When /bin/sh Attacks: Revisiting "Automate All the Things"
summary
Argues that automation often hides failures and erodes operator skill: manual skills deteriorate if automation runs tasks for months, auto-scaling and Kubernetes can camouflage system state, and ML or vendor black boxes make root cause opaque. Shows engineers use three heuristics—change, go wide, convergent search—and that postmortems delay beyond 72 hours lose vital context. Recommends simulation, game days and designs that buy time so operators can understand, interrupt, and recover when shell scripts or autopilots misbehave.
AI-written summary · about
chapters (34)
- 0:00Intro
- 0:52Safety Sciences and Resilience Engineering
- 1:30Context
- 2:58Thought Experiment
- 3:39Emergency Landing
- 5:58Success or Failure
- 6:16Act 1 Resilience
- 7:11Change
- 7:56Going Wide
- 8:34Convergent Searching
- 9:22Engineers Use Incident Memory
- 12:05How Do You Get Better
- 12:18Elements of Expertise
- 14:04Experts see what is not there
- 15:07Transformation of experience into expertise
- 16:32Experience comes from failure
- 16:45Rasmussen triangle
- 17:04Economic Failure
- 19:07Review Failure
- 19:18Example
- 21:18Correspondence Bias
- 21:49The IKEA Effect
- 22:21Bias
- 25:29Questions
- 27:16Ironies of Automation
- 30:51Tracing Decisions
- 31:59Joint Cognitive Systems
- 33:53Technology Mistake
- 34:13AnimaC Paradox
- 35:54Whats the difference
- 36:40How do you deal with these ironies
- 38:28Epilogue
- 39:00Success and Failure
- 40:28Incident Remediation and Prevention
related talks
41mWhen /bin/sh Attacks: Revisiting "Automate All the Things"
55mThe Troubles of Automating “all the Things”
38mWhy Automating Everything Adds to Your Toil
33mSystems Automation: Past, Present and Future
32mManual work is a Bug!
31mExploring the Unintended Consequences of Automation in Software
28mThe Evolution of Automation
44mContinuous Stupidities - Myths of DevOps