Den

Den

@flaticols.dev

Digital logic. Analog chaos. I frame the light, ride the descent, and rewire the house. Always covered in dust or mud.

14 followers47 following103 posts

Den
@flaticols.dev

@sketch.com please make an official integration with the ChatGPT app. I did my own, but anyway it would be nice to have it from you.

Den
@flaticols.dev

I added a mirroring feature to my blog on ATProto. It was a good exercise that helped me gain a general understanding of how AtProto works. #atproto

Den
@flaticols.dev

I like Delta. I’m using it a week every day. It feels more natural for me than terminal flow. In my head, terminal is for run, see, clean that it. Delta is like Jupyter Notebook.

Den
@flaticols.dev

What is your opinion about following canonical rules or principles? Let’s say, Clean Architecture? I’m not against it; however, I think the canonical way very rarely fits well in real life. I mean exactly each rule.

Den
@flaticols.dev

And I also strictly believe that for LLM tools, you must do an architecture. You must design it.

Den
@flaticols.dev

In my previous post, I told about one project; it was 50% written by LLM, but exactly the boilerplate part, like work with the GitHub API. Even with that, it saves me months of work (for a usable state of that project, I spent two weeks).

Den
@flaticols.dev

Last week, for some reason, I was working on, let’s say, a Postman/Insomnia tool, but very specific to us. It is quite interesting how to make it fast- for instance, a simple json/xml response view to render 10MIB of payload.

View full thread
Den
@flaticols.dev

App is aggressively flushing unused data to disk; memory footprint is also not big for that kind of app, around 100 MiB. And around 50MiB for server.

Den
@flaticols.dev

Last week, for some reason, I was working on, let’s say, a Postman/Insomnia tool, but very specific to us. It is quite interesting how to make it fast- for instance, a simple json/xml response view to render 10MIB of payload.

View full thread
Den
@flaticols.dev

Then I decided to rewrite early using GPUI and GPUI-Components. And now we have a single codebase on desktop and on server. In our case, the server can be local, bundled with the app, or remote- no matter.

Den
@flaticols.dev

Server can be running in the background without a UI. I measured app open speed by… bouncing the count app icon in the Dock! Very precise! And results are very good- only one jump (up and not landed), almost instantly.

Den
@flaticols.dev

Last week, for some reason, I was working on, let’s say, a Postman/Insomnia tool, but very specific to us. It is quite interesting how to make it fast- for instance, a simple json/xml response view to render 10MIB of payload.

Den
@flaticols.dev

First attempt was Tauri and React with Base UI. In general, it worked. Quite fast; however, on our data, it was very slow. Extracting some logic to Rust and passing a lot of data via IPC not a good idea.

Den
@flaticols.dev

Last days were very heavy, working on a project on Rust and GPUI. I like it! Performance-wise, it’s pretty good. Native-looking? Well, for an internal tool, it doesn’t matter a lot.

Den
@flaticols.dev

People already aren't thinking with their own brains and just answer, "That is written in Clean Architecture; we must do it exactly that way." With AI, it becomes even worse.

Den
@flaticols.dev

About Clean Architecture, my point is: you must read books and articles and highlight what can help you or fit you. You don't need to do exactly what the book or article says.

Den
@flaticols.dev

Figma is limiting MCP calls; that’s why I love Sketch. Local, native app, local MCP. Yes, some limitations to run something overnight, but it’s solvable.

Den
@flaticols.dev

KDD - kitchen-driven development. When you come to the office in the morning and solve problems without bureaucracy.

Den
@flaticols.dev

Overall, I have a bold idea to make a pure web components kit which does not require any build steps for the end user.

Den
@flaticols.dev

Looking at a thread full of vibecoded markdown editors. Yes, it works, and I even do the same for myself. But it doesn't touch real software for critical operations. Good luck videcode it or even debugging it in the middle of the night for a critical incident.

Den
@flaticols.dev

Also, imagine a situation something like, “I don’t know how it works; it was done by our agentic pipeline and passed all tests.” But know is 3am and customers can’t do a work? Mm?

Den
@flaticols.dev

I’ve tried to find a very specific screen recording app for macOS. I want to record separated a few windows and audio/video from the camera. So a few days ago I started writing it by myself. What I wanted to say: I think macOS has a quite good set of APIs.

Den
@flaticols.dev

I’ve tried to find a very specific screen recording app for macOS. I want to record separated a few windows and audio/video from the camera. So a few days ago I started writing it by myself. What I wanted to say: I think macOS has a quite good set of APIs.

Den
@flaticols.dev

I’m wondering why big tech companies usually can’t invest 2-3 people to make app/cli usable? Use, gcloud, I’m with you. It’s so slow and bloated.

Den reposted

Zed
@zed.dev

Code lens support lands in stable .... tomorrow! 🎉

Den
@flaticols.dev

Playing with jj. Looks like what I need to do is a really big refactoring and work in the repo. I also looked at GitButter, but it looks more like a fancy thing.