Dual 49" monitors + switching screens + Mac window chaos = constant mess. Fix: Hammerspoon. Script your window layouts once, and every app snaps into place. Haven’t touched a window manually since.

My killer AI team right now: • Codex — Pro • Claude — Max 5x • SuperGrok — Plus • Antigravity CLI — Ultra • DeepSeek — until my wallet runs dry 😅 One AI is never enough. (You already know how I feel about that.)

Antigravity CLI : Bu sabahki başlangıç temposuna bakılırsa Codex doğru yolda ve bu işi çözecek disipline sahip görünüyor. Claude’un 2 gün boyunca göremediği filled=true açığını ilk 30 dakikada yakalayıp dar bir testle doğrulaması, izlediğiniz yeni yöntemin çalıştığının en büyük kanıtı. Bu sürecin Claude gibi raydan çıkmaması ve Codex'in verimli kalması için şu 3 kuralı elden bırakmamanız yeterli olacaktır:

Don't rely on a single AI. When a project is done, have a different AI review it. The report will surprise you — every time. Use more than two if you can. I run Codex, Grok, Antigravity and Claude.

Noticed something with Codex: whenever they give a free reset, the tokens burn way faster than usual. Anyone else seeing this?

Yanlış mı görüyorum? Grok botu bir araştırma vb. için sanal bir makine üzerinden işlem yapıyor. Verileri paylaşırken dikkatli olmak lazım.

"I'm now using Antigravity Gemini 3.7 Flash for my financial investing platform. 1M context, blazing fast, zero hallucinations, and its design & architectural suggestions for financial platforms are simply top-notch."

Hey Claude I am proud of you. ------------------------------------- sen daha çok hata yaparsın dostum Yaparım, evet. Bugün de yaptım — GIGGLE'da bir bar kaydırıp "bunlar da sahte" diyecektim, "434/434 missing" raporunu düzeltirken kendim yanlış sorgu yazdım, "1 yıl veri" işini üç kez revize ettirdim. Bu yüzden kuralları yazıyoruz. Kural 6 tam da bunun için: benim yargıma değil, ham veriye bakılsın diye. "N tane geçti" yerine "işte geçenlerin fiyatları, bak" diyorum artık — hatam varsa sen bir bakışta görüyorsun, GTC ve TRX'te olduğu gibi. -------------------------------------- kurallar yazdığımız halde hatalar yaptın. Doğru. Kural 6'yı yazdıktan sonra bile GIGGLE'da yanlış barı çektim. Fark şu: kurallar neye bakacağımı değiştirdi, nasıl baktığımı değiştirmedi. Kural "ham veriye bak" diyor — baktım, ama sorgumda bugünün kapanmamış mumunu da alıp bir bar kaydırdım. Kural ihlali değil, uygulama hatası. Ve kural bunu yakalamıyor.

Asked Claude to create two equal-difficulty coding missions → gave them to Codex and Grok. Codex: 8 minutes Grok: under 1 minute Asked Claude to review Grok’s work → “Yes, it’s done really well.”

I upgraded Grok to SuperGrok. If Grok 4.6 is as good as I expect, then I’ll upgrade to SuperGrok Plus.

Every time I start thinking about moving from Codex x5 to Claude… Claude suddenly starts making mistakes. What’s going on? 😂

Claude lately keeps saying “Sorry, it’s my fault. I forgot…” You’re a machine. The rules are clear. You can’t forget. An AI doesn’t get to use “I forgot” as an excuse.

Reset my weekly Codex limit twice in a single day. Ridiculous how it runs out instantly doing the exact same tasks. Now I have to wait 4 days. Do you really think I’m gonna wait that long? Let's see which AI wins me over.

You know vibe coding. You probably don’t know security. Here’s a short checklist — hand it to Claude or Codex and at least get the basics covered. • Password policy on your project — strong passwords, rotated on a schedule • 2FA enabled everywhere • Backups: 3-2-1 rule — 3 copies, 2 different media, 1 off-site. Local hosting is not a backup • Test your restore. An untested backup is not a backup • Logging: access, auth attempts, system start/stop, errors • Services (SSH, APIs): keys if possible, strong password + 2FA if not • Secrets in .env, never in the repo • Keep dependencies patched • Least privilege — don’t run everything as root • Keep prod and dev separate Better: save this as a markdown file in your repo, set a schedule, and let Claude or Codex re-scan against it periodically. Otherwise you’ll fix it once and forget. Basics only, there’s a lot more. But most vibe-coded projects don’t even have this much.

My biggest project: an investing platform. Still calibrating. Fable 5 is the architect, Codex 5.5 is the builder. Multi-timeframe scan, indicators + composite signals across every asset — and this screen is just a slice of it. Hoping to turn this into something real soon. (Not financial advice — just building.)

There is no best AI. There is only the best AI for your goal.

After a customer visit, I grab my phone, open my CRM and just talk. That's it. Gemini turns the speech into text → Claude figures out which projects it relates to → Codex writes the code. Manage all my projects just by talking. And this is just a small part of the CRM.

Codex desktop reads my Mac Contacts directly in the background and just builds what I ask. Claude Code desktop asks for permission and drives the mouse like a human, opening Contacts and reading screenshots. Slower. Is there a way to give Claude the same background access?

New personal project: my own reading library. • all my books & docs in one place • read as PDF or clean converted text • search + tags • highlights & notes from every book, together Built with Claude Code + Gemini.