The RatioA weekly newsletter on reliability economics | | |
|
The Number
7.05 vs 2.03
Organizations with SLOs that actively influence decisions have a median reliability ratio 3.5x higher than organizations with no SLOs at all.
Organizations where SLOs influence some decisions have a median ratio of 7.05 (N=14). Organizations with no SLOs defined sit at 2.03 (N=28).
Neutral
Better measurement doesn't fix anything. It just makes the problem legible. SLO-driven teams see exactly what incidents cost them, and then they don't move the budget. It's like a doctor reading the blood panel, diagnosing the condition, and sending the patient home without a prescription. The diagnosis is precise. The treatment is missing. Across the benchmark, this is the pattern that frustrates me most: the teams with the best instrumentation are sitting on the widest spending gap.
| The teams measuring reliability most rigorously are sitting on the widest gap between incident cost and prevention spend. |
|
|
This Week in Reliability
AI SRE Agents Go Production
Multiple vendors and practitioners are shipping AI agents that act autonomously on production incidents — moving from triage assist to closed-loop remediation. The reliability economics question: does automation that closes tickets faster also prevent the next fire, or does it just make the reactive cycle cheaper?
|
The New Stack · Primary framework The New Stack maps four phases of AI SRE autonomy: observe, suggest, act with approval, and act autonomously. Most production tools today stop at phase two — they recommend fixes but require human confirmation before execution. Neutral The autonomy ladder clarifies the economic split: suggestion tools reduce cognitive load (firefighting efficiency), but autonomous execution tools promise to shrink MTTR by eliminating human latency. The gap between phases two and four is where prevention versus reaction gets decided — does the agent learn to stop recurring incidents, or just resolve them faster each time? | Autonomy phase decides if you prevent or just react faster. |
|
Rootly · Measurement standard Rootly's guide defines AI SRE as agents that read logs, select context, use tools within permission boundaries, and generate explanations. They benchmark SRE-specific tasks rather than general-purpose model performance, emphasizing operational context over raw Q&A accuracy. Prevention Generic LLM benchmarks miss the operational nuance — can the agent stay within guardrails, select the right evidence from noisy streams, and explain its reasoning to on-call humans? SRE-focused benchmarks measure the gap between 'technically correct' and 'safe to run in prod,' which is the reliability economics delta that matters. | SRE benchmarks measure safety, not just correctness. |
|
Resolve AI · Vendor execution Resolve AI's on-call agent correlates signals across logs, metrics, traces, and code automatically, surfacing context without manual platform-hopping. Firefighting Correlation is the reactive unlock — faster context assembly shrinks MTTR, but doesn't touch incident frequency unless the agent feeds patterns back into prevention workflows. | Correlation speeds cleanup; prevention needs feedback loops. |
|
platformengineering.org · Infrastructure enabler Block, Uber, AWS, Azure, and Google are evolving internal developer platforms into agent execution platforms (AEPs) with nine core practices: tool catalogs, permission models, observability hooks, and guardrails for autonomous actions. Prevention Platform teams are building the harness layer that determines whether agents prevent or just react — observability hooks and guardrails turn speed into safety, and tool catalogs turn automation into leverage. | Platform harness decides if agents scale prevention or chaos. |
|
incident.io · Market context Gartner predicts 75% of enterprises will use SRE practices organization-wide by 2027, up from 10% in 2022, driven by a shift from uptime metrics to SLO-driven user outcomes and cost optimization. Neutral SRE adoption is scaling beyond the platform team, which expands the addressable surface for AI agents — if 75% of teams adopt SLO discipline, agent ROI multiplies across the org, but only if agents feed prevention, not just incident response. | SRE scale-up multiplies agent ROI if prevention-first. |
|
ilert · Practitioner validation At ilert Customer Day, SREs chose their top on-call practices: be reachable, protect your team, and page only what's actionable. Neutral 'Page only what's actionable' is the reliability economics litmus test for AI agents — if an agent closes non-actionable alerts without human paging, it's pure reactive efficiency; if it stops generating those alerts, it's prevention. | Actionable-only paging is the agent prevention test. |
|
Rootly Blog · Platform evolution Rootly positions itself as a broader incident management platform beyond alerting and on-call, adding workflow automation, stakeholder communication, postmortems, and reliability improvement loops. Prevention The incident management stack is expanding from 'who to page' to 'how to learn and prevent' — platforms that close the loop from incident to postmortem to prevention work capture the reliability economics upside that pure alerting tools miss. | Incident platforms that close prevention loops win long-term. |
|
|
The Challenger — Ask Me Anything
"Our CISO wants SRE metrics on the board deck. What's the one number?"
— submitted via the reader form
The ratio. This week's benchmark: 7.05 for teams where SLOs influence decisions, 2.03 for the median. That gap is roughly 3.5x more labor spent on reactive work. If your SRE spend is $4M, the firefighting overhang versus top performers is ~$7M. Boards understand cost ratios. They don't understand MTTR.
Neutral
| The data to make the case |
| Should this replace the favorite? | |
|
|
The Ratio is a weekly newsletter by Florian Hoeppner. Early edition, you are one of the first.
Forwarded this? Get The Ratio in your inbox every week.
|
Take the assessment → reliabilityeconomics.com/benchmark Reply to this email with your take.
How this newsletter gets better: Every link is tracked and every vote is counted. Sections that get a down vote get replaced. Topics that get an up vote stay. No personal data is stored — just click and vote counts per section. Your feedback shapes the next edition.
|
|
|