Being able to see what your system actually did and why — every step, decision, input, and output recorded and inspectable. Not a dashboard bolted on later; a design property.
Why it matters: an agent you can't observe is an agent you can't trust, debug, or improve. For AI features it's the feedback loop: log what went in and out, track quality, fix without guessing.
More in “Keeping it running in the real world”