| The RatioA weekly newsletter on reliability economics | |
|
The Number
7.05 vs 1.94 (SLOs that act vs none)
Organizations where SLOs influence real decisions carry a failure cost ratio 3.6x higher than organizations with no SLOs at all.
The benchmark has three SLO maturity tiers. Organizations with no SLOs: median reliability ratio of 1.94 (N=27). SLOs exist but don't influence decisions: 2.24 (N=58). SLOs that influence some decisions: 7.05 (N=14). The group measuring most carefully has the widest gap between failure costs and prevention spend. Not the narrowest.
The Ratio Take:The Ratio
You'd expect SLO maturity to track with better economics. It doesn't. The organizations with the highest ratios, the ones where every dollar of prevention sits against seven dollars of failure cost, are the same ones who built SLOs that actually change behavior. They didn't adopt SLOs because things were going well. They adopted them because the bleeding forced it. Think of it like a cardiac monitor in an ICU: the patient hooked up to the most sophisticated equipment isn't the healthiest person in the hospital. They're the sickest. The SLO didn't close the gap. The gap demanded the SLO. Adopting one without shifting where the money goes is paperwork. The organizations measuring failures most carefully are the ones bleeding the most. The SLO is the smoke alarm, not the sprinkler.
|
|
This Week in Reliability
AI SRE Agents Hit Production
Multiple vendors shipped autonomous AI agents for SRE workflows this week, moving from incident triage prototypes to production remediation. The reliability economics question: are teams buying prevention capabilities or just faster firefighting with a chatbot?
|
StackGen · Primary evidence — vendor production deployment StackGen describes their AI triage system for on-call engineers, focusing on what patterns work in production incident response versus what fails under pressure. The Ratio Take:Firefighting The honest headline here is 'triage' — faster root cause detection, not prevention. If your AI agent arrives after the pager goes off, you're still paying the full cost of the incident minus fifteen minutes of human confusion. The ROI case depends entirely on whether those saved minutes prevent escalation or customer churn. Triage speed is not prevention — it's discounted firefighting. |
GitHub (agamm) · Market landscape — shows adoption breadth Community-maintained list of 100+ AI tools for SRE workflows including autonomous incident response, intelligent observability, self-healing infrastructure, and chaos engineering agents. The Ratio Take:The Ratio This list proves the market exists but reveals the taxonomy problem: 'autonomous incident response' and 'self-healing infrastructure' sit in the same category despite opposite budget implications. Self-healing that triggers before the alert is prevention; autonomous response after the page is cleanup cost reduction. Until vendors segment by when-in-the-timeline they act, buyers will keep conflating MTTR tools with MTTF tools. 100 tools, two economics — nobody's labeling which is which. |
Steadybit · Vendor response — chaos engineering angle Steadybit launched SteadyBuddy, an AI companion for orchestrating chaos engineering experiments and reliability testing workflows. The Ratio Take:PreventionChaos engineering with an AI co-pilot is preventive spend if it runs before deployment, reactive if it diagnoses why prod broke. AI chaos tooling lives on both sides of the curve. |
Anyshift.io · Supporting infrastructure — agent context layer Anyshift released a Graph API that lets teams query production state relationships — rank a flapping monitor, trace a bad deploy to its failures — designed to give agents structured context instead of dashboard screenshots. The Ratio Take:The RatioGraph APIs for agents are infrastructure for both sides: trace-back-to-commit is preventive, rank-the-flapping-monitor is reactive incident triage. Context infrastructure serves firefighting and prevention equally. |
Grafana Labs · Adjacent signal — observability agent tooling Grafana Assistant now queries across 30+ data sources to answer questions without manual tab-switching during incident investigation. The Ratio Take:FirefightingMulti-source query assistants compress war-room time but don't shrink the number of wars. Faster answers during outages — not fewer outages. |
The New Stack · Market structure — platform convergence Amazon, Microsoft, and Google are standardizing enterprise agent architectures, creating portability contracts that let organizations move AI workflows across clouds without rewriting integration logic. The Ratio Take:The RatioPortability reduces lock-in risk but doesn't tell you whether the portable agent prevents incidents or just triages them faster across three clouds instead of one. Standardized agent plumbing — economics TBD. |
SRE Weekly · Community synthesis — emerging operational risks SRE Weekly curated stories this week covering agent scale, blind spots in distributed job runners, and live state visibility across throughput-heavy environments. The Ratio Take:The RatioCommunity curation flags the visibility gap — agent scale creates new blind spots even as agents promise to close old ones. More agents, more places to run, more to monitor. |
|
The Challenger — Ask Me Anything
"Our team has SLOs defined, but nobody reviews them in sprint ceremonies. Does that count as 'having SLOs' for benchmarking purposes?"
No.
The 7.05 ratio belongs to teams where SLOs trigger actual decisions. Rollbacks. Feature freezes. Escalations. Defined-but-unreviewed SLOs sit closer to the 1.94 baseline. Here is the test: name the last decision your SLO changed. If you can't, you have documentation. Not a reliability contract. The gap between 1.94 and 7.05 isn't SLO existence. It's SLO consequence.
The Ratio Take:The Ratio
The data to make the case
|
|
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 👎 get replaced. Topics that get 👍 stay. No personal data is stored — just click and vote counts per section. Your feedback shapes the next edition.
|
|
|