Quick note before we dive in. I skipped last week's send. I was off-grid at a 4-day hypnosis training, deliberately rewiring my own decision-making. Sounds woo. Bear with me. It's actually on-topic.
Because if your code isn't your moat (and it isn't), what is? Increasingly, the answer points back to the one piece of infrastructure you can't outsource to Claude: your own cognition. Taste, judgment, speed of pattern-recognition. The thing I just spent four days deliberately upgrading.
So let's talk about what's actually defensible in 2026.
Three months ago I rebuilt a $200/month OpenClaw setup for $15. I wrote it up. 38,000 people read it.
The uncomfortable part wasn't the savings. It was how easy it was. Two evenings, one Claude Code session, a $5 VPS. The original took someone months.
Here's the thing nobody wants to say out loud: if I can do that to someone else's stack, someone else can do it to mine. And to yours.
What used to be a moat
Five years ago, the moat list was simple: a clever architecture, a hard-to-build feature, a polished UI, a custom integration nobody else had bothered to wire up.
Every single one of those is now a weekend project for anyone with Claude Code, a CLI, and 20 hours of focus. I'm not being dramatic. I'm describing what's actually happening in my own GitHub history every week.
Your prompts? Leak the second a screenshot hits Twitter. Your codebase? An LLM can read and rewrite it faster than your senior engineer. Your "secret sauce" workflow? I bet I can reverse-engineer it from your public output in an afternoon.
The code isn't the moat. The code is the table stakes.
What actually defends you in 2026
I've been auditing my own work for two months. Here's what survives the LLM-commoditization test.
1. Distribution you own. Not Twitter. Not LinkedIn. An email list. A Discord. A Beehiiv. Something where the algorithm can't decide tomorrow that you don't exist.
2. Taste, and the speed to apply it. LLMs generate average. The moat is knowing which 5% of the output to keep, and shipping it before anyone else finishes their planning meeting. Taste is a compounding asset. Every shipped artifact sharpens it. This is also why I went to that hypnosis training. Taste lives in your nervous system, and your nervous system is upgradeable.
3. Receipts. "I did this thing, here's what happened, here are the numbers." Public track record beats any pitch deck. People don't trust your code. They trust that you've been right before.
4. The compound interest of being early and consistent. I've been writing about Claude Code, OpenClaw, and CLI agents weekly since February. Not because each post is a banger. Most aren't. Because the 50th post compounds in a way the 1st never could.
What this means for you this week
Stop polishing the code. The code is rented from Anthropic anyway.
Pick one: ship something publicly that proves a point you have receipts for. Send an email to the list you've been "growing" for six months without sending anything. Write down what you've actually built, with numbers, somewhere indexable.
The moat isn't in your repo. It's in your inbox, your archive, and the speed of your next ship.
Which moat would you actually bet on for the next 12 months?
P.S. On vibe coding
I wrote a book about this exact shift. Not for senior engineers. For the founders, marketers, and operators who got handed an LLM and told to "just build it."
It's called Vibe Coding, For Real. It's about how to ship real software when the code itself stopped being the constraint. If this newsletter resonated, it's the long version.
— Phil