
AI is a cheat code, and the skill file is one of the specific buttons that makes it one. A skill is just a markdown file (plain, readable text) of written instructions you hand your AI coding agent so it can carry out a task it couldn't before. Real example: my team couldn't get Claude Code to analyze video, so I wrote a skill telling it to split any .mp4 into frames to look at and audio to transcribe. Went from "can't touch video" to "no problem." Think Mario power-ups: each skill is a new ability. It's how Joe 3.5 turns my voice memos into published blogs, and because skills can teach an AI to use an API, it's how AI stops just talking and starts interfacing with your real systems.
Read More ›
I said the AI receptionist companies lost their moat when OpenAI made the hard part easy. Now it's the vibe-coding builders' turn: Lovable, Replit, Cursor, Base44. Their whole value was stitching GitHub, Vercel, and Supabase together and handing you a working app. That convenience was the product. But Claude Code and Codex now do that same assembly for you, and better, so the markup people paid for is exactly what they stop paying for. They're not dead, they've got distribution. But here's the lesson: if a fast-moving AI can absorb your whole product as a feature in three months, you don't have a moat. You have a demo. Build the platform AI comes to, not the wrapper it replaces.
Read More ›
Let me be clear up front: I'm not an AI hater, I think it may be the most powerful technology we've ever invented. This is a different complaint. A lot of the people selling it to you are outright charlatans. At its core an LLM is fancy autocomplete, a prediction machine that can't say "I don't know," which is exactly why it's confidently wrong. Meanwhile the funnel gurus posting "my 300-agent swarm runs my business" have never shipped a real product in their lives. I even watched a nine-figure entrepreneur I respect push a tool that would get your machine hacked, then bury it when it stopped trending. There's a reason 94% of AI initiatives fail, and it isn't the tech. It's the snake oil.
Read More ›
Bold claim: the AI receptionist companies, Vapi, Retell, Synthflow, just lost their moat overnight. And Vapi's no joke, reportedly valued near $500M and working with Amazon's Ring. So let me prove it. An AI receptionist isn't one brain, it's five layers duct-taped together: telephony, speech-to-text, the LLM brain with its tools, text-to-speech, and telephony back out. The moat was never the parts, it was making the seams play nice, and that was brutally hard. Then OpenAI's Realtime API collapsed the middle three layers into one speech-to-speech model. I built my own copy in a day. When your moat is "this is hard to build," and AI makes it easy, that's a problem.
Read More ›