Last night I dispatched 235 agents to review a codebase. The model wasn't Claude. The harness was though. Claude Code ran the whole thing. The model underneath was a budget one. Here is the thing nobody tells you about dynamic workflows. They are not a model feature. They are a harness feature. ————————————————————————————————— Claude Code has a mode called UltraCode, and UltraCode runs the dynamic workflows for you. Turn it on and it fans out tens or hundreds of agents by default, runs them in parallel, verifies each finding adversarially, then synthesises the survivors. That whole dance is orchestration Claude Code performs. The model just fills the seats. Which is also a cost bomb. Automatic fan-out to hundreds of agents on a premium model is a serious bill arriving by default. So I started renting the conductor and swapping the orchestra. The conductor is Claude Code, the harness: UltraCode, dynamic workflows, dispatch, the fan-out and verify loop. That stays fixed. The orchestra is the model underneath, and the model is swappable. Claude is one player you could seat. GLM and MiniMax are others. Point the same harness at a cheaper provider, GLM or MiniMax, and you keep the orchestration while the token bill collapses. Same UltraCode, same automatic dynamic workflows, a fraction of the bill. You utilise the same UltraCode, you just stop paying premium rates to run it. ————————————————————————————————— Now the catch, because there is one. You do not get all the benefits. A cheaper model reasons worse per agent. One GLM agent is not one Opus agent. What you are buying is the right to run fifty of them for the price of one, and to let structure do the work the raw model cannot. That is the trade. Depth per agent goes down. Breadth goes way up. And the workflow shape, fan-out then adversarial verify then synthesise, claws most of the quality back, because three cheap sceptics catch what one of them misses. ————————————————————————————————— The numbers from one run:

You guys keep liking and commenting on my confession posts, so here's another one. I wish at least one of these confessions wasn't real. They're all real (sad face). For the AI nerds in here (so, all of us): I think I'm addicted. It's worse than being hooked on a video game. A game at least has the decency to feel like a waste of time. This feels productive. Sometimes it actually is. At 1am, running six seven sessions at once? Not so much. ------------------------------------ What my nights have turned into ------------------------------------ Multiple Claude sessions going at once. When it got late and I knew I should be in bed, I'd flip every one of them with the /remote-control command so I could keep feeding the machine from my phone. Lying there in the dark. Waiting for the little dot to show up that means it's done thinking. Fire off the next instruction. Wait for the dot again. It's a slot machine. Drop the coins, pull the lever, watch for the dot. Like a freaking addict. One evening this week (I think it was Tuesday) I had three sessions all editing the same end-of-day file, and they kept overwriting each other's work. I'm sitting there getting genuinely angry. Then I caught myself cursing out a piece of software. Out loud. "You BLEEP, you broke it again." And I stopped. It's a machine, Ruben. Why are you getting triggered by a machine? And who's really doing the breaking? ------------------------------------ So I asked the machine why I can't quit the machine ------------------------------------ I did what any self-respecting addict does. I used the thing I'm addicted to, to figure out why I'm addicted to it. I'd read The Goal a while back (the Theory of Constraints novel everybody in operations swears by). I reopened it, had Claude walk me through the main concepts and tie them back to my business. And it clicked: I'm the bottleneck. My time, my attention. Not my team. Not my tools. Me. Then came the part that actually stung. The optimizing itself was the bottleneck. I'd been spending multiple two-hour sessions buffing an end-of-day routine whose entire job is to take fifteen minutes. The thing I kept "improving" stopped being my constraint years ago. I just couldn't put it down.

I turned one of my competition entries into a real business. That has been part of my goals with these comps—to see what I can drum up that’s been sitting in my head, waiting to escape into material reality. Like a true ADD entrepreneur, I have tried many other things while building my primary business. With every endeavor, I’ve either burnt out, or realized I didn’t have enough capital, or that I needed to stay focused on my original business. Well, I got my first lead ever (outside of the primary biz) for something I spun up here!!!! I will be mentioning ICM to them as “the final puzzle piece” since the core service I’m selling is marketing related, and I hope to upsell the client into the ICM later. I have the call on Friday. Even if they don’t buy, I’m so happy to get a lead. I have spent only $60 to test that business model and I got a lead. this is the best test I’ve conducted since my original business (luxury house cleaning, if you’re wondering) and I’m pumped. I’m looking for a 100/1 LTV/CAC ratio. I know it’s out there. Could this be it??? Let’s find out!

🎯 WE READ ALL OF YOUR RESPONSES. HERE'S WHAT'S NEXT (AND A BIG ASK) A couple weeks ago we asked you what you wanted Clief Notes to become. You showed up. You wrote real, thoughtful answers, some of you wrote essays. We read every single one. Twice. So before anything else: thank you. This post is us answering you. 💬 YOU TOLD US. HERE'S WHAT WE'RE BUILDING. You said you want a clearer path from learning ICM to actually getting paid for it. → It's coming. A real learning-to-earning track, plus a talent platform we're building to connect you with people who want to hire what you can do. Heads up: the talent platform will be Premium and VIP only, one more reason to lock in below. You said competitions without feedback don't help you grow. → Fixed. Going forward, every single entrant gets tailored feedback on their build, not just the winner. We're moving to two competitions a month so we can do it right. And both monthly winners get a 30-minute call with me. You said the best builds get buried in the feed. → We're building a tagged library so you can actually find "ICM setups for solo operators" or "small team, non-coder" instead of scrolling for an hour. You said you couldn't map all the pieces, Skool, Discord, the Lyceum, ICM, what each tier unlocks. → A single orientation page is on the way. One place that breaks all of it down so nobody's lost. You said you want to connect with each other. → We hear you. Meetups, pairing, and a recorded "After Tea" hangout are on the table. South Florida alone has over 1,000 of you. Let's use that. You said we've felt spread thin. → The most honest one. You're right. We're bringing on real help to run the day-to-day so the community gets consistency, and so Matt and I can keep building the things above instead of dropping balls. You called it, and we're fixing it. 📍 QUICK NOTE ON ICM: a few of you asked for "ICM with Copilot" or "ICM with n8n, Sheets, GoHighLevel." Good news, ICM already works with any model and feeds straight into the tools you're using. It's the structure underneath all of them, not a separate thing you have to relearn per tool. If that's not landing for you, that's on us to make clearer, and we will.

I rebuilt my site today with the focus on folders over agents. It's geared toward providing the workflows for businesses. Not sure if this is allowed here, but I'm sharing anyway. Take it down if it breaks any rules. So many great minds here, so I thought it was worth risking criticism or advice on how to improve. Perhaps it will give you ideas for your own site. Check it out: optimarketai.com