Resilience Coffee/2026/08/21

• Availability and Reliability:
• Orthogonal:
• Systems can be unreliable but high availability
• e.g. self-hosting may be more reliable because the change rate can be smaller or off-peak
• Not that the failure modes are reduce, but rather that change rate is reduced
• "How can you have a critical system and not control it?" e.g. GitHub
• Tax Season deployments for a tax preparer software
• https://incident.io/blog/we-turned-off-pub-sub-and-nobody-noticed
• Scale of impact and duration of impact
• Object to self-hosted as more available: maybe!
• Steve McGee's Inverted pyramid model https://wiki.resilience-coffee.org/wiki/Inverted_reliability_pyramid
• Reliability beyond availability: e.g. incorrect search results
• Or LLMs hallucinating
• Should SRE be able to control e.g. feature flags?
• Help teams help themselves vs. separating responsibilities
• Connection to leadership maturity and "rowing in the same direction"
• SRE as limiting options:
• Paved path, guardrails, desire lines
• Use RFC process
• Can this happen post AI?
• Maybe in an ADR style approach where the LLMs pull the RFCs into context
• AI can make the "quality" of a RFC better, but social benefits lost: alignment of decision making process across a group of humans
• "Let people touch the hot stove" as a precursor to group learning
• Related to educational philosophies
• "Looking for trouble" meetings
• Categorization of incidents: is it valuable?
• re: https://direct.mit.edu/books/monograph/4738/Sorting-Things-OutClassification-and-Its​