Do you struggle with agent state continuity across restarts?

Quick sanity check for people running agents in production:

When an agent restarts (crash, redeploy, scale, upgrade):

  • Do you lose behavioral state?

  • Do you rely on prompt replay or ad-hoc memory?

  • Do you have deterministic replay or tamper detection?

If this is a real pain, how do you handle it today?
If it’s not a pain, why not?

(Not selling anything. Pure problem validation.)