Mahdi Yusuf

At 250 billion DNS cache entries, one wasted byte costs 250 GB of RAM. Five Rust optimizations later: 100 TB freed, inserts 43% faster, lookups 19% faster. We didn't trade speed for space. https://cfl.re/4wOegyp

@larchuk.bsky.social did you reach out to me on twitter? dm?

ssh ssh.mahdiyusuf.com check it out.

ssh ssh.mahdiyusuf.com check it out.

@softwaredoug.bsky.social can I buy some of your old courses still or enroll in them and get the content now? It’s inconvenient to have to wait for course dates.

www.mahdiyusuf.com/looping-casi...

iddqd, or the hardest kind of unsafe Rust. My first post for the @oxide.computer blog! Goes deep into the pitfalls of unsafe Rust and how to reason about it. A lot of labor went into the post -- please have a read! oxide.computer/blog/iddqd-u...

Speed Tax www.mahdiyusuf.com/the-speed-tax/

So what do people think about sliced tomatoes on a pizza?

@rickypadil.la sup

America outsourced manufacturing, then the skills to design what gets manufactured. Software is next. mahdiyusuf.com/the-journey-...

America outsourced manufacturing, then the skills to design what gets manufactured. Software is next. mahdiyusuf.com/the-journey-...

I built more with AI agents in two months than the previous year. I used almost none of it. The real productivity gain isn't building—it's using. mahdiyusuf.com/your-agent-i...

Why your coding agent keeps undoing your architecture mahdiyusuf.com/why-your-cod...

Why your coding agent keeps undoing your architecture mahdiyusuf.com/why-your-cod...

Wrote something about the importance of the code generation and library use in the context of agents. mahdiyusuf.com/agents-didnt...

Experimenting with a Touch ID wrist rest 👀 Early prototype, lots to refine—but the idea feels right.

MCPs give me the same energy as GraphQL: works just well enough that people adopt it, just poorly enough that you inherit a new class of problems. Both exist because building proper context and tooling takes effort. Most would rather reach for something "good enough."

Playwright MCP is too token heavy what else people using?

First custom keyboard build ✅ Split layout, fully custom case + PCB — designed from scratch and 3D-printed. Still a work in progress, but seeing it come together is fun. Firmware next.

Anyone have a cool set up for iPadOS for running terminals for using Claude code?

Has anyone set up a self hosted ghost instance using 6.0 and manage to get up with new built in activity pub integration?

anyone know of any good image generation LLMs that are good with icons or generation extension logos. I am surprised this isn't an area of focus.

I was looking for ways to vibe with llms using voice to text in rooms with others. this product did not disappoint.

OH: Does CI really need to run on every merge? Can't it reasonably be once a day or something?

This industry gives me a headache sometimes.

I wonder which happens first, we start building tools for agents to use to get better context or agents can hook in. I am surprised the things they do everyday to get context they need.

Anyone have any cool workflows for Claude code?

What do people use for photo editing on the iPhone?

Goodbye, Fingerspitzengefühl? architecturenotes.co/p/the-cost-o...

anyone doing any writing on how to think about what to include in an vector embedding? most advice I read is just throw everything in there. 😬 I know most things will be specific to your project, but would be interested in what people have done.

top of the mornin’

PS2 startup still hits me hard.

I just finished printing this cool 3D holder for my desk. It’s a wire holder that keeps my wires from falling off the back of my desk. I think it’s pretty neat!

We are back!

verifications are here!

if you write a comment and there is only an llm to read it? was it even read?

honestly 2fa in emails is so annoying and unsafe.