Fresh out of school, I was hired as a junior developer by a car dealership data company. IBM had spent three years and roughly $6 million failing to deliver the system. Toyota gave the company six weeks: get it done or the deal was gone. I was told, directly, that if this did not work there would be no job.
On day one, my boss wanted to write code for every report. About forty reports. My first thought was simple: that is too much work. There had to be a better way.
I saw the shape of a recursively dynamic system. A universal architecture that could absorb variation through configuration instead of new code every time. My boss said no. Then he went to lunch.
So my buddy and I stayed behind. I designed the database. I coached him through the connecting code. I built the report configuration page that proved the idea. Once my boss saw it, the answer was obvious.
On day fourteen, I wrote code that wrote code. It translated every Toyota financial statement, from dealers with different charts of accounts, into a universal chart of accounts that could work for every manufacturer. When it ran, three years of data came alive. Every report lit up. The in-house accountant verified it.
Then we were told to pretend it did not work for four weeks and deliver it on the last day of the six-week deadline. While we waited, we configured Honda, Hyundai, and Kia too. Not by rebuilding the system. Through configuration.
That is still the work: solve the thing once, configure it forever, and stop making people execute what the system should carry.
The current version of this site used to soften that story. It said I built it on a lunch hour. It made the whole thing cleaner and smaller than it was. That is exactly what AI does when nobody extracts the truth first: it writes the plausible version instead of the real one.