
One reason we expeirence it differently is that some of us were already an order of magnitude more productive than average, without AI.
AI is a great tool for catching up. It slurps up popular patterns and spits them out, sometimes in novel contexts.
For everyone who used AI to catch a productivity technique(s) they had not yet encountered, I can see how it feels life changing.
I suspect we have a wave of realizations coming from folks whose token costs get too high, and realize they can get 95% of the AI productivity gain they experienced, with zero use of tokens - just by copying and pasting the patterns and tools AI already introduced them to. We don't talk about that aspect enough - genuine acceleration is happening, and some of those folks will stay more productive after the AI hype wave ends.
Of course, there's a whole other category of folks genuinely benefitting from AI because they need needlessly verbose language output to bullshit their dumb bosses. I don't currently have a dumb boss, so I'm not making use of that. But I 100% will start, if needed. Lol.
Well yes that's the case with me i was never trained in coding just started out as a low level employee in an accounting department teaching myself how to do vba scripts in excel to do basic stuff and now im building data pipelines and web UIs with claude to automate reconciliations and rollforwards that used to take me a week without using Claude. I also have friends who are senior devs real master programmers, and yes they can do everything I do with claude faster and better.
But, while one claude agent isn't faster than them, they have the ability to run a dozen agents at once getting 10x the work done they used to. Instead of paying ten good programmers $100K/yr each to be his code monkey slaves and not think too luch for themselves, they pay claude $2,500 a year to be ten codemonkeys and not think too much for itself.
It really is as game changing as the automobile was. That doesn't mean an idiot can't drive it off a cliff like they literally can with a car, and horses had some self preservation instincts that would save an idiot from riding over a cliff, but the answer is to let these idiots drive themselves off a cliff with AI, not to ban a new technology that helps a lot if you're not an idiot about it.
It's faster to go from the 50th floor of a building to the bottom by jumping out the window than it would be taking the elevator, but that just makes a mess of things. Similarly, you seem to be going faster with AI, and in some respects you are, but there's also the matter of technical debt, which is the messy aspect that someone has to deal with later.
I have used AI to quickly write up small functions here or there, but even then I've had to go in and clean up the code because even in small tasks it can be messy. The mess scales with the size of the problem, even if you do split it up among more agents (in fact it can be worse if you use too many agents).
Especially when people claim 10x productivity gains (a suspiciously oft-repeated claim, by the way), alarm bells ring in my mind, because I've seen the garbage that AI generates, and no one is actually reading that garbage carefully and cleaning it up while maintaining 10x productivity.
And yet how much of our industrial engineering uses free falling gravity whenever safely possible? What's wrong with a garbage chute straight from the sixth floor to a dumpster in the basement? What's wrong with a water tower using gravity to store energy and let the towns water literally fall straight down 50ft into the plumbing system?
That's what I'm saying, if an idiot breaks stuff letting it fall, that doesn't mean we have to stop everyone from letting anything fall ever.
Im just building rest api clients, what technical debt? Then i have a simple streamlit server with a clean UI.
Like this is all cookie cutter stuff but now a layperson can do it. It's basically ikea furniture for coding, and yeah im sure ikea uses a ton of water and energy and put a lot of furniture makers out of business, but that's the way of the world. And yeah a complete moron finds a way to fuck up building ikea furniture too, but there is now a way for moderately smart people to put together furniture and also put together simple computer apps without dedicating their lives to the study of woodworking or coding.
So idk, i mean once every twenty minutes maybe i correct something Claude is doing, sometimes they look for stuff on the wrong hard drive, sometimes they request network access when it's categorically not necessary nor what I asked it to do. Sometimes it wants to ping a server continuously until it no longer gets a 521 and i have to tell it "HELL NO STOP" but then my human coworker is like "yes this is why i love claude" and i just have to make sure that branch he's working on never sees production.
But the semicolons brackets pythonwhitespaces are always in the right spot, variables are always instantiated properly and then dumped from memory based on best practices, while/for loops incremented correctly. And yeah it all costs me some number of gallons of water. It's the water cycle it all ends up in the clouds and rains back down, idk sounds terrible for people who live in irrigated deserts but i could never live somewhere like that because i don't do well in hot weather wcyd
Very cool. Welcome to the trade! I can tell you caught the passion for it! Lol.
