akitaonrails

Clean Code was originally written for human programmers. But how does it change when we now write for AI agents?

I spent the whole Sunday on this, and it was one of the most productive Sundays I’ve had in a while. The mission was specific: close the loop on the simcade racing games that are hardest to emulate, the ones I’ve wanted to run properly for years, and only now got to a reliable state.

Two monsters sit at the top of that list. First, Driveclub on shadPS4. It’s a PS4 exclusive, never got a port, Evolution Studios was shut down, no remaster. To play it outside the original PS4, emulation is the only option, and PS4 emulation is still the most immature part of the ecosystem. This is by far the hardest on the list.

TL;DR: Yes, the title is clickbait. The answer is no, it’s not worth it. Keep using Claude Code with Opus 4.6 or 4.7. Details below.

A few weeks ago I wrote a detailed LLM coding benchmark comparing 33 open source and commercial models on the same test: build a Rails app using RubyLLM. The conclusion was that only 4 models generated code that works on first try (both Claudes, GLM 5, and GLM 5.1), and that for anyone who doesn’t want to waste time fixing API hallucinations, Claude Opus via Claude Code remains the most rational choice despite the price.

This article is a follow-up. I’ll keep that benchmark updated as new models come out (Opus 4.7, Qwen 3.6, GPT 5.4 via Codex are already in). This one tackles a different question that shows up in my feed every week: what if I combine two models in the same project? Opus to plan, GLM to execute. Does that work?

I bought a Lenovo Thinkpad T14 Gen 6 and installed Omarchy on it. It’s not my main machine, and it’s not supposed to be. It’s a companion: a notebook I can open on the 3D printing office desk, SSH into the desktop, fire up Claude Code, access files on the NAS, debug the network through ethernet without hunting for a USB dongle and a long cable. This article covers the hardware choice, the Omarchy setup on top, the customizations specific to a laptop and to this Thinkpad in particular, and the architecture decisions that might not be obvious.

I updated my report and ranking on the LLMs benchmarks to add the newest Claude Opus 4.7 and Qwen 3.6. Check it out (all the benchmark code is open sourced in my GitHub)

ByteDance opened up the public release of Seedance 2.0 today, April 15. They had promised the model late last year, but access stayed locked to commercial partners for months, with delay after delay. As of today, anyone can try it.

Every time someone asks whether juniors are going to stop learning how to program because LLMs can write code for them, I have the same reaction: you’re asking the wrong question.

You’re confusing code input with software engineering.

Those are not the same thing. They never were.

This post asks a different question. After almost two months running that setup in production, shipping an episode every Monday, last night I finally pulled the plug on Qwen3 and switched the whole thing to ElevenLabs v3. Let me tell you why.

For the last 37 days I locked myself into a vibe coding immersion. The goal was simple: actually understand what the current generation of LLMs and coding agents can do. Not on a weekend toy project, but building real things, with production deploys, users, and the maintenance pain that comes along.

The result was 653 commits, ~144K lines of code across 8 projects published on GitHub, and a series of articles documenting each one.

This post is a direct follow-up to my recent articles about the new home server with openSUSE MicroOS and the Minisforum MS-S1 Max. Those covered the foundation. Here I want to show one concrete use for it: putting together a decent Bitcoin stack at home, focused on privacy, operational sovereignty and safe transactions on my side.

For the last 37 days I locked myself into a vibe coding immersion. The goal was simple: actually understand what the current generation of LLMs and coding agents can do. Not on a weekend toy project, but building real things, with production deploys, users, and the maintenance pain that comes along.

Classic RAG (using vector database) is over. With the most modern frontier models offering 1 million or more tokens for the context window, we should change our approach to knowledge organization.

Coincidentally I wrote about this same subject this week. But I went a step further than that. It's not about the pen.

http://www.akitaonrails.com/2017/06/01/conference-talks-inte...

It's not that we are all "pissed", it's not the tone of the article. I wrote it because that's what I see happening. And I will be the first to be proven wrong. But right now, it's not that Apple products are "bad", it's more like everybody else caught up while they're in standstill. That's the gist of the article.

That's one of the points. Back then, using Xcode and develop for iOS were bonuses, yet another reason to buy a Mac. These days, it feels more like I am obligated to buy a Mac if I want to develop for iOS. Not the same pleasure it once was.

thanks, corrected :-)

Did you know that at one point X had a freaking printer server? These are all complaints from the maintainers of X themselves. I am not making that up.

I knew the title itself was going to be both good and bad for my intentions, but as I responded in another thread down below, the point of the article is not to be a statistical report.

It's a reflection on history, decades-long, using Apple as the protagonist.

I am a very long time Apple fan, having used everything they released. And I am not 'frustrated' by recent releases. I am a technologist, I like technology, and I will use whatever is the best for now and the future. Which is why I chose Apple years ago. And which is why I am no longer enthusiastic by its future roadmap.

You're right about that. I didn't want to extend too much on that part and I can see how it can be misinterpreted. I didn't mean that Apple's direct reason to build Quartz was because of X. But it is true that X didn't have any reasons to be used back then in the mid-90's as well, specially coming from NEXT with it's Display Postscript compositing manager.

You're not wrong. The idea was not for the article to be a statistically correct report of the present day status. That's easy to get and Apple's marketshare is nowhere near bad.

This is a decades-long reflection on computer history as a whole, having Apple as protagonist and important player.

The title reflects where the early-adopters - the same people that brought Apple back from the dead in the late 90's - will go from here.

I really meant X11. The desktop manager fragmentation is an issue but not so much. It doesn't help for the end user if the system is difficult to maintain, if out of the box it's easy to break, it stutters, crashes, doesn't draw correctly, and overall is difficult to make fluid. Latency is high, one app can hang the windowing drawing, etc. So yes, it's a lot to blame on X11.

I was not expecting it to reach Hackernews frontpage :-) Sorry if it was a confusing read, I just wrote my personal thoughts. It was not the intention to make an objective checklist. Feedback is appreciated.

I did a small change in the opening statement to make my intention clearer.

Exactly! I am using GNOME 3 with Wayland in Arch and enjoying it very much.

This was also one important reason why I am switching. The price, the enjoyment of hacking are primary reasons, but the politics pathway bothers me as well.

I am using LFS on GitLab right now, works perfectly and I mention it in the article.