The 2am test: every workflow should survive the night

Here's a test for every workflow you own: if it had to run tonight at 2am with nobody awake, would it survive until morning — or would you find the wreckage in Slack?

Most processes fail. The ticket queue fills with no triage. The deploy waits on an approval nobody sees. The follow-up sequence pauses because step three needs a human decision nobody documented. We staff around this fragility with coverage, rotations, and the unspoken rule that somebody is always slightly on call. That's not a system. That's a vigil.

What passing looks like

A workflow that passes the 2am test has three properties. First, defaults: every routine case has a pre-approved path, written down as guardrails, not tribal knowledge. Second, a complete log: everything the night run did is readable in the morning with reasons attached, so review takes minutes. Third, a short escalation list: the few cases that genuinely need a person wait patiently in a queue instead of dying silently or, worse, proceeding blindly.

Notice what's missing from that list: heroism. Nobody gets paged. Nobody "stays on top of it." The work either runs on rails or waits in line.

Running the test

Pick your most apologized-for queue and ask three questions. What would the night run do with the routine 90%? (If the answer is "it depends," your guardrails aren't written yet.) What would you read in the morning to trust it? (If the answer is "I'd have to redo it," your log doesn't exist.) What waits for a person, exactly? (If the answer is vague, your escalation list is "everything," which means nothing.)

Agents are good at the night shift the way you'd hope: tireless, literal, and honest in the log. But the test comes before the tooling. Write the answers down, and you'll know whether you need software, staffing, or just a decision you should have made months ago.