AI is bringing my friend out of retirement
I have a friend that is lucky enough to have retired at 40. A year ago he was adamant he'd never work again, having been burnt out from his time at big tech. Back then he was also an absolute AI hater and wouldn't listen to anyone who claimed LLMs were useful for programming.
He finally tried LLMs when Claude Opus 4.6 released and immediately changed his mind in the face of the overwhelming evidence that LLMs can in fact program pretty well. And now with the release of Fable 5 he's giddily creating all sorts of things that would have taken far too long to make prior to AI-accelerated software development. He actually plans to try and found his own business now. He's a very smart guy, so I hope he can make something interesting that people want.
There are a lot of AI doomers and haters. In person I mostly see people doing the same thing they've always done, but now saving time on various tasks. But this is the first time I've seen someone go from grumpy and checked out to giddy and optimistic thanks to LLMs.
I think I'd be classified as a generic AI hater, because I really dislike generative AI and I find the reliance and push of the apps like chatgpt and gemini and whatever to be a really bad thing.
At the same time, I work in industry and there are very good implementations for LLMs possible, similar to the coding implementations described in the OP.
It just makes me weep for the world every time I hear someone say "I asked AI and it's like this"... Fucking use your own brain.
It’s frustrating being in healthcare. I work in an emergency department, and the number of patients coming in and immediately presenting a ChatGPT summary is increasing daily. They won’t even state why they are coming in- they just shove the phone into my face. “ChatGPT says I have this…”
When the work-up is finished and I’m going through their results, I get fact-checked in realtime by ChatGPT. Then I have to explain why they won’t be receiving extremely expensive and unnecessary testing (PET scans, NM studies, send out labs) for their, for example, viral gastroenteritis. Or worse- they pleasantly inform me that their LLM agrees with me so I must be right with my diagnosis. All thinking has been outsourced to these tools.
Meanwhile in Brazil (Gizmodo)
Tech can definitely save lives and have wonderful applications, but in the wrong hands, like triage that ignores science based professional input or hypochondriacs that ignore science based professional input, we speed run medical care back to the dark ages before science based professional input.
I got into a ridiculous back and forth with a coworker today on a ticket. At every stage he'd use an AI conversation as his evidence and each time he'd sprinkle in little requirements that are not actual requirements to steer it to the specific answer that makes him right. Very frustrating, I don't care what chatgpt says I just proved possible the thing you said was impossible in my last message!
For shits n giggles I tossed the whole unedited exchange to an llm and asked its opinion and it said I was right so that felt nice. Kinda. I didn't add that to the thread, maybe I should have? 😏
The goal is to have your AI arguing with his AI, then you both can go back to work.
While I do use Claude Code and Codex (successfully) at work, the biggest change for me is that I'm having more fun programming with modern LLM-assisted tooling than I've had in years. There are so many little projects I've wanted to do over the years, but couldn't dedicate the time to do the traditional way, that are finally in reach! It's a blast, and genuinely makes me happy. I feel like I'm getting to exercise my computer creativity again in a way that I haven't been able to since the old Commodore days, when I could fit the entire machine in my head.
I've had a similar experience.
Doing a lot of agentic coding (voluntarily, not because company mandates it even though they do provide the subscription) at work, and it genuinely does help with productivity at the cost of shifting a lot of the actual load to reviewing code instead of writing it. Going to be interesting times ahead figuring out how to do that sanely in the long run.
But the real benefit I'm seeing here is that I've finally been able to dust off one of my gamedev projects in my free time, as I just don't usually have the energy to do even more coding after doing it all day as part of my day job. I'm going a lot more full YOLO mode with the free-time coding, but also a single-player game does not have as strict security requirements as my real-job coding does, so maybe it's fine? Anyhow, it's pretty fun working at this abstraction level, and I'm not missing artisanal crafting of for loops and async event handlers one bit.
Holy fucking shit balls!
I started using Claude Code.
I blame you and @teaearlgraycold.
In less than a week, I created an entire site.
OK. I made the entire site in 30 minutes. I spent the next week completely fucking with it.
Two years ago when I tried this, I had to fit the entire thing in my head. I created each thing by hand.
Each bit of code. Each shell script. Each HTML template. Each table. The entire relational schema and all the data entered within.
Now I am asking AI to make stuff. And when it doesn't succeed, I simply ask again.
I've worked in product management for most of my professional life.
I punched machine code, byte by byte, into the Amiga.
I wrote entire websites in pure notepad.
I have the ideas, the technical chops.
I have so many fucking ideas.
And I can make all of them.
Are any of them good?
Probably not.
Am I about to have fun?
