Some come from work, like a bookkeeping close that took the better part of a fortnight. Others come from everyday life, like a poker night that kept stalling on arithmetic. I am not a software engineer, and I am not trying to become one. What I am is someone who understands a problem properly, and modern LLMs have made that enough to build the working fix.
That has been true the whole way along. I started out running camps for young people with disabilities, co-founded an inclusive employment project, then joined Visory in revenue operations and moved through operations into finance automation. The jobs have changed more than I have.
I am happiest when I can see a problem clearly enough to do something about it. Rather than writing a requirements document and waiting, I sit down with agentic tools and build the fix myself. In practice that means I will:
The exceptions are the job. The system takes the routine. A person takes the judgement calls, and only those.
Stop before the point of no return. Nothing is paid, sent or submitted without someone confirming it.
Hold, do not guess. An automation that quietly picks the likely answer is worse than one that stops.
Show the working. Per-run logs by default. If it cannot be audited it will not be trusted, and it should not be.
Boring technology. No framework, no build step. It has to still run in two years when nobody is thinking about it.
Ship the live one. A working thing beats a document describing a working thing.