The final day also turned out to be heavy, accumulated oversaturation made itself known, I've only just recovered.
The first talk of the day from Anna Zharkova about automating UI creation using AI disappointed. The same message as the day before: from Russia your ChatGPTs don't work, so we are forced to cry, prick, and poke around in GigaCode. Well, and in Gemini for some reason we'll also poke around a bit, although it's also not particularly available. And the whole talk is just about how they don't generate properly and make up a lot of non-existent things. I don't know about you, for me this wasn't news, but it was interesting to see that it's alive at all and there are even some plugins for the studio. Claude and ChatGPT with the latest updates are currently alternative-less favorites for me and if we check something for strength, then rather them.
Moreover, "UI creation" is not just generating a bunch of basic Compose components. For me, normal layout implies reusing concepts already existing in the project like tokens and components from the design system, otherwise it's throwaway code. And this wasn't touched upon at all. There's an intuition that for some Claude, digesting in context function signatures and tokens from the design system + json exported from Figma API into tolerable Compose code is quite a feasible task, although my hands haven't gotten to checking this yet, however I left myself a note.
That is, both talks about AI, both about tests and about layout, implied some specifics of Android development, but turned into a review of plugins and models that have nothing to do with it. This was not very pleasant.
The final session of the week from Misha Levchenko was about writing your own automation service. Also quite high-level and therefore just easy to perceive for anyone, but leaves a bunch of interesting thoughts to think about. If you put the message in a few sentences, it will be approximately like this: write automation of everything in the form of code, by analogy with any product, because developers are more familiar and comfortable with this. It's easiest to write in the form of a backend, without any user interfaces. Well, and you need to choose technologies as developers like, but a little with a look around at what's accepted in the company. This is unconditionally the headline performance of Podlodka because besides interesting info, the format with real paper "slides" instead of a regular presentation looked super original. How many trees were spent on preparation is unknown.
TIL that this even exists somewhere. By default, when you think about automation, what comes to mind is rather customizing CI and all sorts of existing tools at the Jira or GitLab level. Although not even quite like that, the idea of using APIs of all these services is in the air and from time to time you write something like that auxiliary, but to take this and everything together in the form of a backend with a database to deploy somewhere I haven't gotten to yet. And now it's clear that this to some extent solves many problems. And creates them, yes.
Overall for the whole week, what can be said. Due to the topic, almost all talks for me turned out quite simple and often repeated my own experience. This is both a plus and a minus, the plus is that it perfectly validates my decisions, the minus is that I didn't hear much new. But between the lines I sometimes managed to pick out interesting details. As always, I took away more benefit for myself from the Zoom chat, Telegram, and the question block after sessions. The preparation for the talk from some guys I really liked, I will clearly re-watch about a third of them precisely from the point of view of improving my own talks in the future.
Thanks to the program committee and speakers, it was interesting.