Probably not. But will that stop me?
Probably not. There is a book I recently read.
Letters to a young poet. The young poet writes to an older poet.
He asks for life advice. The advice is to never write poetry. Unless he must.
The rest of the book is complete crap. Because the younger poet never listened.
We don't make because we want to.
We make because we must.
Because we never listen.
Happy to see your friend thrive! I sorta relate to them as well. Last year I was pretty against using Claude Code and Codex and still did all my coding by hand. At the start of this year, my company gave us slightly formal training on using Claude Code in our day-to-day and it really changed my outlook on coding agents. There were always menial/annoying coding tasks that I'd have to do from time to time that I could now delegate to Claude Code instead. Since I was also a new joiner to the company, Claude Code really did help me onboard to the codebase. It not only used CLAUDE.md files that other engineers had left around, it went through the files in the sub-directory I was in and explained the purpose of each file and how they related to each other. It also helped me understand how the code I was working on would interface with our system at large and its various moving parts.
I think with the right guard-rails and usage patterns, coding agents can genuinely be helpful.
I guess I'll be the grumpy guy in the room.
I think even if you personally find LLMs to be useful, and are using them "responsibly" in a way that doesn't involve just blindly accepting extruded code, there are still good reasons not to use them. The unsustainable environmental costs, and not wanting to be beholden to mega-corps whose interests lay 100% in maximizing their own profits just to name a couple big ones for me. I get the thrill and excitement of being able to relegate the tedious aspects of your work to a magic black-box that can do it for you, but I can't help but feel that everyone has blinders on to the wider consequences of mass adoption of these tools in the ways that and extent to which they're being used right now.
It’s common to say that that the environmental costs are “unsustainable” as if it were proven fact, just something everyone knows, but I’ve only seen weak evidence for it. There are prominent examples such as xAI where they cut corners, but it doesn’t seem like it’s inherent. They run on electricity and there are good alternatives. And xAI is already installing solar. It’s only about 10% of their usage, but a carbon tax on data centers would be a quite doable and politically popular way to get them to install more. (Though, probably not politically doable nationally until after the Trump administration.)
Also, we are going to need much more electricity generation anyway to convert to electric cars. There is a surplus of solar panels available from China and instead of tariffs we should be buying all we can and installing them everywhere.
I guess my general feeling (and I am not any kind of scientist nor do I have any real information outside of the sporadic news articles and scientific paper abstracts that I occasionally come across) is that what we're doing to the environment is already unsustainable even if AI data centers were completely removed as a factor, so the idea that the addition of a bunch of resource-hungry data centers is somehow not going to make things worse isn't something I can really comprehend, even if they commit to some percentage of those resources coming from renewable sources.
And as for carbon taxes, again I don't have anything outside of being vaguely aware of the general discourse, but those have always smelled to me of a way for rich corporations to continue their environmentally-unfriendly practices while spending money to make it look they're not quite as bad as they are (I think the term is "greenwashing"). Money can only go so far to clean up the environment, which I imagine is far surpassed by the fact that the corporations aren't actually curbing their harm while being allowed to claim that they are because they're buying carbon credits. Just smacks like there's a lot of room for smoke and mirrors there.
I'll admit that I am inherently suspicious of corporations, and that suspicion grows linearly with how much money they're worth (and I would argue that time and time again they do everything they can to earn that distrust). AI companies are basically nestled firmly around the upper right corner of that graph right now.
Companies do cost comparisons when deciding what to build or buy. A carbon tax would make solar look much cheaper than natural gas.
(It already is often cheaper, but more so.)
I can’t imagine any AI data centre being willing to accept fluctuating power supply at any scale, isn’t that exactly why some of them are building their own on-site gas power plants? And while solar is cheaper than gas, I don’t know if industrial scale solar and battery is also cheaper than gas turbines or how hard you’d need to tax in order for the choice to push in that direction.
I've seen it, i'm just not sure how long it's going to last. Opus is still "cheap". Github copilot's costs went from too good to be true to 60x overnight. I'm curious to see how much more of that will happen. Given your friend is retired at 40, they've probably got the money to burn on something like that, but I'm not sure how long it's going to stay remotely affordable for the rest of us.
We really need less of this.
It’s not a business about AI. More so a business he can explore more leisurely because it exists. Of course at some point he’d have to start working hard on it because everyone has access to the same tools. Half-assing a new business will almost certainly kill it before it starts.
What's the business?
I'm also very curious what sort of business can be created when the tools are now widely available
One that helps others use AI to make things they sell, ie, shovel and jeans selling? Your friend does has a very persuasive profile that others with Claud Opus does not