How this channel uses vaults ## How this channel uses vaults In this channel's material, the term vaults appears in the context of the production infrastructure that agent platforms must provide. When Katelyn Lesse (head of engineering, Claude platform) describes the real wall teams hit when moving agents from demo to production, she points at the unglamorous plumbing: keeping servers running, storing transcript data, and secure sandboxing — the kind of safe, isolated storage where an agent's records and secrets can live without exposing the rest of the system. [moment: 1181] [^4872db36-e98a-5f6b-bc5c-b11e48c9411b@1181] ### Why this matters Lesse's broader argument frames why secure storage like vaults is not optional: - A dying sandbox takes your whole agent with it, so isolation and reliability are one concern, not two. - Reliability, not peak performance, is the real value a platform delivers — a Mac mini going down inside a bigger company is unacceptable. [moment: 692] [^4872db36-e98a-5f6b-bc5c-b11e48c9411b@692] In other words, vaults belong to the same family of production concerns — uptime, memory, transcript storage, sandboxing — that Lesse identifies as the actual bottleneck, in contrast to the harness engineering most teams expect to be the hard part. [moment: 813] [^4872db36-e98a-5f6b-bc5c-b11e48c9411b@813] - Vaults are part of the production-infrastructure layer Lesse says is the real wall for agent teams. - They sit alongside transcript storage and secure sandboxing as safe, isolated places for agent state. - Their importance follows from the platform's core promise: reliability over peak performance.