We run every release on our own agents first
Before an update reaches a customer's agent, it has been running ours. Here is what that catches, and what it does not.
There is a version of hosting that treats an update as somebody else's problem. New release lands, it goes out, and the support inbox becomes the test suite. It is efficient right up until it is not.
We do the boring version instead. Every OpenClaw and Hermes release runs on our own agents before it is rolled out to anyone else's.
What that actually catches
Not correctness bugs, mostly. Those are what tests are for. What it catches is the category that only shows up in a running system with real history behind it:
- Configuration that no longer means what it meant, so an agent comes back subtly different.
- Restarts that take longer than expected under a real workload rather than a fresh one.
- Channel reconnections that work on a clean setup and stall on one that has been live for months.
What it does not catch
Our usage is not your usage. We do not have every channel combination, every skill, or every integration in play, so this is a filter and not a guarantee. That is exactly why the dashboard keeps update, undo, pause, restart and reinstall as things you control rather than things you request.
Testing on ourselves buys you a shorter list of surprises, not an empty one.
It also changes the incentive in a way we like. When a release is rough, we find out first, at our own desks, before anyone has to write a support message about it.
Stop reading about it, build one
Set up an agent, pick a channel, and have it working inside the app you already keep open.