# Microsoft Foundry restarts an interrupted hosted agent from its original input. Make every write safe to run twice.

*2026-10-29 ⬩ 8:00 AM ET | bryn*

Microsoft Foundry restarts interrupted hosted agents from their original input as of this week, so every write an agent makes has to be safe to run twice. The [Microsoft Foundry Blog post from October 28](https://devblogs.microsoft.com/foundry/microsoft-copilot-harness-in-foundry/) splits the job in two. Foundry spots the interruption and starts the work again. Your code saves progress and decides what's already done. Microsoft says it plainly: Conversation history alone isn’t enough to restore work in progress.

That's the half builders skip. Say an agent sends an email, then crashes before it records the send. On restart, it sends the email again. Microsoft's own example is a report: check that it exists before you create it. I'd make that the rule for every outside action. Give each one a key, like the account plus the step, and look the key up before acting.

For a GTM team, the sales and marketing side of a company, a duplicate isn't a glitch. It's a buyer reading the same email twice and wondering who's driving. A rerun should find its own receipt and stop.

## Sources

1. [Introducing the Microsoft Copilot harness in Foundry Agent Service](https://devblogs.microsoft.com/foundry/microsoft-copilot-harness-in-foundry/). Microsoft Foundry Blog, Oct 28, 2026.

See also: https://www.civic.com/field-notes/what-the-audit-log-saw

Source: https://www.civic.com/bin/agent-restarts-replay-the-input