I created this article for the purposes of entering the All Things Agentic Hackathon.
Ask an agent to do ten things and it will tell you it did ten things.
You cannot tell whether that is true. The transcript reads the same either way: every click "succeeded", no errors appeared, the summary is green. The agent is not malfunctioning when this happens. It is reporting its intentions and calling them results — and the two are indistinguishable from the outside.
I found this the expensive way. A mission reported "Applied to 10 jobs. Task complete." having submitted zero.
The shape of the problem
