The amount of fun I’ve had in the last 6 months building anything I dream up for my home lab stack is unparalleled.

I’m a professional dev. 35 years experience. C, C#, Java, Rust, JS, PowerBuilder, COBOL, pascal, Python, etc. I’ve done it all and am an expert at most languages.

Coding has been my life, and there is a little grief I feel from losing a quiet hobby I once had. I think I’m still in awe about how the magic pattern machine can bring my ideas to life so fast.

I feel my skills disappearing over time, but I don’t really care too much. Retirement is in sight for me, and magic pattern machine is always available (I run local models on a handful of dgx sparks).

15yrs dev here.

Im not yet set for locally running agents but my hands are itching to build a decent low-mid budget machine for this. I do a lot of agentic coding at work and I’ve honed my skills at creating custom MCPs for whatever I want.

Lots of small utils that can be custom-tailored to my needs can be done with agents in 2-4h to prototype, 1-2d to good solution, 2w of usage to eliminate all the rough edges and bugs. I want that stuff for my personal needs.

As for the main hobby project I do, it’s a godot game. Coding is the worst part of it. I do my coding at my job, now do the coding too after work hours?

You might ask me why did I even start. Gamedev feels different from usual dev in about 70% of activity. You draw assets, you create scenes, you tweak mechanics or some constants until it feels fun etc.

But hitting coding problems is inevitable. And when it’s some complex shit, it can go for weeks, months. It’s a grind. Its not fun.

I’d gladly eliminate the grind. The core architecture of it would still be under my guidance and supervision.

Please don’t meat-proxy.

What is meat-proxying?

A meat proxy is a human who forwards questions to an AI and forwards the answers back — adding nothing but latency.

• https://meatproxy.me/

I want my computer to do things. Sometimes I want to challenge myself with logic and word puzzles. Not normally both

There has to be two ways to look at it. Either you’re doing some recreational coding. Or something more complex, so code quality matters to you, or the project being some kind of art. Or: you’re just interested in some result. I’ve seen people vibe-code something like a Voice over IP stack and now my weird attempt at creating an Alexa can call people. I think it’s awesome! But… Don’t touch the code. I inevitably ran into issues and you really don’t want to poke at AI code, that’s super annoying to read. And it’s going to refactor everything 5 minutes later, anyway.

Also, a lot of the vibe-coded hobby stuff I’ve seen gets abandoned after a few weeks. Not all if it. But I guess it’s not that important to people, or connected with a deeper meaning. I guess it’s really two very different approaches at “doing” something.

It takes 30 seconds for it to write a script that would take me 2 hours.

Takes me another 5 minutes to review it.

I’ve watched it spit out a script in 30 seconds that would take me 1-2 hours to write after I created a block diagram to make sure my logic and sequencing was sound.

And then that would’ve taken another day to validate it.

Instead I watched/read the script as it wrote it and could tell what it was doing well enough, and what I’d have to tweak.

I am a professional software engineer and have codebases to support hobbies, but I’m not building them for the joy of programming and am instead building a codebase to make the actual hobby itself more enjoyable. I’m happy to offload the work to local LLMs in that case and have the LLM follow my existing patterns with me reviewing the code and driving functional and non-functional requirements instead of coding everything by hand.

I also have purely vibe-coded projects that are things I wouldn’t otherwise build due to lack of time, but since I can let the machine do the work while I attend to daily responsibilities or sleep, it’s nice to let a local LLM build it for me. In that case, I try not to look at the code too much so I don’t get pissed off and want to redo it. This works well enough for small projects that serve a purpose and that I don’t intend to build on much long-term.

Does the architect construct their home out of artisanal beams they carefully carved themselves, selected from trees they personally chose and felled?

They could, but there’s degrees to everything.

In programming, it’s fun and educational to reinvent the wheel sometimes but after you’ve been doing it for decades, sometimes you just need a beam and you don’t give AF what it looks like as long as it holds up the wall.

Example: If I ever write another terminal emulator, I’m going to let the machine fill out all those escape sequence functions once I get the framework in place. It’s so damned tedious! It uses zero brainpower and requires hours and hours of careful typing. For things like that, AI is a boon that frees up your time so you can focus on the fun parts.

for some people the result matters more. or they lack the skills to achieve what they want without using these.

as an example, someone i know (over the net) is an old NeXT/OPENSTEP fan for almost 30 years and recently having fun making related stuff with LLMs. like making hw accelerated graphics possible on NeXTstep

few other examples…

• this where the author made USB3 controllers’ drivers for Win9x/2k. the aim is to get ancient OS running on a modern machine, not necessarily programming a USB controller driver on their own
• this where the author made Magic Trackpad 2 drivers for OS X 10.9. This is the last OS X with classic looks, and MT2 support comes since 10.11 so it makes sense somebody wants this.

