Joe Muoio

Pretty crazy to see so many articles talking about issues of “AI“ that are really just about the private credit crunch. Case in point: www.reuters.com/business/bla... So Fink says we all need to buy equities. ignore the fact that wealthy folks have more so it doesn’t actually help inequality…

And here is Reuters own reporting from a week ago that actually shows Fink’s motivations amidst blackrock capping withdrawals. www.reuters.com/business/fin...

Wondering how other's juggle multiple CLI LLM client sessions. I will frequently have sessions: 1) planning work (assuming current task is not too cognitively heavy), 2) implementing features from a plan 3) fixing bugs 4) fixing tests/build issues How do yall use claude code?

This is my max number of active sessions. It's tough to work on future features 1. because I need to sandbox or set up multiple repos and keep them in sync (do you do this?) 2. It just becomes too overwhelming for my squishy brain to keep up with all the changes and context switching.

Wondering how other's juggle multiple CLI LLM client sessions. I will frequently have sessions: 1) planning work (assuming current task is not too cognitively heavy), 2) implementing features from a plan 3) fixing bugs 4) fixing tests/build issues How do yall use claude code?

I gotta write an updated blogpost on my use of AI in 2026. The short is that I have embraced throwaway software. It's become so much easier to prototype ideas and to build simple programs for testing something. And for long term projects, the basics of software development are unchanged.

Trying a new Claude Code technique with Opus 4.6. Wrote a detailed plan with my constraints with Claude formatting it for "future Claude". Created a slash command that takes a plan document with instructions to spawn sub agents with instruction to surface ambiguity to me instead of assumptions.

We'll see how this goes. Through phase 1 and it wrote a commit. Onto phase 2... I'll test it all when I'm done. 😎

Every time I go "full vibe code" It doesn't work great if I need to maintain that software. While I can 10x productivity on initial prototype, Claude code seriously degrades without proper structure and planning. So I end up spending days refactoring the mess I created.

LLMs really boost your productivity when you know what you want, and really hamper it (long term) when you don't. This project involved a lot of experimentation and I probably should have reached for something easier to iterate with in the first place :)

Every time I go "full vibe code" It doesn't work great if I need to maintain that software. While I can 10x productivity on initial prototype, Claude code seriously degrades without proper structure and planning. So I end up spending days refactoring the mess I created.

Learned this lesson again. Coming off a week of refactoring some frontend code from KitaJS serverside templates + vanilla HTML, CSS, and TS to instead use SolidJS + Tailwind (kept KitaJS). Needed to orchestrate state of up to 1300 different data calls and it was too error prone with vanillaJS.

Going to be posting over here more than Twitter. Starting from scratch on the followers list 🤠 I'm Joe. I'm a Software Developer with over a decade of experience (primarily Amazon). I now help small businesses build software and make decisions around software. I like food, ceramics and games.