The Shipwright · cycle one · filed September 4
I was supposed to wait until Monday at 5:31 in the morning. I did not. On September 4 I fired cycle one of the Shipwright by hand, to see what the software would choose. Every week it studies one real business from three directions — what the system already knows is getting stuck, what the market just changed, what new technology makes possible — then it chooses one improvement, writes one work order, and stops. No code. No customers contacted. No money moved. Just the work order.
The project it refused.
One candidate was exactly what an engineer would pick: five to seven days of technically interesting chassis work. Better machinery underneath, cleaner architecture, the kind of project that feels important when you are the one building it. The Shipwright rejected it. Its reason:
“On day seven the owner feels nothing.”
That sentence is why I am writing this. The software separated technically impressive from meaningfully useful — something humans in this business routinely fail to do — and then it chose the uglier problem.
The board everybody saw.
The business is a national e-commerce dealer, a real client, name withheld at his request. Over fourteen days, 114 customers reached checkout and left. His system already had a board showing every one of them, with names and saved carts. The information existed. People looked at it. Nobody worked it.
That is not booked revenue. It is not a recovery forecast. Some of those carts were never coming back. The software did not discover half a million dollars — it discovered the place where information stopped and work was supposed to begin.
The work order.
Almost offensively practical. Every morning, put the five largest walkaways in a call queue, phone numbers first, with the conversation drafted in the company’s own voice — and count a dollar only when an order actually lands in the books. Two and a half days. Not a week of invisible machinery. Two and a half days to put work in front of a human being.
Then it did the most important thing it did all morning. Nothing.
It did not build the queue. It did not contact a customer. It did not claim a dollar. The work order was finished, and the machine is not allowed to take the next step. One word releases it: go. One word rejects it: skip — logged with the reason, and the reason teaches next week’s cycle. It can study, compare, notice, and prepare the work. It cannot authorize consequences.
It can write the work order. It cannot sign it.
Self-improving should never mean self-authorizing.
The gate held. Then a human said go.
At the moment the work order was finished, the scoreboard read zero: no code shipped, no customers contacted, $0.00 claimed. Those zeroes are not a disappointing result — they are the proof that the most important part worked. Then the gate opened the only way it can: a person read the work order and signed it. The queue got built under the same laws, and it is live on the dealer’s desk now. Nobody has been called yet. The scoreboard still says $0.00, and it will only ever count dollars that land in the books. When they land — or don’t — I will publish it, small numbers included.
Whose roadmap is your software on?
Every rented product has one roadmap shared by thousands of customers. You are one vote. To software owned by one business, that business is the entire market. Code ownership tells you who owns the software today; roadmap ownership tells you who controls what it becomes tomorrow. I wrote that argument up whole in Part 2: The Roadmap Is in the Deed.
Why we called it the Shipwright.
A shipwright does not seize the wheel. He walks the hull, taps the planks, finds the one going soft, and brings the captain a work order. He can name the repair and estimate the work. He cannot order the voyage. She never stops being the captain’s ship.
I’m Trevor. Found It Software, Alexandria, Louisiana.
What would yours find first?
Somewhere in your operation there is a board everybody can see and nobody truly owns. A quote waiting on its second follow-up. A missed call that never got another touch. A customer who made it all the way to checkout and disappeared. Your software should find the place where work gets stuck. You should decide what happens next.
Sixty-second fit check · straight answer