I’m a programmer, not a software engineer. I like being a code monkey and just hitting the keys and troubleshooting problems. I obsess over the styling and readability of my code, and the minutia of language standards and idioms. I’m not much of a fan of software architecture and high level planning; I keep things flexible and find that the best architectures bubble up from the constraints imposed by implementation details.

I’m kind of the antithesis of most people here “just doing work”, but even I’ve started using agents for my hobby projects.

I’ve used LLMs in chat and in-editor code completion since the early days of copilot and tab-nine and local llama; but I only started using them for agentic work since 3 months ago when I judged 5.6 Sol to be meaningfully capable (unlike the models available just half a year prior) so I shelled out the 20-something for a sub to evaluate the frontier for the first time rather than just running chat with simple tools locally.

(Edit: I tried many of the frontier labs now and I actually use local for agent tasks too now, but mostly just the simple stuff. I did a deep dive into customizing various harnesses and prompts and making my own memory and skill systems to replace harness-specific ones; this wasn’t a surface-level expedition)

There are projects I code for the sake of coding, but it feels a bit more of an active choice now. I found that I started a ton of projects; things that I only care for the results that I wouldn’t have been able to do before, and speed up things that were almost a year away on my timeline and knock them out in a single week.

However, it also changed what projects I’m working on entirely. There are things that I was doing before that only made sense to when I was writing by hand.

I actually sunk a good three weeks into building up infrastructure that I then immediately threw away because I realized writing C# in Lua was dumb when I could just write C# in C#; but I wouldn’t have before, because I was writing Lua in the context of a larger third-party program that was interfacing with C#; I didn’t want to rewrite everything that that program gave me just for the sake of being able to write native C# so instead I was trying to alias C# objects in Lua, which worked fine… but why? Don’t like writing C#, and I really don’t even like using Lua that much; the only things I really enjoyed about it was how flexible it was and how I could kind of abuse it to accomplish whatever outcome I wanted.

I was trying to avoid reinventing the wheel, but now I have a wheel-design printing factory. It became a non-issue, which changed how I allocate my time and my effort.

I said fuck the sunk cost fallacy and I just dropped all of it and completely started from scratch, leveraging LLMs to write all the busy work. Frontier LLMs have made (much) code disposable. It always was, but now more.

I quickly prototype designs that I wouldn’t have bothered before because the grunt work was just too laborious and I just didn’t have the time. I had it throw together a language server for proprietary DSL with syntax highlighting and embedded language support (multiple langs in one file); something I never would have bothered to do if I had to do it by hand (could never justify the time-spend), but that made my actual writing code experience better. Now instead of burning a weekend or maybe even a week on such a QoL project, I only have to commit maybe a couple hours of hands-on time and the rest is just the LLM.

I am not a programmer. I find coding fascinating. I do not find passion in writing code or in debugging. However, I enjoy game making. LLMs enable me to do so on a new level.

One of my hobby projects over the past couple of years is aikemi.eu, and while its ~70.000 lines of code might be 100% technical debt to me, it’s been - and still is - a fun challenge for me to work on.

because some of us can’t actually code but we can use claude to build some neat little apps. i also can’t draw so i use microsoft paint to make illustrations. idk to me it’s the same idea.

Well, I code for hobby and passion. And I also use llm both for documentation and also coding.

The point is that maybe I really have no interest in deep diving into yet another API or library shenanigans and having an ai wrote up the code is a massive step up. I will then revise and adapt as needed, and learn the API or library in the process cutting down time by a factor of 1/4.

Or maybe there is a boring task o don’t want to do or redo, ai is also great at that.

Lastly, maybe somebody ask for a feature or something to be added to my software and I don’t want or have to waste time on it, but it’s actually useful so an ai quick start let me implement that faster.

Just because it’s a hobby doesn’t mean you have to grind trough it… It’s even more fun to focus on what you love and delegate what you don’t love so much.

Ai is a tool, better learn how to use it properly…

There are people who like hand sewing stuff and there are people who would rather use a sewing machine to go much faster to create their final products. These people may then be able to spend more time on designing and testing different solutions for example.
What is hard to understand in that (since we’re not considering the ethical issues here)?

I am not running a local LLM, yet. But I plan to. Because I could use the skills for work, but also because I need a decent AI for research and general assistance (like debugging why my dual stack is not working, malware compromise forensics, security lockdown and performance optimization). I could imagine an agentic AI could provide a daily update on these niche things that a care about, with full privacy and ability for it to remain available. I’d very much like to avoid using the usual suspects.