Blog
Email security for AI agents.
Mailbox setup, human review, document workflows, and what it takes to run agent email in production — written by the people building it.
2026-09-20 · James McIntosh
A safer document-email workflow for AI agents
A staged approach to invoices and other attachments: inspect metadata, expose only the content the agent needs, and govern the reply.
2026-09-20 · James McIntosh
What should happen when an agent reply needs human review?
Design the held-message path so the agent stops retrying, the reviewer gets useful context, and the final action stays attributable.
2026-09-20 · James McIntosh
How to give an AI agent its first mailbox
A practical first-run pattern: one scoped mailbox, one agent key, and a simulator send before any real recipient is involved.