Automation is ubiquitous—it is entwined in our daily lives in ways that we aren’t always aware of. It has been woven into all aspects of modern software by being presented as a utopian vision: a way of making human lives easier, doing repetitive tasks faster and with fewer errors, freeing us fallible humans up to do other ostensibly more important work. But anyone who has worked directly with automated systems knows that we are still very far from such a dreamy reality.

This talk delves into detailed research about how automation is involved in software incidents. My focus on this area stems from the growing portrayal of automation as a panacea for various software incident issues, despite its limitations in effectively addressing these challenges, such as reliable detection and resolution of software issues or analyzing and disseminating learnings from these incidents back into the organization and its products and services.

Drawn directly from public incident reports (collected in the VOID), the research revealed multiple, often competing, roles that automation can play over the course of an incident, and most importantly underscored how important humans are at understanding, troubleshooting, and recovering from automated software issues. It’s a slightly vexing perspective, but all is not lost! I close with research-backed suggestions for ways we can coexist and build better systems in partnership with automation.