10 days in, and this one might be my favorite so far. When I first started with OpenClaw, I had no idea what tools my agent actually had. I was installing random stuff and hoping it would work, and it took me way too long to realize the whole system clicks once you separate two things in your head. Tools and skills are not the same thing. Tools are the raw capabilities, like exec, read, write, and web search. They're the actual hands. Skills are just the instruction manuals that teach your agent when and how to use those tools for a specific job. So a Gmail skill, for example, doesn't give your agent any new powers, it just teaches it how to use web fetch and exec to talk to the Gmail API. Once that clicked, everything else got easy. I also get into Claw Hub in this one, which is honestly kind of wild. 3000+ community skills, one command to install, and you can bolt new capabilities onto your agent in seconds. But the part I really wanted to dig into, and the reason I almost split this into two videos, is the security side. Skills are the single biggest attack surface in any OpenClaw setup, and a bad skill runs with the same permissions as your entire system. I walk through what to actually look for in the source code before you install anything, plus the 4-step process I run every single time. If you're installing skills without reading the code first, watch this one before you install one more thing.

Your agent doesn't know what it's NOT allowed to do... 8 days in and your agent is actually doing things on its own. But here's the problem - it's unsupervised. No priorities, no delegation, no guardrails. Day 9 is where that changes. Most people think the soul.md file is where all the important config lives. It's not. agents.md is what your agent reads first, every session, including every background task it spawns. Get this file wrong and everything downstream inherits that mess. The part I find most underrated is subagents. You can literally have your agent spin off a worker mid-conversation, keep going without waiting, and get a summary dropped back in when the job's done. And the security model is cleaner than most people expect. Three independent layers: where code runs, which tools are available, and who can override. They don't overlap, they stack. Day 10 we're getting into tools and skills, so hat your agent can actually do beyond the defaults. 🎥 Watch Day 9 below.

Just shipped two packages that make working with Vapi a lot faster. mcp-vapi connects your Vapi account directly to Claude Code, Cursor, or any MCP-compatible AI client. List assistants, fire calls, pull transcripts, without ever opening the dashboard. One install command and it just works. n8n-nodes-vapi is a proper community node for n8n. If you're building outbound call workflows, you can now trigger Vapi calls, manage assistants, and pull transcripts as native n8n nodes, no custom HTTP requests needed. Both are free, open source, MIT licensed. Built these because I was copy-pasting IDs and writing the same boilerplate over and over on client campaigns. Figured other people are doing the same thing. If you're building AI voice agents or running outbound campaigns with Vapi, these will save you some time. https://github.com/adhirajhangal

A lot of people are reacting to the Anthropic change by ripping Opus out of their stack completely, and I think that’s the wrong move. Yeah, the old subscription setup is gone, but that does not mean OpenClaw is dead. It just means the architecture has to change. I just posted a video breaking down what actually happened, why the old setup stopped working, and the exact stack I’d use right now if I were rebuilding from scratch. In the video, I walk through the new setup, how I’d split the work across models, and the mistake I think most people are about to make. If you’re building agents for yourself or for clients, watch this before you touch your stack.

Made this one because a member in my Skool community asked for it. The only n8n tutorial you'll actually need... We build three automations from scratch in one video, starting simple and ending with an AI research assistant that pulls live data, analyzes it, and logs everything to a spreadsheet on its own. Build 1 is a daily weather email that fires every morning at 6am. Build 2 is a client intake form that reads someone's revenue, routes them down the right path, sends a personalized email, and logs them to Google Sheets. Build 3 is the real one. Live crypto data, filtered for the big movers, sent to GPT for analysis, saved to a sheet, running every morning without you touching it. By the end you'll see the pattern behind all of it, and once you see the pattern, you can build pretty much anything inside n8n. The workflow JSONs for all three builds are in the classroom. Import them, swap your credentials, done. Go watch it and tell me in the comments what you'd actually want to automate first.