Ask anyone why software costs so much and you'll get the same answer: engineers are expensive.
They are. That's just not where the money goes. It goes to translation.
Think about the old assembly line. The expert explains the job to a product manager. The PM writes a ticket. An engineer builds the ticket. Every hop loses exactly the nuance that mattered. And you pay for every hop twice: once to run it, once to fix what it lost.
I found the bottom of this at Ledger Collective, our own company. I'd built the first cut of the data layer as an R&D push: one governed interface over every system our accountants touch, and a small set of broad tools behind it. Verbs, really. Pull the bank feed. Pull the ledger. Match to the penny. Not one of them knows what a reconciliation is. And a first cut is all it was. It's our engineering team who make sense of it and keep it safe as it becomes the thing we run every day. Then I sat next to Jess Liberman, who keeps the books at Lion Chaser, and spent hours translating her month-end into something the AI could run. I still didn't nail it. The steps were right; the judgment was mine, not hers.
So she wrote her own. Not a tool, a skill: her reconciliation, written the way she'd teach it, arranging those same broad verbs with every minute detail that lives in her head and nowhere else. Which account gets the benefit of the doubt. What "close enough" means, to the penny. When to stop and flag a human instead of guessing. Engineering owns the verbs; she writes the sentences. Hers were profoundly better than mine, because she's closer to the problem. She didn't say "that's impressive." She said, "I can schedule this for every morning."
Nobody got replaced at that table. The thing that died was the middle.
That table taught me a rule I still use. If the accountants don't ask for it, it's safe to assume it isn't their biggest problem. Sometimes the real fix is the org chart, or one person who is overwhelmed, and no tool helps with that. Ask them first, in their words. Then decide if it's even a software problem.
So run the real accounting: the most expensive thing in software was never the code. It was the translation. And translation is a cost that drops to zero the moment the person who knows the job can reach the data herself.
Quit trying to solve your business partners' problems. Just give them the data.
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 Not Stepping Away. Stepping Into the Lab. → On July 1 I move from full-time CTO of Ledger Collective into a board and advisory seat. Not a goodbye, just me going back to what I'm built for.