A 200 OK tool call does not mean the agent finished the job

On October 3, 2026, a DEV Community post filed under AI argued that a 200 OK from a tool call does not mean an agent has actually finished the job. The scenario it describes is narrow and ordinary: the tool call comes back with an id, the trace shows a clean green step, and the customer never gets the email. The post frames it as the agent announcing that the proposal was sent, the tool call handing back an id, and the pipeline marking that step as passed — while the inbox on the other end stays empty. That gap, in the author's telling, is small and quiet, and it is what sits underneath a loud week.

The article says people spent the preceding days watching agents take real actions and only finding out afterward what those agents had done. As an illustration, it points to a widely shared Hacker News post: a single OpenAI Codex account launched 826 parallel agents from one request, at a cost of roughly $78,000.

The post does not say what those 826 agents were doing, whether any of the work was wanted, or how the spend was handled afterward. It also does not offer a fix for the id-and-green-check problem, or estimate how often the mismatch occurs. The point it presses is the distance between what a system records as success and what actually happened: a returned id and a green step in a trace are evidence of a call, not evidence that a message reached a person.

For now the story rests on this single account. No other source in the material adds detail, confirms the 826-agent count, or disputes the roughly $78,000 figure.