The tool still responds
There may be no obvious hard outage, so you keep working.
How token waste happens
The output gets weaker
Plans become shallow, assumptions drift, and the code needs more review.
You pay twice
You spend tokens and then spend time fixing or replacing the result.
A practical prevention step
Before large implementations or refactors, check Claude Code and ChatGPT Codex status. If the signal is normal, proceed. If it shows caution, keep work small or wait.
The small popup can keep the status visible during long sessions.