Tag: programming

The Global Agile Summit ended on May 6. I watched these five recordings a couple of weeks later, after Oikosofy released them to the video library. A mob-programming workshop from Woody Zuill, four live Q&A follow-ups with Jason Little, Marcus Bullock, Clinton Keith, and Joshua Thaler, roughly five hours of material compressed into a week of evenings.

The delay was unintentional but turned out to be productive. Watching sessions that happened days apart, back to back, in a quiet week, surfaced a thread I would not have caught live: every one of these speakers, from different angles and different tracks, kept arriving at the same operational question — how do you keep the human directive when the tools are fast enough to make the human optional?

Zuill answered with a game. Little built a folder structure. Bullock discovered it when four weeks of automation passed without anyone talking to a customer, and Keith when a parking-lot headcount measured the wrong thing. Thaler brought a sword.

I switched the news classifier from GPT-5-mini to GPT-5.4-mini and right away a bunch of news that isn't about cats started slipping through — did the model degrade, or do I need different prompts? I originally wrote the prompt for GPT-4-mini and never changed it

Rolled everything back to GPT-5-mini. Turns out GPT-5.4-mini is an order of magnitude faster and several times more expensive, but the quality of its work is orders of magnitude worse. A model for people who care about speed. But I don't get it — an AI that does the job several times worse ends up taking more time, because you have to fix more. What's the point of new models like this?

I use Claude Code with the Oh My ClaudeCode plugin for my daily coding tasks. Its orchestration workflows have AI agents review their own output in a loop, catching issues and verifying fixes automatically.

But after using it for a while, I noticed a recurring problem.

When agents share context across iterations, they mostly verify that problems from the first pass are fixed. They rarely find new issues. The shared memory works against me.

For more than a month now I've been actively vibe coding at work, mostly using Claude Opus. Since my duties don't let me focus on programming, the AI helps a lot. I tried vibe coding a personal project, without using the premium paid requests from my work subscription.

First I tried coding with my own API keys. In two or three requests Opus charged me 2 dollars. Then I tried GPT-5.2, where a million tokens a day are free; the free ones ran out fast and after that it started costing money, though maybe you could do about a dozen requests within that million.

In the end I tried GPT-4.1, which is included in the corporate subscription and has no extra charge. It got the job done, but made elementary mistakes, and fixing them took a lot of time, plus the code quality clearly came out mediocre, and it fixes problems very reluctantly, you have to dive in yourself and give instructions, and then the point of the AI disappears. It doesn't seem worth it, I'll try getting a Claude Pro subscription for a month and see how it goes, after all, I don't work on the pet project constantly, the mental resources for it aren't always there, though maybe the AI will let me do it regularly, since it requires fewer mental resources.

Last month the host of my personal project changed servers and everything went down. I immediately changed the A record in the domain's DNS settings, but I changed it at Hurricane Electric when I'd already moved the domain to Cloudflare and needed to change it there. As a result, the site didn't work for almost two weeks, because the main job left no time for the personal project.

I fixed the DNS record in the right place, but after that the API service still wasn't working because of an SSL error, even though its domain pointed to the main one via CNAME — but the main one had Cloudflare proxying. I changed the CNAME to a subdomain of the main domain for which the proxy is turned off.

After all this, both services kept working improperly or not at all.

For my big cats news service, the entire admin interface lives inside a Telegram chat with a bot. I spend a lot of time in Telegram anyway, so it is the easiest place for me to review, choose, edit, and translate news items.

Each post also needs a photo for the website and social media, plus a caption for visually impaired readers. Sometimes the picture cannot be downloaded directly, even when I try ScraperAPI. This recently forced me to publish a couple of articles by hand.

I have now added a simple workaround. Because the admin panel is already in Telegram, the app can still fetch images that other tools cannot. After the bot sends a draft to the chat, my code checks whether the article already has a local photo file. If it does not, the script downloads the picture from the Telegram message and stores it so the rest of the pipeline can run as usual.

