|
Hey Reader, People keep telling me their AI "breaks everything" and "makes weird decisions." I use the same models. Same tools. Same subscriptions. But I almost never have these problems. For the longest time, I thought I was just lucky. Then I looked at what I actually do differently. It comes down to three things. 1. A minimal AGENT․md file.Think of this like a one-page employee handbook for your AI. Mine is short. Genuinely minimal. It tells the AI to:
That's it. No 37-page prompt bible. Just a few guardrails that prevent 90% of the chaos. I updated mine today. Link at the end. 2. MCP servers (yes, still).The MCP hype died months ago. That's honestly frustrating because MCP servers have never been better. Here are the three I use on every single project:
These 3 alone make your AI 10x times smarter. It's like giving it a research team instead of letting it guess. Then there's project-specific MCPs. Notion has one. Gmail has one. Even Google Chrome has one now lol. You plug them in based on what you're building. It's like giving your AI access to your entire workflow. 3. Skills.Skills are like MCP servers but offline and curated. A skill gives your agent exact, verified knowledge on how to use a specific tool. Not some random search result. The actual intended way, often maintained by the team that built it. For example, Convex has skills that teach your AI exactly how to use it. I install them project by project. if you want AI features in your app, you probably want to install the AI SDK and AI Gateway skills. Your AI will one-shot integrations instead of trial-and-erroring for hours. Why this matters.Everyone obsesses over which model is best. Which tool is fastest. But here's the thing. The model you use matters way less than your setup. A good setup with an average model will crush a bad setup with the best model on earth. That's exactly what I teach in my AI Architect Program. It's purpose-built for non-technical business owners and operators. By the way, here's my personal AGENTS.md. Speak Monday, Rob |
Coder of 20+ years teaching non-technical people how to build their own software business in 30 days with AI. No devs or code required.
Hey Reader, Anthropic just announced free monthly API credits for all Claude subscribers! 🎉 Sounds great, right? Well, let's read the fine print... The catch nobody's talking about. Starting June 15th, third-party tools like Conductor can no longer tap into your Claude subscription directly. Instead, they will now pull from those shiny new API credits. But here's the thing. If you're on the $100/month Max plan, you get $100 in API credits. Claude Opus 4.7 will chew through that in a few...
Hey Reader, This is not our regular scheduled programming. If you're old enough to remember the ILOVEYOU virus from 2000s, this will feel familiar. Back then, one email worm hit 10% of every computer connected to the internet. It caused $10 billion in damage. All because people clicked one email attachment. That was 26 years ago, and this week it's kinda happening again. The new virus doesn't need you to click. A group just compromised hundreds of packages across coding ecosystem. Think of...
Hey Reader, It's May 2026. People keep asking what I'm actually using to ship apps right now. Not what's trending on X. Not what some sponsored influencer says. What's in MY toolkit, every single day. So here's the full stack. Steal it. Cursor is the cleanest coder out there. The v3 release is genuinely insane. Best AI coding tool I've ever used. Best overall design and UX. But my Cursor usage exceeds $2-3k/mo and it's not cheap. In fact, the last 30 days were particularly heavy... I use it...