With or without Claude, you're a developer now.
If I can offer some advice: Don't let anyone tell you which tools to use. And also never let anyone tell you that all the magic is in the tool.
Oh, yes. For the places that were already getting by on slinging CRUD (create/read/update/delete) calls, I can see how this is a game changer.
Although, the pattern I see over and over is that the boss-man buys cheap code for a few years, then pays ultra-premium prices for a consulting company to dig them out of their costly mess of spaghetti code.
They usually repeat this every few years.
These AI tools are promising to solve that, but so did Web 2.0 frameworks, and so did memory safe languages, and so did COBOl and BASIC. All of them helped, and none of them really solved the good/cheap/fast trade-off.
Lol. Yes!
Agree AI is as overhyped as the internet in the late 90's was. I also think AI or some descendent of it will likely be as ubiquitous as the Internet is now. There's quite a few problems right now that AI just seems really well suited to solving, unlike the blockchain where it really only solved one sorta esoteric problem. I look at AI as being the bridge between the real world where things are fuzzy, rules are inconsistent, things don't have clear cut answers, etc and the digital world where everything is precise and well defined. That's not something that's going away.
However, what I see happening with AI is much the same thing as what happened with the Internet. To use the Internet in the late 90's was frustrating. The computers sucked, they were huge, they used a bunch of power, the connection was slow, connections dropped, they weren't always on, they took quite some to establish, etc. It wasn't till CPUs got good enough to be able to be battery powered and still render full websites (in other words, the key building block of a smartphone) that the Internet really became a ubiquitous thing for most people. Today's AI uses way too much power, requires hardware that's way too expensive, is less smart than people think it is, has problems learning, has problems with hallucinations, etc. What I see happening is the AI bubble crashes, like the dotcom crash, but then it comes back once the technology is really ready.
As far as law and IP go, the Internet often had lots of issues with that too. Lookup the origins of why we have Section 230. It's still something we're arguing over. We'll figure out the legal issues. And IP law is broken, has been for a long time. It needs a revamp to bring it back to some sanity. I have no problem with AI breaking IP law. Much of that shouldn't be under copyright anyway.
That's a great point.
Edit: (Sarcasm incoming:) But we can all agree with the tech bros, that I became obsolete during each of these transitions, because I didn't drop everything else and focus on it completely to the exclusion of all else.
I guess my paycheck disagrees, but who are we going to trust - cold hard cash, or some con artist tech bro CEOs? Lol.
I actually think that AI might be closer to Web 2.0 than blockchain. Where I see it's biggest potential though isn't in the stuff most people do everyday, but in specialized applications. I see some of the uses in stuff like medical research and find the potential there to be wild. In alot of ways, it's sort of a revolution in how we think about doing computing. We're still really early on it, so its hard to know just where it takes us. Even on the more mundane stuff, it really does help programmers be alot more productive (though it's hardly a replacement for talented programmers). Which is a huge for helping us build the next generation of tech.
There's alot of stuff they say is obsolete but really isn't. For example, in my day job, I'm an analog IC design engineer. The most advanced process I work on is sort of 90nm (but not really). The previous silicon process I worked on was basically a 0.5um process. Ask most people and that's all stupidly obsolete - should have died out in the 90s or something. But I work on power products. Power is analog, not digital. We may have some digital stuff, to be sure, but what we fundamentally do is analog. And you can't use 5nm processes to deal with "high" voltages - that's all on "old", "obsolete" silicon processes. Oh and you know what we power with all this? Among many other things, those fancy AI chips. So yeah, alot of these transitions do obsolete stuff, but there's often still important niches in older technologies. I mean you still have people learning COBOL so they can program mainframes for banks.
I 100% agree. I use AI on purpose for things, once in awhile. I've never had a use for blockchain, only had it shoved at me.
Exactly. Ilike Cobol as an obvious point against both extremes of the AI argument.
"AI is going completely away..." - Sure. Right after Cobol and the Fax machine.
But also "AI has eliminated all need for X, Y and Z." Sure. Right. And it also just got rid of Cobol and fax machines, right?
Things change surprisingly fast, at the same time as things change shockingly slowly.
What doesn't go out of style is learning and expertise.