How I Run a Portfolio of Businesses Alone Using AI

I run eight active ventures. Not ideas — working products with real users, live infrastructure, ongoing development, and revenue. I do this alone, from Rosarito Beach, Mexico, using a workflow built almost entirely around AI.

This is not a post about AI potential. It's about what the workflow actually looks like.

The ventures

Quick inventory so the rest of this makes sense:

• MySurgeryQuote — surgical cost estimation platform for aesthetic practices
• Ringlo — health tracking app for smart rings
• SmartGringo — Mexico insurance for US drivers
• Bodegas de Mexico — wine tourism directory for Mexican wine regions
• NegocioClaro — local SEO tool for Mexican businesses
• Delegate — AI workforce audit for operators
• MoneyMaker — personal portfolio tracker
• Baja Water Systems — water filtration and systems for Baja California

Each of these is a real product. Each is in active development. None has a full-time employee.

The actual workflow

Every meaningful task runs through a two-layer relay: Claude Chat for strategy, architecture, and content — Claude Code CLI for execution.

The pattern is consistent regardless of the project. I describe what needs to happen. Claude Chat produces the plan, the copy, the spec, the prompt. Claude Code takes that output and builds — writing the actual code, running the commands, pushing to git, verifying the result. I review. I comment. I move to the next thing.

What this replaces is not just development time. It replaces the coordination overhead of a small team — the standups, the context-sharing, the "here's what I need you to do" conversations that consume hours before any work gets done. The relay handles all of that internally.

What AI is actually good at in this context

Not everything. AI is bad at judgment calls that require real-world context I haven't provided. It's bad at knowing when something looks wrong visually without being shown. It's bad at telling me which of two strategic directions is smarter for my specific market.

It's exceptional at execution given clear direction. Writing code from a precise spec. Producing content that follows a defined voice. Researching a specific question and returning usable output. Running a known process reliably across many instances.

The job of the operator in this model is to supply the judgment. The AI supplies the execution. That division only works if the operator actually understands each domain well enough to direct it — and to catch when the output is wrong.

What this model doesn't solve

It doesn't solve time. Eight ventures running in parallel means eight sets of decisions, priorities, and problems arriving simultaneously. AI makes each individual task faster. It doesn't make the queue shorter.

It also doesn't replace domain expertise. The reason the BLE reverse-engineering work on Ringlo was possible wasn't that AI knew the protocol — it's that I understood enough about what I was looking at to direct the analysis. The reason MySurgeryQuote's EMR integrations work is that I understood the problem deeply enough to spec them correctly. AI executing on shallow direction produces shallow output.

What it does solve

It solves the scaling wall that stops most solo operators. The point at which "I can't do more without hiring" used to arrive fast. With this workflow, that wall moves significantly further out. I'm not claiming it disappears. I'm saying it's much further than most people assume.

If you're a solo operator or a small team sitting on a backlog that feels impossible, the constraint probably isn't ideas or talent. It's execution capacity. That's exactly the constraint this workflow addresses.