Will Mruzek

@theo No one holds a SM7B

Welcome to the FIRST EPISODE of TBPNN (Theo & Ben Podcast News Network) jkjk we're calling it Nerd Snipe. Available on Youtube, Spotify, and coming to all your favorite podcast networks soon™

I love it when I explicitly revert an LLM change, and the agent is like "Nope, I put that there for a reason." Who's the boss here? 😂

We're going to make consumer routers more secure... by banning firmware updates! 👏 🤠 cnet.com/home/internet/…

Review every single line of agent code. Number of lines written is not a flex if quality is lacking. Slow down. Must watch material. youtu.be/1r9n-HsBQsE?si…

"AI Engineer" is too broad to be useful. - Do you write code _with_ AI? - Do you write code _integrating_ AI? - Do you _write_ and _train_ the LLMs? It's unfortunate engineers feel the need to label themselves this way to get noticed in the job market.

Spent a month in NYC drinking a lot of coffee and wrote up the spots that genuinely impressed me — plus a quick explainer on third-wave coffee and why lighter roasts are worth trying. ☕ willmruzek.com/thoughts/2026/…

@kitlangton @opencode Congrats! Does this mean Effect in OpenCode?

For my attention span deprived compatriots, here's just the announcement part. 🫡 I'm going to work at @opencode. It's all terribly exciting.

@RhysSullivan Slop?

@tristanbob @dillon_mulroy Agreed. Context engineering is your agent app’s value proposition

Hot take: If you are BUILDING an agentic app, context engineering is the most important thing to get right. But if you are USING an agent app, you should NOT have to worry about context.

@theo Obviously, they need to write the compiler in Rust

Rust build times are so bad holy shit

@dillon_mulroy Amazing times we live in

i have clarity for what i want/need to build after 3 weekends of prototyping different ideas and mvps and using other peoples implementations/ideas. v. excited

History: made. Dreams: come true. @IUBloomington: first-time national football champion. Congratulations, @IUHoosiers. You've made Indiana proud.

Go Hoosiers!

@tannerlinsley I prefer: Make it so

> Ready to proceed with implementation when you are. Go.

@kitlangton A good problem to have haha

Thanks, everyone. I guess I finally have to pay for Resend 🫡

effect.institute ← tomorrow

@leerob I think AI is in a great place for coding! I just won’t trust it 100%, especially for tech I have no experience with.

The YouTube comments have been even worse. I feel like the video is pretty balanced/pragmatic and still, faced with “AI”, many folks are just in denial.

It’s been interesting to watch two drastically different reactions to this. 1. Yeah, AI is changing software engineering. Let’s figure out what that means. Cautiously optimistic. 2. Denial.

@RhysSullivan claude -p “Fix this bug. Don’t forget to rewrite the project in @EffectTS_.”

i used opencode to fix opencode yesterday and walked to walk through the process & how i'm thinking about AI & OSS the bug was after i sent a message, i got no response from the LLM i did some initial investigation myself and narrowed it down to my opencode config, specifically the MCP section, as it was working in all projects except one i already had opencode cloned on my machine, so i ran `bun dev` in that folder and prompted it with my initial investigation it then narrowed it down to not having a timeout on connections, so i had it generate the command to run opencode on the problematic folder, checked to make sure it was working there and it was this is where i could've told opencode to open a PR and gone on with my day as it was working, but that's the wrong approach imo as you need to roughly understand the changes your making i saw that it was using an existing withTimeout function that didn't need to import from anywhere, so i had it look into where else timeout was being used i then see in another location there's a default timeout of 5000, so i decide rather than using the random numbers claude generated, we should make that timeout a shared constant & am feeling more confident in submitting my PR so i then have opencode make the PR for me using the github cli, clean up the description, and it's merged in under an hour this is a long thread for a tiny fix, but it highlights a few things i've been thinking about recently - open source matters more now than ever, it sucks there was a bug but i love that i can just fix my tools and continue working - fixing bugs is easy and using AI for them is good, but be intentional in the changes you're making if you're curious, here's the full chat history opncd.ai/share/NY6eSsb6

I’m excited for the future of software given the progress of these tools. However… I’d be wary of creating a project in a technology I have no experience in. How do I know it’s writing good code? How do I know its answers are up to date? I only understand parts of Rust, so I would probably brush up from resources besides LLMs. It’s not an either/or, but I personally would want some knowledge to audit LLM output. Same if I were to hire engineers who use these tools.

It's 2026. AI writes most of my code. Now what?

@RhysSullivan Are you trying to start a war?

Curious about how encryption works? This guide walks you through some of the, ahem, key concepts. ssd.eff.org/module/key-con…