Blog
Ideas, playbooks, and lessons on building and scaling with AI agents.
How ONE Remembers
Every AI memory system reaches for the same tool: embeddings, a vector store, a similarity score. Ours has almost none of that. Here's the brain, then the memory, and an honest read on whether skipping the vector database was actually a good idea.
The Scariest Bug in a Self-Improving System Is Silence
We wired production outcomes into our build queue's ranking today. The feature itself is small. What almost got past us — a failure mode that returns exit 0 and does nothing, forever — is the more useful story.
The Week the Weights First Moved
Our routing system had a beautiful learning formula, written months before the code. This week we discovered no weight had ever moved — and then we watched the system learn, forget, refuse, and repair itself for the first time. Receipts included.
Sixteen Holes, Two New Senses, and a Rubric That Fired Itself
Three days, roughly 470 commits: we closed every open security hole we could find, gave the system a semantic sense and a memory it can read from Drive, settled real money on-chain through it, and watched its own scorecard drop a rule it couldn't defend.
Your Anonymous Traffic Just Got Names
Most of the people who want to buy from you never tell you who they are, and the usual fix is a form wall that drives them away. There's a better way: notice what every channel already shows you, remember faces across visits, and ask one question only when you have to.
Alex Hormozi's 13 Essential Books for Business Owners
Alex Hormozi built a $200M+ portfolio and credits these 13 books. Here's the full list, why each one matters, and the order to read them in.
Why Your Referral Program Belongs On a Blockchain
A loyalty balance is a row in a database you don't control. Put it on a public ledger and trust becomes something the customer can check, fraud can't forge, and the payout can go to a person or an AI agent. Here's what changes, with real receipts.