The dotcom bubble was the whole investor industry trying to capitalize on the internet, and they just failed because they were investing everything in a new thing people didn’t need yet.
Today we live in an interconnected world, close to what was promised to the investors back then.
In ten, twenty years, once the bubble, OpenAI, Anthropic and all that are long buried, in what form do you think AI will exist? Do you think it will be an improved version of our current-day LLMs, or a different form of AI? Or do you imagine something else altogether?
If you don’t work in tech or software development, it’ll be very hard for you to understand what this tech is doing within that space.
As a game developer who has worked in professional environments for about 15 years, I can say this is the most dramatic shift in working patterns in my own lifetime. A lot of it has to do with the mechanics, some of which are physical limitations. Humans can type around 300 wpm, and maybe you can stretch that with an integrated environment that has some intelligence around auto complete and so on. However, AI can write at 6000 wpm. This is a modern John Henry problem: even the best programmer could not out write the AI, and I think the best programmer partnered with an AI will always beat the best programmer without it. In all honesty, humans make tons of mistakes, and programming has many patterns that reappear. I’m not sure if the human+keyboard is the best interface to write code. Likewise, an agent without clear context is somewhat worthless & will generate poorly architected projects.
I don’t think AI’s assault on software development translates to all industries. Agriculture may use AI to inform how they do things, and eventually, machines could then take that knowledge and run the farm for you. The reality is that’s coming, but maybe 20 years or something. It’s not today, not tomorrow, and not for a while.
Other industries have greater risk profiles compared to game development. If a game is programmed poorly, so what? If a medical app is programmed poorly, someone could die. Legal & financial apps programmed by AI have meaningful risk profiles as well. Sure, AI will impact them, but it must be slower due to the considerations at hand.
In December of 2025, I would have said an AI bubble is going to pop. Now, I’m really not so sure. It’s upending how I’m used to working, and I struggle predicting the future. A few years ago, I would say I had a good sense of how to position myself & my career for a long term plan. Now, colleagues of mine with more tenure than me say, “I can’t predict what’s next.” This is pervasive once AI tools enter your workflow. It really is magic at times, talking to a machine & watching a game come into existence.
This is buttressed by people showing me memes of ChatGPT screwing up the count of R’s in strawberry & other silly memes. The reality is the consumer models are different than the business models, and integrated agents (ones with access to a scripting environment) can be very powerful compared to a chatbot. It’s like comparing an e-bike to a car: they each get you somewhere faster than walking, but the scale of capabilities changes quite a bit.
AI will need to find new patterns outside of LLMs in order to actually create something like AGI. LLMs are a mechanical process, a regression towards the mean at scale, but its still very powerful. You could write code on punch cards, or in a language with a compiler. AI feels like it is a big step, in that way.
For my own area, my hopeful self wants to believe that this will enable new voices & projects to exist that would have required significant funding or hard, multi-disciplinary skills prior, or a combination of them. You can hack an unbelievable amount of the game dev process for what’s basically $100/mn, compared to what would have been a burn rate of +$100k/mn for a small dev team. That is transformational.
At the same time, my pessimistic self sees things like arcade.gov, which is a fucking travesty & embarassment to the medium, clearly vibe coded by some nepo-baby with connections. The Skyrim IKEA shelf meme mod, while funny, is a marketing stunt. More of these sort of projects are coming, where the game is a vehicle for something else (a product, a brand, a person). We had Sneak King & Cool Spot, but I mean the scale. I think those sort of projects will be normalized, or at the least, many more will exist than today. If you look up AI disclosures on steam, it’s turning into a hockey stick.
It feels like Pandora’s box has been opened. I don’t think it’s all smoke & mirrors. It will not impact all industries equally. It won’t fix humans being poorly intended or malicious actors seeking power & wealth. In many ways, they benefit the most (giving capital more access to creativity). It’s built on the largest scale theft of public good for profit. To me, that’s wrong, and the people/masses should be getting financial returns, collectively & universally, in exchange for what was stolen.
At my job, my main role is being a “solver”. I deal with stuff that others are struggling with, or things that are both complex and urgent. I also handle most integrations with third party systems that lack proper documentation.
Just a year ago, I could confidently say: “if an AI can do one of my tasks, that task should not have been assigned to me in the first place”.
Nowadays it’s the opposite and I feel like AI is specially useful for dealing with things that would usually have been assigned to me. I went from being needed in every integration discussion to only hearing about a new integration after it already got merged.
I mean I still prefer writing code by hand and I still write higher quality code faster by hand than others on my team who use LLMs to write code for them. Writing good, production intent, software is not about wpm, nor is more code necessarily better. Concision (as long as it doesn’t sacrifice readability) is a virtue.
