15 posts
Four of Google's top AI and systems engineers — Jeff Dean, Sanjay Ghemawat, Oriol Vinyals, and Quoc Le — just left together to start Discovery Loop. Alphabet's stock dropped 4% the same day. Here's why that reaction was correct.
ZAI quietly added a feature that lets you schedule tasks to run when their servers are idle — effectively free or near-free compute. It's a brilliant move, and every AI platform should copy it.
After two days of heavy coding and switching between the new models, here's my current setup for choosing the right model without burning your Codex limits. Luna for everyday work. Terra when you need to explore. Sol when it actually matters.
At $0.053 per task, GPT-5.6 Luna-Max outperforms models that cost 6–8x more. Opus 5. Sonnet 5. Gemini 3.6 Flash. All beaten. All out-priced. Here's why this changes what "good enough" means for your daily AI stack.
The new Luna pricing + Max reasoning tier made it an obvious choice. Terra is now nearly pointless. Here's my reasoning, the numbers, and why this is genuinely a great time to be an AI user.
OpenAI quietly dropped GPT-5.6 Luna by 80% and GPT-5.6 Terra by 20% today. Luna is now a fraction of what it cost a week ago. This isn't a sale — it's a structural shift in how AI pricing works. Here's what actually changed and why it matters for how you build.
OpenAI temporarily removed Codex's 5-hour usage limit on July 12. It was never a gift — it was incident response. And now it's probably coming back. Here's the full timeline, what it means for heavy users, and why the "temporary" framing is the most important signal.
Eighteen months ago open source AI wasn't a real alternative to GPT-4. Today it is. The performance gap is almost gone. Here's the honest breakdown of what's changed, what hasn't, and what it means for how you build.
The performance gap between open source and proprietary AI models has shrunk to almost nothing. Here's what's actually worth using, what it costs to run, and where this is heading.
I was paying $100/month for Claude Pro. Now I split the same budget across four specialised tools and get more done — here's exactly what I use and why.
Three months of AI automations gave me back my mornings. It also gave me three months of content that sounded like everyone else's. Here's the difference between getting time back and just getting faster at producing noise.
OpenClaw, Hermes Agent, and CrewAI — what each one actually does well, where each one breaks, and how to pick the right one for what you're building.
Anthropic quietly extended Claude Fable 5 access to July 19, and kept Claude Code's rate limits 50% higher. I break down what that actually means for your daily workflow — and what happens after the 19th.
Anthropic's Fable just turned the 'Pandora's box' argument from theory into something I'm dealing with every day. Here's what it means for the people shipping with AI right now.
Claude tracks you on two clocks at once — a five-hour rolling session and a weekly cap per model family. Once you see them both, your usage stops feeling like a mystery.