The Two-Claude Workflow That Runs Nine Ventures

Last week I was featured in The Rundown AI's Community Workflows section. The summary they published was accurate: I'm a solo founder running nine active ventures from Rosarito Beach, Mexico, using a two-Claude relay as my entire engineering and content operation.

A few people asked me to explain the relay in more detail. Here it is.

The setup

Two tools. That's it.

Claude in the browser handles everything that requires judgment — strategy, architecture decisions, content, specs, diagnoses, prompts for the next step. This is where I think out loud, describe what I'm trying to build, and get back a plan.

Claude Code in the terminal handles everything that requires execution — writing code, running commands, pushing to git, verifying the output. It doesn't decide what to build. It builds what I've already decided.

I'm the relay between them. I review what Claude Chat produces, decide what's right, and paste the relevant prompt or spec into Claude Code. Claude Code runs it, reports back, and I return to the browser with the result.

That's the whole system.

Why two instead of one

Claude in the browser is optimized for thinking. Claude Code is optimized for doing. They're different tools built for different parts of the workflow, and treating them as one collapses the separation that makes the system work.

When I try to do architecture decisions in the terminal, I get code before I've thought clearly about what I want. When I try to execute in the browser, nothing ships. Keeping the two roles separate keeps the work moving.

What it looks like in practice

Last month Ringlo — my smart ring health tracking app — got rejected from the App Store. The rejection cited a guideline violation in how the subscription was disclosed.

I pasted the rejection email into Claude in the browser. Within a minute I had a diagnosis: the paywall screen wasn't showing the subscription terms prominently enough before the purchase confirmation. I had a fix plan: three specific UI changes, the exact copy Apple requires, and a note about which screen needed to be updated.

I relayed that to Claude Code. It found the relevant component, made the changes, rebuilt the binary, and flagged it for resubmission. I reviewed the changes, confirmed they matched what Apple needed, and resubmitted the same afternoon.

A normal founder without this workflow would have spent a day reading Apple's guidelines, another day coordinating with a developer, and submitted a fix by the end of the week at best. The whole cycle took a few hours.

The nine ventures this runs

For reference — the current portfolio:

• MySurgeryQuote — surgical cost estimation for aesthetic practices
• Ringlo — health tracking app for smart rings
• SmartGringo — Mexico insurance for US drivers
• Bodegas de Mexico — wine tourism directory
• NegocioClaro — local SEO tool for Mexican businesses
• Delegate — AI workforce audit
• MoneyMaker — portfolio tracker
• Helm — field service CRM
• Perfect Photo — HIPAA-compliant medical photography

Each is in active development. None has a full-time employee.

What the relay doesn't fix

Time. The queue of decisions across nine products doesn't get shorter because I can execute faster. If anything, the bottleneck has shifted — I now have more ideas than I have hours to evaluate them.

It also doesn't fix bad direction. If I give Claude Code a poorly thought-out spec, I get a well-executed version of the wrong thing. The judgment still has to come from me. The relay amplifies good thinking and amplifies bad thinking equally.

What it does fix is the execution wall — the point where most solo operators run out of capacity to ship. That wall is much further out than most people assume.

---

Scott Tobin is the founder of Tobination. He builds AI-powered ventures from Rosarito Beach, Mexico.