Observability

Being able to answer "what did it do, and why?" after the fact — every step, decision, input and output written down somewhere you can go and look.

The job file with everything in it. Who was on site, what was ordered, which day the delivery slipped, who signed off the change. When the customer rings up annoyed six weeks later, you can actually answer them.

The one catch is that it has to be built in from the start. You can't go back and observe last Tuesday if nobody was writing anything down on Tuesday.

Why it matters: an agent you can't observe is an agent you can't trust, fix or improve — you're left arguing about impressions. It's also the question worth asking before you buy: when this gives a customer a wrong answer, can you show me exactly what happened? If that answer is vague, so is everything else.

Related: Telemetry · Tracing · Logging