Everyone's out hiring prompt engineers.
Skip it. You don't need better prompts. You need better memos.
The unit of software at Ledger Collective isn't a ticket anymore. It's a skill: how we do something, written down so the AI can use it. Ours read like a memo to a sharp new hire: here's what you'll be given, here's which tools you may use, here are the steps, here's what "done" looks like. They live in the repository next to the code. They get reviewed like code. They get versioned like code.
The rule. Every good skill ends with one. Our bank-reconciliation skill closes with a single line: "Does NOT post adjustments to any system." But the rules that matter most don't live in the skill's memory. They live one level down, set once as defaults on the tools themselves, at the org level. Nothing posts to a system of record without a human approving it, right there in Claude. Every client, every skill, every time. A clever prompt can't talk its way past it, and a new skill inherits it for free. And when a rule matters to the code, we write it as a test that fails the build. A wiki asks you nicely. CI stops you.
The author. This is the part that changes everything: the person who knows the job best can finally write the software. The accountant who knows what a clean journal entry looks like doesn't file a request and wait. She writes the memo. Her judgment, captured once, used every morning. Nothing lost in translation.
The habit. You don't prompt harder. You teach the machine the way you'd teach a new hire — in writing, under version control — and you get it the data.
The programming language of this era is prose. Write it down.
— Ashley
First shared on stage at the Capital One AI Speaker Series in Plano, Texas. Thank you, Howard Dierking, for hosting. One of four essays from that talk.
Keep reading It Was Never the Code. It Was the Translation. → For decades we priced software by the engineering. The expensive part was always the explaining.