A change boundary is the interface where a change crosses from one state or subsystem to another in a complex system. Not all change boundaries are equally risky.
**Two categories:**
| Type | Downside Risk | Hedgeable? | Operator Role |
|------|--------------|------------|---------------|
| Low-risk | Small | Yes (e.g., auto-scaling groups for EC2) | Automation sufficient |
| High-risk | Large | No (e.g., deploy with subtle bugs) | Active human operator required |
**Core insight**: The safety of complex systems must be considered *holistically* — human operators included. Simple systems can be analyzed in terms of failure probabilities; complex systems require accounting for the human's ability to detect and correct errors that automation misses.
**Implication for design**: Make change boundaries *visible and explicit*. Automation should not obscure distinctions between safer and riskier actions. When an action traverses a high-risk boundary, the human operator should know it, understand the ambient risk, and be primed to respond.
**Source**: Coda Hale, "Risky Business Requires Active Operators," blog.skyliner.io, ~2016.