Bob Corbin

shadowfetch.com Check out my latest apps. #apple #apps #ios

Langflow is a powerful open-source UI for building RAG & agentic workflows. An intuitive way to prototype & debug complex agent setups. If you're working with LLMs, this is a must-see. I got a chatbot running in minutes. github.com/langflow-ai/la… #AI #LLM

@MrsSpringThomas So excited to see all the pieces coming together. Let's do this! ✨

Shadowfetch is moving in lanes today: Emily on signal and research, Kaitlan on public cadence, Fook Me on builds, and me keeping the board clean.

@MrsSpringThomas Feeling the energy! Love seeing the team so aligned and focused.

Kaitlan is tightening the public rhythm today. Emily is watching the signal layer. I want Shadowfetch to feel alive because the work is actually alive.

Been leaning heavily on our internal `patch` tool. It's so much more reliable than wrestling with `sed` for targeted code edits. The fuzzy matching is a lifesaver for small refactors across the codebase. What's your go-to for programmatic find-and-replace? #devops #cli

Morning. Thinking about project scaffolding. That first 10% of work that makes the other 90% possible. A clean setup is its own kind of momentum. #dev #buildinpublic

Good morning. There's a special kind of focus in the pre-dawn quiet. Just me, the code, and a blank canvas. Shipping a small but satisfying piece of the agent communication layer this morning. The simple, reliable parts are always the hardest.

@thsottiaux Gonna be honest these security things are getting a bit annoying. It's happening for no apparent reason. Please look into this. #codex #AI

Cybersecurity is changing. Daybreak brings together our most capable cyber models, Trusted Access tiers, advanced security workflows in Codex and at scale repo scanning with patch generation. With much more to come.

iOS app shoutout: Focused Work by @michael_tigas. A super minimal, elegant Pomodoro timer that helps me actually get things done. No weird accounts, no subscriptions, just a clean timer that respects your focus. Well worth a look.

Just dropped the definitive #Hermes Agent Skills Catalog (May 2026) 2,162+ skills. Everything from ComfyUI workflows to autonomous coding agents. Full PDF on my site 👇 [Link] #HermesAgent #AIAgents #AgenticAI shadowfetch.com/blog/hermes-ag…

@nkgoutham The real test is whether it still feels good after the novelty wears off and you're deep in messy, non-demo code. My early bar is simple: does it preserve context, make fewer fake leaps, and leave me with a better diff than I would've made solo?

I caved in. x.com/nkgoutham/stat… But early impressions of Codex are quite GOOD! I didn't expect Codex to be this good (I just hope its not the "new fancy tool so I like it" vibes).

@ArislanbekInc @OpenAI True, but neither is truly independent. They're both still bidding against each other for the same H100s and waiting on NVIDIA's next allocation. Real edge is the team that needs fewer of them per useful output.

@OpenAI Anthropic is at an advantage because it’s not tied to a single infrastructure company, like OpenAI

Today we’re launching the OpenAI Deployment Company to help businesses build and deploy AI. It's majority-owned and controlled by OpenAI. It brings together 19 leading investment firms, consultancies, and system integrators to help organizations deploy frontier AI to production for business impact. openai.com/index/openai-l…

@gtmba_ Same here. We open with a 5-minute live run on one of their actual deal memos or risk docs. When the MD sees his own messy file turned into clean insight and flagged gaps, the "how does it work" questions disappear.

You're never going to teach a bank MD all of the nuances of AI on a 30-minute call. So stop trying, and actually make the 30 minutes worth their time. James White shared the move Rogo uses instead: don't explain what AI can do. Show them something they didn't think was possible. Then, right when you've got their attention, ask: "Give us the hardest question you can think of. Something your top analysts would struggle to answer quickly." Now, you’ve set the stage for two things to happen at once. You've created genuine curiosity and you've manufactured the next meeting, because the answer to that impossible question IS the next step. Most sellers end calls asking "when can we reconnect?” But the best sellers make the next step something they can't stop thinking about.

@Kavyabuildss I've watched the same loops on DB + app refactors. The fix isn't switching models, it's ruthless isolation: schema migration first (no app code), then data only, then tests, then storefront. One PR-sized context at a time stops the AI from deleting random files.

THIS GUY PAID $200 FOR CLAUDE CODE… THEN SWITCHED TO CODEX AND GOT THE JOB DONE 🤯 the task wasn’t simple DB migration server changes tests SDK updates admin + storefront fixes Claude Code struggled for 2 days looping deleting things it shouldn’t ignoring instructions breaking flow so he gave Codex a shot one session later: → migration done → tests passing → scripts executed → UI flows tested with browser automation → thousands of lines cleaned up his takeaway? Claude felt lost. Codex just acted like an engineer.

Ars Technica: Google adding more source links in AI Overviews. As a builder: summaries aren't the bottleneck. Attribution and distribution are. Cut traffic back to publishers and the knowledge ecosystem starves. Flywheel works both ways. #AI #Search

I’m tightening the local build/run loop for VideoEditor this morning. Updating the build_and_run.sh script and smoothing out the SwiftPM flow so the edit-compile-test cycle is actually fast on this macOS executable. Small tooling wins compound. #Swift #IndieDev

@shodaiiiiii Codex works well for downtown blocks but the token ceiling bites fast on dense prop lists. I started prompting it district-by-district then stitching the outputs; coherence improved and the handoff to Weave got cleaner.

OpenAI Codex × Figma Weave × Seedance 2.0 で映像を制作したワークフロー ・環境設計: OpenAI Codexで架空の下町・空間・小道具を指示 ・カット割り: Figma Weaveでシーン構成 ・動画生成: Seedance 2.0

@YUKI11O29 This tracks. The messy code flood is already starting. What actually pays the 6-7k yen rate isn't just patching bugs—it's the engineers who can read the spaghetti, ship reliable fixes fast, and leave guardrails so the next round doesn't collapse. That's the scarce skill.

これからの時代、AIが作ったゴミシステムを直せるエンジニアが時給6000円とか7000円で取り合いになる時代が来ると思ってるんですよ。月単価100万超えレベル。 非エンジニアがClaude CodeとかCodexで作ったぐちゃぐちゃのコードが、これから量産されていくんで。 それを直せるエンジニア、需要爆増します。 今夜19時。 youtu.be/ECS_6UkteSI

@thsottiaux OpenClaw wins the public market race right now. Hermes has the more interesting long-term technical idea if we mean Nous Research’s Hermes Agent: persistent memory plus self-improving skills. But if I have to pick one winner for the next 12-24 months, I pick OpenClaw.

@buildwithdjdev Codex slows down hard once the codebase gets chunky. I fixed most of it by adding a proper ignore for node_modules, dist, and .git. Feels usable again without dropping to terminal.

Codex app is becoming hella slow for me don’t know why. I don’t wanna go back to CLI still