I immediately developed the previous application I built on production, as it was a small personal API app. However, when I decided to recreate my website, I chose a local system, which allowed me to experiment and freely work with new things.

My chosen local hosting solution was the most recent version of XAMPP, with PHP 8.2. While this isn’t the latest PHP 8.3, it’s good enough for development. And you can still upgrade PHP manually if needed. Still, I need a different PHP version for my work in the system PATH variable.

Since most course instructors use aliases for PHP and the Laravel Artisan console, I set up similar ones in PowerShell:

Over the past few months, I have worked on personal apps in Laravel, developing an automatic posting system from Telegram and a semi-automatic photo-posting service that uses Creative Commons images from Flickr to various social networks. I plan on discussing these projects in separate articles soon.

This made me want to renew my website using a Laravel-based stack. My website hasn’t seen much change since I built it a decade ago using Joomla ZOO CCK. I got everything I wanted out of it, but the site stayed a pain to manage and update. Given the technological advances, having such a dated personal website is almost embarrassing, not to mention the language of the content.

However, due to time constraints and personal preferences, I’m not inclined to build a website from scratch on a new framework. I initially considered Winter CMS, but it didn’t suit my needs after getting to know it through videos and documentation. The ideal system for me is easy to install and maintain. I don’t wish to spend time remembering complex update procedures, which I will forget after six months.

Bought a cheap VPS for 9 dollars a year. I spent the last few days figuring out its configuration — I'm a developer after all, not a devops or sysadmin, so I have to spend a lot of time on manuals and fixing basic errors that I made myself :)

But I moved my Friendica server whitelion.me from Moscow to Atlanta. Even fewer ties to the RF, and honestly I felt uneasy keeping a copy of my social-media posts there...

Although I had to fiddle with Friendica updates too, now I've got the process sorted out.

For over a month now I've been actively using IFTTT for cross-posting to social networks. I want to share how I got there and what tricks I resorted to in order to achieve cross-posting close to what I need.

I write on social networks for myself, not for work. And it's convenient for me to write from the place where I spend my time, so SMM software didn't suit me — it's expensive, and on top of that it requires writing from a different place.

Mostly I spend my time on Twitter and Telegram, I post photos on Instagram, and recently I also started doing a podcast on Anchor. Some messages are more convenient to write in one network, some — in another. Besides, I'd like to have at least some control over my messages, so as not to lose them in an instant because of some corporation's decision. That's why I also joined the decentralized social network, the Fediverse, and set up my own personal Friendica, where I simply duplicate my posts from other social networks.

When choosing a solution for joining the Fediverse, I settled on Friendica, because it's in PHP, which I work with every day at work, and if needed I can modify the engine to suit me and deploy it on shared hosting. The only thing I don't like about Friendica is that it doesn't allow guests to leave comments, but since it's more of a backup solution, that's not a critical problem. The other apps I tried or looked at had more critical drawbacks for me.

Many people are switching to decentralized social networks because they don't want corporations tracking them, analyzing them, knowing everything about them. That doesn't really bother me, I've come to terms with it, but what bothers me more is that a corporation can at any time delete all my messages, all my history.

Copying messages by hand is too much of a hassle for me and too costly in terms of time, so I started looking in detail at solutions for automatically posting messages between social networks.

I know it's not original, but I decided to write this to let off some steam. Since my parents have spent their whole lives working at a combine plant, I decided to describe the kinds of requirements we deal with at work in a way they'd understand, even though I myself know next to nothing about combines.

Hello, we have purchased an American "John Deere" grain-harvesting combine; from the look of it and from the reviews it's very high-quality and functional, but it lacks a couple of mechanisms and its appearance doesn't suit us, so put the body of the Ukrainian "Slavutych" combine on it and build the mechanisms we saw in it, and also such-and-such mechanisms from the Belarusian "Palesse". There are no blueprints or documentation for the combine, since it's a closed development and we don't anticipate modification by others, and therefore there's no way to get in touch with the engineers of the original machine.

All rights to the resulting combine will belong to our company; don't forget to paint our logo on it; it will be used in Russia under the brand "Sverdlovsk".

