Buddy Hadry

After a chat with Grok dissecting why it feels like I'm always posting into a void here, Grok came to the conclusion my account is likely permanently throttled. Some early stupid follower-churn moves around 2010, followed by ~7+ years of account dormancy, resulted in an unremovable 'low-trust' scar. My reply reach and engagement was good, but originals pretty much always capped at 40-100 views. So following Grok's advice, I'm moving to a clean slate. New account (same name): @buddyhadry If you were one of the 42 people 😂 that ever saw my posts and liked them, follow me there, where I plan to post more about my current Claude orchestrator & personal OS setup.

After a chat with Grok dissecting why it feels like I'm always posting into a void here, Grok came to the conclusion my account is likely permanently throttled. Some early stupid follower-churn moves around 2010, followed by ~7+ years of account dormancy, resulted in an unremovable 'low-trust' scar. My reply reach and engagement was good, but originals pretty much always capped at 40-100 views. So following Grok's advice, I'm moving to a clean slate. New account (same name): @buddyhadry If you were one of the 42 people 😂 that ever saw my posts and liked them, follow me there, where I plan to post more about my current Claude orchestrator & personal OS setup.

Today’s project: letting my Claude Code based personal OS train itself. Via remote terminal app on my phone, telling Claude to use my orchestrator system to spin up multiple Claude Code sessions in parallel. Each worker analyzes a set of repos and writes project context into a shared sqlite brain (features, pain points, roadmap, etc.). Goal is for the assistant to have enough structured context to suggest plans and future orchestrations on its own.

That’s actually great, I don’t think I even considered keeping it just files when building my system. I jumped straight to a SQLite layer so I could hang a front end on project/task/repo views and have Claude code suggest orchestrated runs, but who am I kidding I mostly just use CC as the UI anyway.

@buddyhadry no database, just files. KISS principle

This is how Clawd woke me up this morning. All the building blocks are here. Took me 2 days to glue things together. Didn't even need 150 Million in funding. warelay.dev

I’ve turned Claude Code into a personal assistant / personal OS: – cron jobs keep an sqlite “brain” in sync with git, uncommitted work, projects, tasks, deadlines & more – daily /workday:brief session that summarizes yesterday + today – commands like /capture (made aware to cc in global settings so it knows to capture ideas, tasks, etc) and /handoff to grow the brain and pass context between days – tmux & mcp based orchestrator that spins up worker Claude sessions (or an orchestrator session to adopt another already active session) and approves non-destructive commands Curious who else is building this kind of personal infra with Claude or other agents.

@thekitze The “oh shit” and “fuck, you’re right” are killing me. 😂 What’s in your rules file?

opus just completely ignored my rules file .... fck thankfully this was with local db

Fully open source, local semantic code search for Claude Code that works. osgrep -v2 is live! 36% faster answers, 23% cheaper, 70% win rate

I'd love to see the Claude Max subscription upgrade stats this week.

@dagon_duke @Timcast I can't decide which I like more. The fact that you were wrong and he posted this video himself, or the fact that you used AI to write this.

The most chilling part isn’t that a major party candidate just ran a blatant deepfake ad it’s that the Acton campaign didn’t even try very hard to hide it. The lip sync drifts the lighting is wrong the cadence is off by half a beat. It’s the AI equivalent of leaving fingerprints on the murder weapon. They’re not testing the tech they’re testing us. How much obvious forgery will we shrug at before we demand consequences Because once everyone knows it’s fake becomes the new normal the next version will be flawless and by then the outrage muscle will be too atrophied to matter. Ohio isn’t just a governor’s race anymore. It’s a stress test for whether reality still gets a vote.

“If we are playing this game, we are done.” A Democrat candidate circulated a fake AI video of Vivek Ramaswamy promoting something he never said.

@thsottiaux @__nmca__

@__nmca__ Will consider if the model is still considered frontier by end of year

Codex team is working on a few experimental projects that are starting to shape up and I’m excited to share more about soon. But I’m curious, what would you like to see ship or improved by the end of the year other than better models?

I've now been testing Opus 4.5 against GPT-5.1-Codex-Max on backend tasks for the last 24 hours, and honestly, I can't decide on a clear winner. Usually when comparing models, the winner is pretty clear very quickly. Not this time. The testing continues.

@theo Bold take

I think Opus 4.5 might become my default coding model...

All I wanted was to talk to Claude Code from my phone. Ended up accidentally building my new daily terminal app: - tracks claude code + codex working states - notifies when waiting for approval - color coded by agent cli - all features available remotely via browser There's a more minimal traditional terminal style view, but I usually keep the sidebar open. It's like having a dashboard for my agent CLIs

@JotaDeRodriguez @localghost Yeah I literally said I didn’t think I could trust it

@localghost @buddyhadry Bro how do people in 2025 don't know you can't trust the model to tell you what it is? At most the only reason it can know is because it's in the hidden system prompt, but you can get any model to say they're any other that way...

Sad to see that AI labs are focusing less on voice. Gemini Voice uses 2 Flash Grok Voice uses 2.0 ChatGPT uses 4o

Yeah I did that and it told me Gemini 3, which I was suspicious of since I didn’t recall seeing anything about an update to their voice mode. I have to imagine at some point they’re all going to care more once one of them releases the first voice-to-vibe-code type solution. Seems like a low hanging fruit that to many will have a ‘magical’ feel. And easy to implement (I myself built a crude version of this for Claude Code).

@buddyhadry I tried this yesterday, if you just push it to find what specific model it uses without leading it, it'll eventually say 2.5 flash. they just kinda lie.

After seeing how good Nano Banana Pro is at creating images and infographics about complex topics, one of my first thoughts was how useful it would have been for so many chats I’ve had in ChatGPT. That alone is probably going to have me using Gemini a lot more for these kinds of conversations. What are some of your favorites you’ve had it create? One of my faves (big Robert Cialdini fan):

@AlexFinn I keep seeing people talk about this, when did the algo change and what’s the actual implementation difference??

Follower counts on X are dead This is AWESOME news for small accounts The new algorithm has rendered follower counts completely meaningless An account with 100 followers can get the same amount of engagement as an account with 1,000,000 followers What does this mean? This means quality of content is absolutely everything If you post horrible content, if you rage bait, if you post meaningless questions, that house of cards will eventually collapse The only thing that matters now is positive content that educates, entertains, or uplifts Small accounts, your time is here