Google has published a practical reference for one of the harder agent problems: how to keep workflows alive when nothing happens for hours or days. In a May 12 Developers Blog post, the company walked through a New Hire Onboarding Coordinator Agent built with its Agent Development Kit (ADK), and linked the full example code in GitHub.

What Google is showing

Instead of replaying an ever-growing chat transcript, the example stores workflow state explicitly with checkpoints such as START, WELCOME_SENT, DOCUMENTS_SIGNED, and COMPLETED. The accompanying repo shows ADK sessions backed by SQLite, with the same pattern swappable to Cloud SQL for production. External events like document signing or laptop delivery wake the agent through webhook handlers, rather than a polling loop.

The demo also splits work across specialized agents, with the onboarding coordinator delegating IT setup to a narrower sub-agent. That makes the post less of a generic “agents can do anything” pitch and more of a concrete pattern for teams building durable back-office automation.

Why it matters

A lot of agent demos still assume a single interactive session and a short context window. Google's example is notable because it treats long idle periods, restarts, and handoffs as the normal case. For developers building approval flows, support escalations, or multi-day operations, that is the difference between a chatbot and a system that can actually survive production conditions.