🇪🇺

I have to migrate my life from Google to Proton.

These token sellers really smell their own farts and find AGI.

Anthropic is the new evil.

I didn't know that, it's handy:
platform.openai.com/tokenizer

I wish it was in Rust, or another modern compiled language.

Distrobox is handy.

It suits well with Fedora Atomic and works fine alongside with Toolbox.

Someday I should write a blog post how I use Fedora Cosmic Atomic, Brew, and Distrobox.

I think I'll give MastraCode and Pi a try.
Curious about pros and cons compared to OpenCode.

code.mastra.ai
pi.dev

What a circus joke, the world is going crazy over Claude leak, while others have been open-source from the start.

Soon token sellers will provide both the poison and the antidote.

I asked LLM to build a breadcrumb system for Svelte. As the result I had very complex design, trie algo, many files and lines of code, and a Vite plugin.

I asked the same thing but by specifying the use of LayoutLoad. Same result, zero complexity, use of built-in features.

Conversation AI powered by Voxtral: Pizza Ordering YouTube video by Mistral AI

Introverts around the world are so thrilled

www.youtube.com/watch?v=qJ9Q...

automatic reload when file changes externally · Issue #1125 · helix-editor/helix It would be great if it were detected when the file you are editing has been changed externally, and if possible, automatically reload the file. I imagine it could work similarly to vim and kakoune...

I give up...
Not handy to use it with opencode without auto reload feature.
github.com/helix-editor...

Otherwise, I really liked it and I think I'll switch to once more features are supported. Or at least I use for minimal tasks.

OK, I'll give helix a try for a month.

stack: TS/Effect/Node/Svelte

It'd have been great if nvim could be as fast as helix.

My agent skills for EffectTS, Svelte, and Tailwind
github.com/akhansari/do...

I'll add Drizzle (once V1 released) and Hono.

My tech usage:
Before 2023: 90% Microsoft
After 2023: 10% Microsoft
Absolutely nothing is missing for me except F#.

I did the same for Effect and Svelte/SvelteKit.
It just asked me how to split to smaller files, and done!

Amazing to see how the output quality has increased.
Much better than opinionated skills that we can found on the web.

Cool tip: You can ask LLM to build skills based on llm.txt

Like:

Build DaisyUI agent skills based on:
- daisyui.com/llms.txt
- opencode.ai/docs/skills/

My favorite dev workflow with or without GenAI:
Function Driven Development

With GenAI:
- I model the domain with types
- I write an empty function with the desire input and output
- Either I let the LLM to complete and I review
- Or I write myself and ask GenAI to optimize and simplify.

#opencode should be rewritten in Rust or Go.
There is too much CPU and memory usage and spikes and performance issues.

It's super rare to see good readme files. It's an art like a good code.

Repos README:
- Before LLMs : Empty readme, 0 value
- After LLMs : 1k lines readme, 0 value

Just blogged. The slop coding era!

akhansari.tech/the-slop-cod...

LLM: You're absolutely correct!

Calling LLMs AI is one of the biggest marketing scams.

people get so mad when i'm harsh on vibe coding

then they get really confused when they realize i work on opencode

which makes them even madder

Agent Skills Are Spreading Hallucinated npx Commands AI agent skills are propagating hallucinated npx commands, creating real security and reliability risks for developers and supply chains.

lol www.aikido.dev/blog/agent-s...

This is the only AI I want to use YouTube video by The PrimeTime

Oh pretty cool, Primeagen has built the perfect tool for my workflow. I have to test it.

youtu.be/ws9zR-UzwTE?...

akhansari.tech/the-right-wa...

Just blogged. Mise: the underrated tool.
akhansari.tech/mise-everyth...