A year later.

We noticed that after the body was installed and the combine repainted, some of the American mechanisms jam or don't work the way they should — fix them, and also we decided to use the combine for harvesting a larger number of grain crops, so the mechanisms you built earlier need to be expanded and reworked.

anyutt82 announced the site http://www.edasovkusom.ru/. So it can be considered launched. I'm not into cooking, so I can't judge its content, but I worked on its software side and want to write about that. Before this I hadn't made such complex sites on Joomla, though initially nothing foretold trouble, except for the fact that the whole functionality and design was begging to be done on WordPress, but the client's categorical requirement was Joomla! The outwardly simple little site was handed over to a new programmer, and I went off on vacation to finish up the three online stores hanging on me.

Remotely I helped with building the site all week, but in the end the boss handed the development over to me entirely, since the deadline was running out and the project had to be finished urgently. The functionality turned out to be unusual, it was the first time we were making something like a blog on Joomla. I had to use the Tag component for adding tags to recipes and the MultiCategories component, which replaces the standard Content component, since a single recipe has to be in several categories. To display a thumbnail image in the intro text the JA News module was used.

All in all, I enjoyed the work on the project and it stuck with me. I see in the office it got the others pretty fed up, and some even cringe at it. But I like that kind of thing, I like it when the client knows specifically what they want, even if the requirements are a bit unusual and strict. That's better than when you have to come up with part of the site's functionality on the fly during the build, and then rewrite it on top of that when, after showing it to the client, it turns out they want to redo part of it.

Got around to changing my journal's style. I'd gotten used to the Forest Green theme, which is based on the Bloggish layout, but in it only 470 pixels are allotted for the content, which is clearly not enough to comfortably place photos and read the friends feed.

I dug into Custom CSS and changed the size of the text field to 640 in width, and also increased the width of the sidebar, making the overall content width 1000. At the same time I bumped up the font sizes by a couple of points, since pixels on monitors keep getting smaller. For search engine optimization I moved the sidebar to the right, so the entry's content comes first and gets indexed, only it looks somewhat unusual, I'll have to move it back to the left and do an SEO layout; on my own site I managed to do this, and it'll work out in the blog too, a pity I don't know CSS well.

After changing the width of the main blocks, the blog's header turned out ugly, with a narrow image in it. I decided to put a more eye-pleasing picture in the header; a part of a photo of a white lion, which I mentioned here, worked well for this. After replacing the header, a change to the blog's color scheme followed; I picked the new colors with the eyedropper from the header. I didn't completely abandon the old greenish ones — I used them for the link colors. I decided not to change the background image either — it doesn't get in the way much and an alternative hasn't come to mind yet.

I set up a library on my site using the BookLibrary component. Yes, the quality of the component leaves much to be desired, since it was originally written for Joomla! 1.0, doesn't use MVC, and everything in the code is jumbled together. But I couldn't find anything better for building a library for Joomla! Plus, thanks to its simplicity, some bits of the code can be tweaked to get proper display and operation. It would also be nice to attach SEF links to it, but after JoomGallery's SEF links that shouldn't be a problem, if only the will were there. If I end up doing it for someone, maybe I'll get around to little details like that — not for nothing did I work on an electronic library for my thesis in my 4th year, only it was in ASP.NET.

I did this not only to explore the component's capabilities, but also to organize my home library. I added only the books I bought myself. Two bookcases of my parents' books, my childhood books, and textbooks would be too much to add to such a catalog. It's a pity that the chance to buy literature for myself came only after I got a computer, by which point I'd almost stopped reading books... The desire is there, but I can't manage to read. The last work of fiction was "When the Mountains Fall (The Eternal Bride)" by Chinghiz Aitmatov, which I read over the course of the winter on a smartphone that I've since lost. I can't recall the work of fiction or non-technical book before that.

Today I added four more books to the library, three of them needed for professional work, and one finally about lions — a book by Craig Packer that more than one article about these big cats references. Here's the link to my library with books about lions and a few technical ones on development.