I have the same problem every builder in a fast-moving niche has: by the time I notice something is trending, it's already stale. Someone else already made the video. So I wrote out exactly what I needed. A tool that watches competitor channels, scans forums and news for what's breaking, and surfaces it so there's always a real, same-day angle for a video instead of guessing. Then I handed that brief to the AI and went to bed. By the time I was back at my desk, it was built. Not a mockup, a working capability inside Yagi Studio: it watches the sources you point it at, checks community activity and news, and scores what's actually rising versus what's just noise. I didn't build this just for me. It's configurable end to end: you add whatever sources make sense for your niche, not just mine, and the AI synthesizes everything it pulls back into what's actually worth paying attention to. It's called Trend Radar and it's part of the same platform I'm building everything else on. The part that still gets me isn't the tool itself, it's what it represents. I described a real, specific problem, and instead of me writing the code line by line, I described the outcome and it existed by morning. That's the whole shift I keep trying to get across: you don't need to be the one typing every function anymore. You need to be the one who knows exactly what problem is worth solving. This is staying inside Yagi Studio for now while I use it myself and see where it breaks. When it's ready, it's built for the whole community to point at their own niches, not just mine. What's a problem you deal with every single week that you've never actually described out loud as something worth building?

Every AI company shipping "agents" right now has the same structural problem, and none of them can fix it: they only trust their own model. I've been building something different for one of my Yagi projects. Instead of one AI wearing different hats, you build a panel of agents. Each one gets its own background and role, like a skeptic, an operator, a customer advocate, whatever the decision calls for, and each one runs on whichever model actually fits that role best. Right now that's Claude, GPT, Grok, and Gemini integrated, with more models going in as I build this out. A panel can be one agent or ten. You decide who's in the room and what job each of them has. Here's the moment that made me sure this was worth building. I was comparing it to Grok Bot, the "always-on AI teammate" xAI just shipped. Grok Bot looks impressive, a chief-of-staff agent delegating to specialist agents that run on their own cloud computers. But every one of those agents is Grok. Same training lineage, same blind spots, same refusals, same way of being confidently wrong. A panel of Grok agents, each playing a different role, still isn't a second opinion. It's one mind talking to itself in different voices. Give the same roles to agents running on four different companies' models and something different happens: when three of them independently flag the same weak spot in your offer, that's actual signal, because they didn't come from the same place. When they split, that's useful too, it means the answer isn't as obvious as you thought. And here's the part no single-vendor company can copy even if they wanted to: xAI will never put Claude in the room. OpenAI will never route a question to Gemini. A neutral orchestrator that treats every frontier model as a hireable agent is a position only an outsider can hold. That's not a feature I built, it's a structural gap only independents like us can fill. It got built out same day, both halves: the engine that streams results back from every agent in real time, and the interface where you actually convene the panel and run a decision by them. Six database tables, real streaming, real disagreement.

Every day I build something, I'm supposed to turn it into posts for five different places. Do that by hand and it becomes a second job. Worse, it's a job that gets bigger as more people follow you, which breaks the one rule I run everything by: no part of the work is allowed to scale with the size of the audience. The day I need more hours just because more people are watching, I've built an agency, not a portfolio. So I built a skill to do the repackaging instead of me. Here's the part most people miss. I didn't point it at my git history. Git only shows what changed in the code. I pointed it at my Claude Code and Codex session logs, because those hold *why* I changed it. The option I threw out. The wall I hit for an hour. The moment I said "wait, that won't scale." That's the stuff people actually want to read, and a diff never captures it. The skill reads a full day of those logs (about 50,000 characters in one pass), pulls out the few things that genuinely deserve a post, and drafts them in my voice. One day in, a handful of drafts out. I set it up so it can be run any time you want, or you can have the Claude desktop app run it on a cadence so it's just done for you. One thing it will never do: post for me. It writes the draft, I read it, I edit it, I hit publish. This post you're reading came out of it and got my hands on it before it landed here. How to get it right now. It's open source, MIT, part of a plugin I'm calling `arc`. In Claude Code: /plugin marketplace add mrrichsylvester/arc /plugin install arc Then run `/arc:daily-content yesterday`. First run asks you a few questions (your voice, your platforms, where to save drafts) and writes its own config, so the same skill becomes yours without any of my paths baked in. And `arc` is going to grow. `daily-content` is the first skill in it, not the last. This is the toolkit I use to build in public, and I'm going to keep adding to it in the open. So if you grab it now, you get everything I add later for free.