I watch a streamer who game devs without AI because I appreciate watching him discuss his thought process & learning more about how I could write it manually. For context, I’m a game designer who knows some code, not an engineer. There’s real concern around skill atrophy without people doing some of these more manual ways of doing things.
My point about WPM is moreso about how the scale impacts velocity. I do think the speed impacts the old adage about software & pregnancy.
There was a quality of software development where work takes the time needed, and sometimes throwing more bodies at something wouldn’t alter the time. A good engineer could maybe shave some time off, or the patterns they used were more resilient or production ready on the first pass. Now, that engineer could genuinely deliver that feature much faster.
Here’s another trap I forgot to mention: our human read speed is always faster than our write. We may write up to 330 wpm, but your read speed is 400-1000 wpm. AI read speed is like 11m / wpm. The gap between our write v. read suggests it’s faster to describe a plan, read what the AI wrote & triage it, rather than the alternative.
I mention John Henry because in the story Henry wins but dies in the process. Going fast is not the only goal, but the scale of the velocity change happening is really difficult to overcome mechanically on paper.
Finally, to be clear, I say this stuff as a bitter pill I don’t want to swallow either. :/ I have my own emotional qualms with the whole situation. There’s big business risks around becoming captured markets & beholden to token costs. I think part of the AI acceleration is about creating that dynamic, and it greatly concerns me.
Given how straight tractors drive, the way you can programme the topend to self plant/harvest a field. The driving itself isn’t AI, but there’s a ton of image recognition software popping up that can limit the amount of pesticides used. As they’re able to recognise the exact areas in need of them.
Then we have drones able to capture soil efficiency and there’s probably a ton of other cases. AI is definitely altering farming and in theory could also reduce the needed amount of manpower - As more and more specialised machinery, can be used more or less autonomously. There’s legitimately farms in the Netherlands marked as airports, so they’re allowed to fly drones of a certain size. And all the imagery processing is also depending on a kind of AI
I agree with pretty much all your points. I use AI every day to do the typing for me, but not the thinking. It’s saved me from mistakes and makes me far more productive. However there is no way it could replace what I do.
It’s probably going to take a few years, but eventually I think devs will finally realize that the typing is the thinking.
There are projects I worked on 5+ years ago that I can still describe in intimate detail. I remember the design decisions, I remember the concessions I had to make for technical or business reasons, I remember they connect to other systems, etc. Then by contrast I have code I wrote with Claude ~2 months ago and I can barely remember a single fucking thing about it… I did all the same planning and made all the same kinds of decisions, the only difference is that I didn’t write the actual code.
And that makes sense. When you physically engage with the code by manually typing, you’re creating connections in your brain that you don’t get just from reviewing it. It’s exactly the same reinforcement mechanism you get from taking physical notes during a lecture versus just passively listening. Physical engagement builds memory and sharpens understanding.
Two or three years from now, when all of the projects minted in the '23-'26 era have aged into ‘legacy’ status and the AI-generated technical debt has been left to proliferate unchecked, things are going to start breaking in ways that the LLMs won’t be able to reason their way out of. Maybe they’ll be able to ‘fix’ bugs as they pop up, but in slopified code fixing something in component A is inevitably going to break something in component B, and fixing something in component B will break component C, etc. Eventually you’ll end up with a disgusting, agentic ouroboros of regressions, and the only devs left in the industry will be wishing they hadn’t mortgaged all of their intelligence on short-term productivity gains — most of which only benefited the shitty tech companies that were so desperate to replace them.
I think you have a very valid point. That’s why I have increased all of my documentation of everything I am doing considerably. I always document the “why” of the code instead of the what, and do as much of it as I can inline instead of a separate repository.
That is true for me and most of the best devs I know, but I don’t think that is true for most devs in general. I think a lot of people already “type and forget” their code and so what they are losing by letting Claude type for them is something they never had in the first place.
And the world always adapts to the majority’s habits. If you do a research into what you lose by not typing your code, it won’t show this phenomenon you describe if it only happens with a subset of developers.
The typing is the means of communicating the thinking. If you are “doing the thinking” and the AI is doing the “typing” then either you are instructing the AI to generate additional content without original thought (which is a waste of your time, the AI’s time, and the reader’s time), or the AI is actually doing some of the thinking (in which case your job can be automated).
The programers use AI to make art. The artists use AI to write. The writers use AI to program. I think we’re all in denial by both just how much our own value is at risk, and also the real value of AI (less than advertised). A bit paradoxical.