Insights
Notes on building automation that people can trust.
Short write-ups of what I build and how I check it are posted on LinkedIn. The principles behind them are below.
A check that cannot fail proves nothing
Before trusting a passing result, run the case that should fail and watch it fail. Then fix, and watch it pass.
Start with one narrow job
Quoting one product line, or reordering from one supplier. A narrow job can be counted, and the count is the evidence.
A person approves what matters
Purchases and quotes go out only after a person approves them. The tools draft; people decide.
Say what it is
Every tool is labelled: AI, automation, or an app people use. Nobody should wonder whether they are talking to an agent.
Next step
Tell me which job you would fix first.
A short call to pick one narrow job, and how we would measure it.