I love seeing this more in libraries. The LLMs.txt and an MCP to configure the library. I recommend using the LLMs.txt since you can't verify the MCP, but still.. www.better-auth.com/docs/introdu...

The plan mode in claude code makes a huge difference. I love how my context-engineering prompts help me plan stuff. But it involves using extra files. The plan mode in claude code doesn't involve any extra files, but does the same job. I wish that github copilot had a similar option.

Finally! Just created steam accounts for the kids so I no longer need to share my personal account. Included parental controls so the 16+ games don't show up. No need for violence just yet :P

Having a lot of fun during hacktoberfest. Figured out how to build a multi-module quarkus application with Maven. Today I'll be writing a blogpost about it. And then it's off writing some more fixes for this awesome library.

About to go out to work on open-source during the Info Support hacktoberfest. Wondering if I should bring the dygma keyboard I use at home and my mouse.

Looking at automating my computer setup a bit more with Ansible to configure packages etc. In case you missed it, I did make a dotfiles repository to store my configuration files. That alone saves me a ton of time: github.com/wmeints/dotf...

People are asking me: Are they stopping work on Semantic Kernel? In a way, yes they are. But they moved everything over to the Microsoft Agent Framework. devblogs.microsoft.com/foundry/intr...

Managed to get the new cosmic desktop environment running from system76. It's in beta and it has a few problems. But so far, it's been pretty stable. I'm missing a bunch of keybindings though :P

I love how the developer behind Ruby on Rails thinks: Go away with the ceremony and give me good stuff. One of the examples that proves this: omarchy.org An interesting approach to setting up a workstation for building web application. And why not?

Building a more complex solution with context-engineering today. One lesson learned: You need to tweak your prompts and instructions more to incorporate the necessary guidelines. But it is certainly doable.

One important aspect that all coding agents haven't figured out is performance. The LLMs are rather slow at the moment and frequently overwhelmed, this slows down the development process quite a bit. I frequently find myself switching to a browser window to read articles while the agent codes.

It's amazing how powerful you can make your coding agent when you include proper instructions. I just added another layer of context to the instructions and now I have to tell github copilot even less.

The AUR repository for Arch Linux is great for installing all sorts of packages, but sometimes I run into issues where the application itself wants to update, but can't because it's installed through AUR. Luckily, you can download the contents of the AUR package and modify it :D

Weekend! Time to get back to my rust app :D

Uh ooh, I split my project into multiple crates. Am I a pro Rust developer now?

Some shiney stuff is happening over at system76: system76.com/cosmic

I'm thinking of saving up for an ARM based workstation with linux. Just because I guess, but I'm also extremely curious how the performance is going to be for generative AI.

⛓️ Still trying to solve complex problems with one massive prompt? Complex chain-of-thought prompts are "chain-of-problems" prompts. The more you ask an LLM to do in a single step, the more likely it is to fail spectacularly somewhere in the middle.

The one thing I learned from building apps fully with AI using a spec-driven approach: The base tier subscriptions aren't enough. I just fired off 45 messages in a 1 hour timespan and now I have to wait 4 hours before I can send another 45 messages.

Renaming an app after building a signifant portion of the docs and code. Not advisable, but the web domain was taken 🤣

I am trying out rust for the third time now, and it is clearly growing on me. I wouldn't use it for things like websites per se, there are more productive ways to build software. Command-line utilities is a different story though. It's great for building terminal things.

Now that I'm using claude code for longer periods of time I'm running into some pretty annoying limits. "Approaching the 5-hour limit"... Oh well.

Working yet more advanced experiments building software with agents using only specs as input. Follow along here: github.com/wmeints/refo...

Extracting the prompts I used for an internal AI lab so that my colleagues can use them for their own projects. Looking back, I've found yet more ways in which I can improve the templates :D

Slowly turning my context-engineering git repository into a more engineered experience. I hope to have a full workflow ready soon.

Building a fun rust CLI utility for an AI project. Because I can :P

This package saved me a lot of time building test data for my demo app: www.npmjs.com/package/rosie

Great tunes for those AI-fueled coding sessions: youtu.be/KoIccnsxayw?.... At least to me they are ;-)

Claude Code with Neovim is awesome! Getting an AI workflow ready that's going to process news feeds :D

Some very "strong" typing there typescript. Poah!

I'm refactoring an application that I made with AI to transform so that we use atomic design principles. medium.com/@benfluleck/... The benefits: The risk of building spaghetti code is much lower than without these strict guidelines. Also, I found myself reusing much more of the code.

Every time I use AI to fully implement scenarios into an application I find that my mind wanders into other things, because I have to wait super long before it is finished. It's really annoying.

Doing some work building an inngest workflow with AI. Those RSS feeds aren't processing themselves you know.

I propose a new term: Excel-driven platform architecture.

🎯 Tired of writing regex patterns to parse chaotic LLM responses that break half the time? Here's a game-changer: You can force LLMs to output perfectly structured JSON that your code can reliably parse—no more "output this in a table" prayers followed by regex nightmares.

Sometimes I have a very fun job. Made an awesome video with Joop showing off an AI lab we worked on before I left for vacation. Now making pretty pictures to go along with the video to help people understand what they're looking at!

Time to finish building a demo for next week. It's fun doing some typescript after a day of discussing data platforms.

One inbox down, three more to go. Starting up after a vacation is a bit of a nightmare if you ask me.

The first day of the new school season is upon us! Wish me luck 😅

Just updated my personal dotfiles repository to include the everything for the hyprland configuration I'm running at the moment. Enjoy: github.com/wmeints/dotf...

The most enjoyable part of being at home during the vacation. You can be all over the place with your programming. Did some golang programming and now I'm back doing some interesting stuff in rust. Because I can!

Just upgraded my neovim configuration to the latest Astronvim v5 base installation. I added in some of my customizations, but wiped the rest to get a clean slate :P github.com/wmeints/astr...

About to run gpt-oss on my home machine to see if it's any good at generating podcast scripts. Just for fun.

Just updated the samples in my book to the latest semantic kernel version for Python and .NET, enjoy! Haven't seen my book yet? Go grab it here! leanpub.com/effective-ll...

I just installed linux on my wife's (old) laptop. Speed times 5-10, and the battery suddenly holds a charge for more than 10 minutes. And the nagging that the laptop should be replaced because it can't run Windows 11 is over :P

Just found GSConnect to link my Android phone with my Linux desktop. Works awesome. And there's a KDE connect too in case you're using KDE. extensions.gnome.org/extension/13...

Yay! Postgres 18 is out. I've been using this database for my personal website for over 15 years now. It's crazy scalable and has a ton of extensions.

Making ribs. Nomnomnom.

Good news! OneDrive works on Linux too! You can get it from here: github.com/abraunegg/on... Make sure to read the usage documentation under docs/usage.md so you get the best results.

Moved my full machine over the Arch Linux instead of dual boot with Windows. Made a BTRFS setup with two SSDs ending up with 4TB of storage. And I moved my home directory to a separate SSD of 2TB storage. Should be enough for the coming years :P