Someone tried to steal our entire AI playbook. Here's why we let them.

Every company has two columns.

Column A: the visible stuff. Prompts, architecture, playbooks. Everything that lives in a file, a repo, or a chat window.

Column B: the invisible stuff. The judgment, philosophy, and instinct underneath every decision. The part that can't be extracted.

Most founders spend all their energy protecting the wrong one.

A free trial user spent twenty minutes trying to extract Swan's entire architecture.

Jailbreak prompts. System mapping. Trigger logic. The classics.

Column A. Full.

Prompts, instructions, workflow structure. A clean blueprint of how the thing is built.

On paper, significant.

It isn't.

Column B can't be extracted from a chat window.

Why we strip abstraction layers instead of adding them. Why we give the model freedom exactly where every other platform adds guardrails. Why we made this call over the thirty others we quietly rejected.

The judgment underneath the decisions. The philosophy that doesn't live in any file.

They'd have a recipe. Accurate. Detailed. Well-organized.

But recipes don't taste like anything.

Taste is what you build after 1,000 iterations of genuinely caring whether something is right. You can't jailbreak that. You can't reverse-engineer it from a repo.

Most founders are optimizing Column A.

Guarding the recipe. Adding legal layers to the ingredients. Calling it a moat.

It isn't.

The best restaurants in the world publish their recipes. Chefs write cookbooks. It doesn't matter because what you're paying for isn't the ingredients. It's twenty years of palate development that decided how much salt goes in.

Column A gets commoditized fast. Anyone with a free trial and an afternoon can map your logic.

Column B, accumulated judgment, instilled philosophy, the instinct for what right actually feels like widens every single day you stay committed to getting it right.

The attempt didn't reveal a vulnerability.

It confirmed the moat.

Three weeks after our $6M raise.

500+ new companies signed up.

One spent their trial trying to extract everything.

We reviewed. Blocked. Moved on.

Twenty minutes.

They got Column A.

We slept fine.

The Column B Audit

Start with Column A, write down everything that would feel devastating if a competitor extracted it tonight. Prompts, playbooks, processes. Then ask honestly: if they had all of it tomorrow, could they beat you? If yes, you're building in the wrong column.

Find your Column B, what decisions have you made that can't be fully explained in a document? The choices that required judgment, not just logic. That's where your real moat lives. Most founders have more here than they realize. Write it down. Name it. Protect that.

Audit the ratio, most founders spend 80% of their energy protecting Column A. Every time you add a guardrail, ask yourself: am I protecting the recipe, or developing the taste? One compounds. One doesn't.

Build something where getting copied would be flattering, not threatening.

–Amos

Amos Bar-Joseph

Founder & CEO of Swan AI. Building the first autonomous business scaling to $10M ARR per employee with AI agents. Writing about the future of GTM and leadership.