Nobody can steal the thing you are actually afraid of losing

Two years of evenings live in one folder on my laptop.

Close to two hundred rules. Roughly as many automated checks. Every single one written after something went wrong, usually while an enterprise client was watching.

Nobody has ever seen it.

Whenever I imagine showing it to someone, my body reads the idea as danger. The same way it does when you leave your bag on a train seat and walk to the toilet.

You probably have a folder like this.

Maybe yours holds the internal runbooks nobody outside your team has read. Scripts you built slowly, from real consulting failures, that would take another person a long time to arrive at on their own.

And you have probably had the thought I had this week.

If I share this, what stops them from taking it.

I spent a week protecting the wrong thing

Here is the embarrassing part.

I had already built the defences. One custom licence forbidding commercial use. One file sitting at the root of the repo that speaks directly to AI agents and instructs them to refuse to clone the work and to contact me instead.

I was proud of that file. It felt clever.

Then I asked a differently trained model to attack my own reasoning, which is a habit I picked up after being confidently wrong four times in a row about something I was certain of.

It answered in one line.

That file is a sign taped to a door that never locked.

I sat with that for a while, because it is correct, and because I had spent real hours feeling protected by it.

Say the uncomfortable part out loud

You cannot prevent distillation. Not slow it. Not make it hard. You cannot prevent it.

The moment a human or a model can read your work, the boundary has already been crossed. Somebody reads two hundred rules, understands the judgement underneath them, and rewrites the same thinking in their own words. No licence stops that. No file format stops that. No warning addressed to a robot stops that.

I want to be precise here, because half-truths on this topic get expensive.

Your licence does buy you three real things. Legal grounds if somebody commercialises your work. Friction with honest people and with tools that follow stated policy. Documented intent, which matters if there is ever a dispute.

That is worth keeping. Keep it.

Now be clear about what it cannot touch. Somebody who reads your work and rebuilds the ideas in their own language. Any local model that ignores your instructions entirely. Screenshots. Copy and paste. Embeddings. The client who absorbs your judgement over six months and then hires somebody cheaper to reproduce it.

That last one is the one that actually happens.

Move the boundary instead of reinforcing the wall

If prevention is impossible, the goal becomes placement.

Almost everyone makes the same mistake, myself included until this week. We treat a knowledge artifact like a software binary. We assume a licence plus a warning protects the value. Once the value is readable, the protection has already failed.

So the boundary has to sit somewhere else. Around access, around execution, around updates, around the relationship. Never around files that somebody can open and read.

In practice that means sharing the running of the thing rather than the contents of the thing.

Somebody sends a task, they get an output. They never receive the two hundred rules. Your source stays private. What reaches them is the minimum they need to get a result, and what you licence is the usage.

Every serious tooling business already works this way. What surprised me was how long I resisted applying it to my own work, because my instinct said the files were the asset, and my instinct was wrong.

What a copy cannot capture

Here is what changed my mind completely.

Snapshots can be distilled. Living systems cannot.

Today alone, that folder gained three new enforcement systems. Every one of them exists because something broke in front of a real person. One client received an email from me that arrived as unreadable markup, in the middle of a competitive bid against another vendor. One layout looked correct on my own watch and would have clipped on half the models people actually wear. Each failure wrote a permanent rule.

Then the part I did not expect. I asked a rival model to attack all three designs before shipping them. It corrected me seven times in one day. Seven. On my own work, in my own domain, where I was confident.

Anyone who copies that folder today owns a photograph of a moving thing.

Six weeks from now their copy is a fossil, and mine has absorbed six more weeks of real failures, in front of real clients, with real money on the line. That gap lives in whoever keeps running the loop, and no file contains it.

The honest answer to what is protectable is the rate at which you improve, never the artifact itself.

Guarding it too hard has a cost nobody warns you about

I need to say the other half, because it is my own weakness and it may be yours.

For years I did excellent private work and shipped almost nothing in public. It felt productive. It was invisible. Nobody hires the person whose best work nobody has ever seen.

Four separate organisations chose me as an ambassador for their tools. That happened because of things I put out, never because of the folder I hid. Every euro I have earned came from something visible.

So the fear of being copied and the fear of being judged wear the same coat, and I have spent years mistaking one for the other. If you are guarding a system that has never earned you anything, you are avoiding exposure and calling it strategy.

The split that resolves it is easy to say and hard to do.

Publish the pattern. Never publish the machine.

Write about the failure and the shape of the fix, in enough detail that a reader genuinely understands the problem and roughly what to do about it. Leave out the wiring that makes executing it clean. If somebody can read your article and ship your exact solution by pasting, you gave away the machine. If they finish it able to see the problem clearly for the first time, you gave away the pattern, and that is what brings people back to you.

I have been on the wrong side of that line in both directions. Too closed for years, then briefly too open once.

The folder is still private. The licence is still there, doing the small honest job it can actually do. What changed is that I stopped believing it was a wall, and started treating the next six weeks of failures as the part worth having.

Your turn

What are you sitting on that nobody has seen yet.

If this was useful

I work through this in public, the wins and the freezes both, mostly on LinkedIn and YouTube. If the real version of building in the open is useful to you, that is where it lives. Find me on X, GitHub, and the work at next8n.com.