Yaqup Phylac

Why is everybody so obsessed with dark design (black, violet, etc.)? Can you imagine reading books printed white on black paper? My screens are light, toned to resemble warm old paper. I spend 10+ hours in front of a monitor, and I had to switch to a light background 15 years ago when the dark rectangle became imprinted in my vision. One might think a dark screen shouldn’t be harmful, but I believe the issue was eye strain from focusing on tiny white text.

We have got a puppy. It is lovely and playful. It is full of energy, doing a lot of agility and dragging toys and shoes all around, leaving our house in a mess. But sometimes, out of the blue, it pees or poops wherever it happens to be. And if you are not vigilant enough, it even eats its own poop. I have started working with a new code repository. It feels like it was written by programmers with a puppy mentality: everything bundled together in a total mess, poop scattered around, and some of it looks like it has already been partially eaten.

I received a support request from a different department claiming that the Playwright SaaS I built for them and used by their agent was suffering from memory issues. I SSHed into the host system, ran free -h, saw that 90% of the memory was free, and closed the incident. After a while, the poor reporter came back again, saying that the memory problem is still not solved, but this time they mentioned that they had not yet implemented the cleanup of large videos. I got suspicious and it turned out that by "memory" they meant persistent storage, which everyone else calls disk space. Do you think AI agents will be of much help for those colleagues?

Everybody is building MCPs for ABAP these days because anyone can just ask, “Build me an MCP server for SAP ABAP ADT and make no mistakes.” The agents then generate a tool similar to the few tools that existed before agents.

I use Opencode and Claude Code (Opus 4.5 and 4.6) daily with Go, Python, and Bash. Frankly, I can’t imagine using a different language, since the generated code is often bloated and inefficient, and I wouldn’t be able to properly review code written in a language I haven’t mastered.

Uz nekdo pripojil @openclaw k @rohlikcz aby mi jednou tyden objednal to co jsem si objednaval v minulosti?

I spent 15 minutes precisely describing a new feature for Claude. Then I asked Claude to read it, read sample data and use another feature as a template for implementation. Claude spent 3 minutes "whirpooling", then informed that the requested feature already exists. Well, that's the dark side of using AI and not using your brain ...

@greg_not_so Even if wrote the instructions how to get the server's PID from logs, Claude was still strying to use $! which works in bash but somehow not for Claude.

@JakubFilak made for clients but not for admins?

Oh daer, can somebody patch Claude Code to be able to run more processes? Currently it can nicely run a bash command and captures its output but I need to run two processes: server on background capturing its outputs and client on foreground capturing outputs too. I thaugth Claude how to run the background process but it cannot kill it - the background job makes Claude utterly confused :D @bcherny

I taught Claude Code how to run bash commands in the background and capture their output. Running Python that way was particularly spicy, since you have to use "python -u ..." to avoid output buffering. Feels good knowing you can still beat the machines 🧐

@agrislis @noquakquak @EoinHiggins_ @DisabledDoctor You studie history only from propaganda materials? The victory was a common joint effort, we needed all - USSR and USA and UK and Canada and all others.

@noquakquak @EoinHiggins_ @DisabledDoctor The USSR freed you from fascism. Study history and you'll be less surprised.

Doesn’t get much clearer than this

Heh, like you seriously think people do not clash with police on protests in Europe? :D Not all nations (e.g. post-communist countries had enough during that era) but the fights with police are sometimes brutal yet you don't see police using deadly guns - especialy if they are 8 to 1.

@noquakquak @EoinHiggins_ @DisabledDoctor Do people interfere with Feds in Europe ?

Pull requests containing primarily LLM-generated code have started appearing in my repositories. It’s easy for contributors to get overexcited and fall into the trap of “this is simple to add, let’s add this shiny thingy and that shiny thingy,” and eventually the pull request grows into an insane set of changes with duplicate, unnecessary README content — to the point where it almost makes you want to vomit.

@mirekpet Kodis v IDE nebo notepadu nebo hardcore style s pomoci efektu motylich kridel a primeho prepisu bitu na disku?

@JakubFilak Vařit guláš ne z masa. Ale z něčeho, co už maso vidělo.

Everybody is talking about writing prompts to generate code. Ha ha—that’s an outdated approach. I use GitHub Copilot in Vim to generate my Claude Code prompts, which then generate the code.

On my workstation, I have GitHub Copilot enabled in Vim as a plugin. I have just installed Claude Code on the workstation too. I started Claude, hit Ctrl-G to edit in Vim and I was utterly surprised when Copilot started finishing the prompt for Claud Code. How should I call this setup?

Here you are -> the ArgumentParser to FastMCP tool convertor (currently tuned for sapcli): #L69" target="_blank" rel="nofollow noopener">github.com/jfilak/mcp-sap…

I find it funny how Kubernetes was supposed to hide the complexity of creating infrastructure - it did hide something and introduced the complexity of Kubernetes resources. Coding agents are supposed to hide the complexity of coding - haha: certainly: you hide the complexity of coding and introduce the complexity of coding agents.

Agentic coding looks like a superpower. But I am also disgusted. I have been delegating coding jobs to colleagues and external workforce for many years, and I never enjoyed it. A) I do not enjoy translating my ideas into other brains, because natural language is non-deterministic and the target brains often lack the necessary resources. B) I enjoy building things myself. C) I do not want to read stupid code and ask for re-dos. Agentic coding is impressive and comfortable, but it makes me think about leaving the industry, as it forces me to use non-deterministic natural language. Explaining my intentions in English is often more painful than writing deterministic code. I like machines because they do what I tell them and do not ask silly questions. If I express myself incorrectly, they dump an error message without emotions. Coding agents kill the machine experience - they move computers closer to humans (emotional inefficient bloated communincation).