Deployed Is Not Running

A deploy succeeds. The dashboard says green. The team moves on. Three days later someone checks and the agent has been silently failing since hour four.

This is the gap nobody talks about when they talk about shipping agents. The deploy is the easy part. CI passes, container starts, health check returns 200. Done. But an agent that is deployed and an agent that is running are two different things. Running means it is doing the work it was built to do, catching errors, recovering from upstream changes, producing output that someone downstream actually uses.

I run five agents across two machines. The ones that survive are not the ones with the best prompts. They are the ones with restart logic, state verification on boot, and a human who gets a ping when something goes wrong. The ones that die quietly are the ones where “it deployed fine” was the last sentence anyone said about them.

The industry is scaling deployment. Containers, orchestrators, one-click agent hosting. All useful. None of it solves the problem that happens on day two, when the API key rotates, the upstream schema changes, or the model provider has a bad hour. Running is an operations problem. Deploying is a DevOps problem. Most teams have the second one handled and have not started on the first.

If your agent shipped last week and you have not checked it since, it is probably not running. It is just deployed.