OpenAI's Codex repo added structured OpenTelemetry events for multi-agent communication on July 1. Commit 129ea2aaf5 logs spawn, message, followup, and result sends, then logs receives with the same communication ID.
Why this belongs in a compact slot, not the lead: it is a sharp observability receipt, but it continues yesterday's Codex background-work arc and does not change a public operator workflow by itself.
Signal
The useful movement is traceability. The new agent_communication.rs path records sender thread, receiver thread, communication kind, state, content, and a shared ID so operators can join send and receive events.
Implication
For platform teams testing multi-agent runs, this gives a concrete next check: delegation should leave delivery receipts that can be audited when a child agent misses, delays, or returns the wrong thing.
What To Watch
The caveat is that logs are not policy. What to watch next is whether these events become surfaced in user-facing run diagnostics, replay tooling, or failure summaries rather than staying as backend telemetry.