The Number: 81% non-operational SLO programs

57 of 102 organizations (58%) report SLOs that exist but don't influence decisions. 26 organizations (26%) have no SLOs defined. Only 14 organizations (14%) say SLOs influence some decisions, and 2 (2%) report SLOs as central to release and investment choices. The benchmark's SLO-to-ratio cross-correlation shows median ratios of 1.52 (no SLOs), 2.56 (non-actionable SLOs), and 5.14 (SLOs influence some decisions).

THE RATIO TAKE

Treat non-actionable SLOs as a leading indicator of measurement theater, not organizational maturity. An SLO that triggers no decision is just a metric with extra steps — it consumes engineering time while leaving the reliability economics question ('are we spending in the right place?') completely unanswered. The 3.4x ratio gap between 'no SLOs' and 'SLOs that drive some decisions' suggests the measurement itself has a forcing function: once you can see the number, you're harder to ignore in a budget meeting.

81% of organizations are measuring reliability commitments they never act on — SLO theater at industrial scale.

Did this land? 👍 More like this👎 Switch it up

This Week in Reliability

AI Agents Need Governance

AI coding and operational agents are shipping faster than the guardrails around them. This week shows the reliability economics gap: agents promise prevention at scale, but without visibility, policy enforcement, and human-in-the-loop controls, they shift unmeasured risk onto production systems.

Deep Reads

1. Who's monitoring the agents?

The New Stack · Primary evidence

AI agents now operate autonomously across infrastructure and code, making decisions that affect production systems without real-time human oversight. Traditional monitoring tools weren't built to track agent behavior, decision paths, or failure modes.

THE RATIO TAKE

Agents are the new blast radius nobody's measuring. You wouldn't let a junior engineer deploy without logs—why would you let an LLM-powered agent modify infrastructure without the same visibility? This is the reliability economics conversation your CFO doesn't know you need to have yet.

Agents are production actors without production observability.

2. Introducing Prempti: Policy and visibility for AI coding agents

CNCF Blog · Solution response

Prempti is a new CNCF sandbox project that adds policy enforcement and audit trails to AI coding agents like Claude Code. It tracks what agents read, write, and execute, and can block actions that violate defined policies before they reach production.

THE RATIO TAKE

This is what happens when the community realizes agents ship code faster than humans can review it. Policy-as-code for agents isn't paranoia—it's the same shift-left thinking that gave us CI/CD gates, now applied to autonomous actors. Prevention spend, finally catching up to agentic reality.

Policy gates for agents, before they commit.

Signals

3. CI wasn't built for coding agents. Here's what comes next. — The New Stack · Infrastructure gap

Traditional CI/CD pipelines assume human-authored commits with predictable review cycles. AI coding agents generate dozens of PRs per hour, overwhelming code review processes and breaking assumptions about test coverage and rollback safety.

THE RATIO TAKE

Your CI budget line just became an agent tax—more runs, faster merges, and nobody's certain what passed review versus what got rubberstamped by fatigue.

Agent velocity breaks human review economics.

4. Why Kubernetes policy enforcement happens too late—and what to do about it — CNCF Blog · Adjacent prevention pattern

Most Kubernetes policy tools enforce rules at admission time, after manifests are written and PRs are open. Shift-left policy validation—checking policies during authoring or CI—prevents violating configs from ever reaching the cluster.

THE RATIO TAKE

If agents are writing your K8s manifests, runtime policy is already too late—you need gates at generation time, before the agent commits.

Policy must move earlier than agent output.

5. Knowledge Graphs: The Missing Brain Your SRE Agents Desperately Need — StackGen · Data infrastructure for agents

SRE agents lack context about service dependencies, ownership boundaries, and historical incident patterns. Knowledge graphs provide structured metadata that helps agents make decisions that align with organizational topology and past failure modes.

THE RATIO TAKE

Agents without context are expensive guesswork—knowledge graphs turn them into informed actors who know which service to restart and which VP to wake up.

Context graphs turn guessing agents into informed ones.

6. Annie meets pup: turning intent into audited Datadog runbooks — Anyshift.io · Vendor implementation

Anyshift's Annie agent translates plain-English operational intent into executable, auditable runbooks that integrate with Datadog's pup CLI. Each action is logged and reviewable, with approval gates before execution.

THE RATIO TAKE

Auditability is the price of admission for agentic ops—if you can't replay what the agent did during the incident, you can't learn from it or defend it in the postmortem.

Runbook agents with receipts, not magic.

7. Three ways operational debt will break your AI strategy, and how to recover — The New Stack · Prerequisite warning

Organizations deploying AI agents on top of poorly documented, brittle infrastructure amplify existing operational debt. Agents can't fix what they can't understand—and they'll automate around gaps instead of surfacing them.

THE RATIO TAKE

Agents inherit your tech debt and scale it—if your runbooks are stale or your CMDB is fiction, agents will codify those lies into production at machine speed.

Agents multiply the debt you already carry.

Did these land? 👍 More like this👎 Switch it up

The Crowd Favorite

1. The Sound of Silence - Simon & Garfunkel — Green dashboards, broken journeys. 81% of SLO programs never alert on what users actually feel.
2. Fix You — Coldplay — MTTR is what you track after an outage. Post-incident reviews are the mechanism that actually lowers it next time.
3. Should I Stay or Should I Go - Remastered — The Clash — Circuit breakers make the dependency-boundary call before cascade propagation makes it for you.
4. Losing My Religion — R.E.M. — When 81% of SLO programs are non-operational, the dashboard exists but no decision flows from it.
5. Comfortably Numb — Pink Floyd — Alert fatigue desensitizes engineers to real signal. Noise normalization is how teams miss the unannounced P0.

Both sides of the curve at once

The playlist — keep or kill? 👍 Keep the crown👎 Dethrone it

The Challenger — Career Spotlight

A mid-market logistics SaaS, 180 engineers, started at 87% non-operational SLOs. Defined, never reviewed, no error budget ever consumed. Over 14 months they did three things. They assigned named SLO owners per service, not per team. They embedded 15-minute error-budget reviews into sprint retrospectives. They deleted 58% of SLOs that had no consumer and no on-call consequence. Ending ratio: 31% non-operational. Still above the sector benchmark. But every surviving SLO now gates a release decision.

Fewer SLOs enforced beats many SLOs ignored

Should this replace the favorite? 👍 Promote it👎 Not ready

---

Take the Reliability Economics Benchmark →

---

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.