The Fractional CTO's Dilemma: When You're Both the Brain and the Hands

I read something this week that stopped me cold. Etienne de Bruin, over at CTO Sub, laid out two extinction paths for CTOs: Token Monkey and Vibe Coder.

The Token Monkey knows every AI subscription, every pricing tier, every prompt trick — but can't explain what the model is actually doing. The Vibe Coder ships fast, demo looks great, production falls apart.

He's right. And he's talking about CTOs with teams. I want to talk about what happens when there's no team. When you ARE the CTO, the lead engineer, the QA, and the one writing the LinkedIn post about it at 10pm.

The One-Person Squeeze

Running Opteia, I live in both traps simultaneously. Tuesday morning I'm deep in an ABI agent's decision tree, fixing a routing bug that only appears when a prospect replies with a question instead of a yes/no. Wednesday afternoon I'm evaluating whether Claude 4 or GPT-5 gives us better extraction quality on YouTube transcripts.

That's Token Monkey and Vibe Coder in the same week. Sometimes the same day.

Etienne drew a parallel to the 1970s, when programmers split into two camps: those who understood the machine and those who only knew the language. The machine-level programmers built operating systems. The language-level programmers built... well, a lot of software that needed rewriting.

The AI era is creating the same split, but faster, and with higher stakes. Because now the "language" isn't COBOL — it's natural language. Everyone can "code." The question is whether you understand what's happening underneath.

What This Looks Like in Practice

Last month, one of our agents started hallucinating meeting times. Not obviously wrong — just 15 minutes off, consistently. A Vibe Coder would adjust the prompt and move on. A Token Monkey would switch models and call it progress.

What actually fixed it was understanding that the agent's timezone context wasn't being passed through the handoff between the scheduling module and the confirmation module. Two lines of code. But you had to know the architecture to find it.

That's substrate thinking. And in a compressed team — where you're the only one looking — there's no one to catch it if you don't.

The Microsoft Study Etienne Referenced

Researchers from Microsoft and Carnegie Mellon studied 319 knowledge workers using AI tools. Their finding: the more confident someone was in AI's output, the less critical thinking they applied.

Not "less time spent." Less thinking. They stopped questioning. The AI sounded right, so it was right.

For a one-person tech operation, that's existential. There's no code review. No colleague saying "wait, why did the agent do that?" Just you and the output.

What I'm Doing About It — A Survival Kit

Four habits I've built over the past year:

1. Go deeper, not wider. I don't need to test every new model. I need to know two of them really well. Depth beats breadth when you're the only engineer.

2. Break things on purpose. Every Friday I run adversarial inputs through our agents. Not to break them — to understand where they break. The edges teach you more than the happy path.

3. Ship less, understand more. One feature I can explain to a client beats three features I can't debug at 2am.

4. Keep coding. The old advice was "get out of the code." Etienne is right — that advice is now dangerous. In the AI era, distance from the code means distance from reality.

Compression Continues

The CTO role isn't going away. It's compressing. Fewer people, more responsibility, higher stakes. The fractional CTO model — my model — exists because most companies can't afford a full-time CTO anymore. They need the brain, not the title.

But the fractional model only works if you're actually in the code. If you're just attending meetings and talking strategy, you're a consultant. If you're in the architecture, debugging the edge cases, shipping the fixes — you're something new.

Something the industry doesn't have a name for yet.

And maybe that's fine. Function over title. That's the compressed way.

Inspired by and sourced from Etienne de Bruin's CTO Sub. His framing of Token Monkey vs Vibe Coder is worth your time.