Reliability Economics

Archives
Log in
Subscribe
September 10, 2026

AI SRE isn't replacing toil—it's exposing maturity gaps

The Ratio #17

Issue 17September 10, 2026

The Ratio

A weekly newsletter on reliability economics

 

The Number

4.26 vs 1.64

Financial Services organizations have a median reliability ratio 2.6x higher than Technology companies, meaning the most regulated industry in this benchmark is proportionally more under-invested in reliability prevention than Tech.

Financial Services: median reliability ratio of 4.26 (N=32). Technology: 1.64 (N=57). In Financial Services, 24 of 32 organizations are classified as under-investing. Only 2 are over-investing. In Technology, 27 of 57 are over-investing, nearly half the industry.

Neutral

Regulation doesn't move prevention budgets. Not in this data. Financial Services is the most consistently under-invested industry in the benchmark. Almost nobody in that group over-corrects, so nothing pulls the median down. Technology looks better on paper, but only because so many Tech companies are over-spending. That's not calibration. That's the opposite problem.

It's like a hospital where every ward tracks infection rates but nobody funds sterilization equipment. The reporting exists. The compliance exists. The actual investment in preventing the thing you're reporting on does not.

The most regulated industry in this benchmark spends proportionally less on preventing failures than the least regulated one.
Did this land?
More like thisSwitch it up
 

This Week in Reliability

AI SRE Maturity Reality Check

The gap between AI SRE hype and practice is coming into focus. Teams are finding that maturity frameworks, agent context limits, and verification loops matter more than model capabilities—shifting budget conversations from 'deploy AI' to 'integrate AI where you already have discipline.'

 

Deep Reads

SRE Maturity Assessment Framework for Teams (2026)

SquareOps  ·  Primary evidence

Framework defines five SRE maturity levels from ad-hoc operations to predictive capabilities. Level 3-4 is where dedicated SRE investment shows highest ROI—SLOs drive decisions, toil is tracked, automation is in place. The gap to Level 5 is predictive capabilities like anomaly detection and chaos engineering.

Neutral

This is the maturity floor that AI SRE vendors assume you already have. If you're still firefighting without SLOs or tracking toil, an AI agent won't fix that—it'll amplify the chaos. The ROI insight here is blunt: get to Level 3 manually before you buy AI automation, or you're paying for a copilot in a plane with no instruments.

AI SRE assumes Level 3 maturity—most teams are still at Level 1.

AI is changing the reliability game for SREs

Dynatrace  ·  Expectations vs. reality data

Survey shows over half of SREs expected meaningful reductions in manual work, faster incident resolution, and lower operational costs from AI. Reality is more restrained than expectations, similar to the platform engineering hype cycle.

Neutral

The gap between vendor promises and SRE reality is narrowing faster than the hype cycle. Teams anticipated AI would cut toil and MTTR—instead they're learning it redistributes work from firefighting to prompt engineering and verification. Budget-wise, you're not cutting headcount; you're retraining it.

AI didn't cut SRE headcount—it shifted work to verification and prompts.

Give the Agent a Fair Environment: Director of Software Engineering on LLM-Powered RCA

StackGen  ·  Agent environment prerequisites

Interview with engineering director on what LLM-powered root cause analysis requires to work in production—emphasizes agent environment setup and context quality.

Prevention

If the agent can't see your telemetry, topology, or deployment history, it's guessing—and you're paying for expensive hallucinations during an incident.

Agents need complete context, not better models, to stop guessing.

Researchers found that 1 in 5 MCP access policies came back broken or missing

The New Stack  ·  Verification loop evidence

Research on Model Context Protocol implementations found 20% of access policies returned broken or missing during vibe-coding workflows—a verification gap in agent-driven development.

Firefighting

One in five MCP policies failing means verification loops are the new bottleneck—AI agents ship faster, break policies more, and cost you incident recovery time if you skip the check layer.

20% policy failure rate: agents ship fast, verification catches up slow.

Kubernetes disaster recovery: Guidance from three reproducible failure scenarios

CNCF Blog  ·  Maturity baseline: DR discipline

Document describes three Kubernetes failure scenarios that separate having backups from being able to recover. All scenarios are reproducible on a laptop from the provided lab repository.

Prevention

Backups don't equal recovery—these reproducible failure labs expose the gap AI agents can't close for you, because disaster recovery is about runbooks and muscle memory, not model inference.

Having backups isn't recovery—practicing the restore is.

How to verify your Azure Application Gateway is zone-redundant

Gremlin  ·  Verification discipline prerequisite

Explains how to confirm whether an Azure Application Gateway is zone-redundant or zonal using Azure CLI and Resource Graph. Notes that a gateway surviving a zone failure doesn't guarantee downstream services will survive.

Prevention

Verification tooling for resilience configurations is the unsexy prevention work AI agents aren't replacing—this is the checklist discipline that determines whether your DR plan is real or aspirational.

Gateway survives zone failure ≠ your services survive zone failure.

How to monitor Cypress tests with Grafana Cloud

Grafana Labs  ·  Adjacent: observability maturity

Guide to monitoring Cypress test suite performance and failure patterns in Grafana Cloud. Helps identify whether slowdowns or failures are spec-level, test-level, or suite-wide trends versus one-off job issues.

Prevention

Test observability is the canary in the coal mine for production reliability—if your CI suite is slowing or flaking, AI agents won't catch it unless you instrument the feedback loop first.

Flaky CI tests are early signals for production incidents.
Did these land?
More like thisSwitch it up
 

The Crowd Favorite

  1. Seven Nation Army — The White Stripes — Two people, no bassist, no redundancy. Single-instance deployments hold until the traffic spike they were never sized for.

  2. Once in a Lifetime - 2005 Remaster — Talking Heads — "How did I get here?" is what your on-call asks when there's no deployment audit trail.

  3. Under the Bridge — Red Hot Chili Peppers — Blast radius expands exactly as far as your architecture permits. Isolation is not a runtime setting.

  4. Bitter Sweet Symphony - Remastered 2016 — The Verve — A sampled loop they didn't own cost them the masters. Undocumented third-party dependencies surface at the worst possible moment.

  5. Don't You Want Me — The Human League — Synthetic checks that only test the happy path miss everything your actual users hit.

Prevention

Know the failure mode before production does
The playlist — keep or kill?
Keep the crownDethrone it
 

The Challenger — Industry Shift

The prevention-to-firefighting ratio is splitting in two. Top-quartile teams benchmark at 4.26:1. The median sits at 1.64:1. That 2.6× spread tracks compounding, not noise.

Early prevention investment reduces incident load. Fewer incidents free capacity for more prevention. Same mechanism as compound interest. Late movers don't just lag. They fall further behind each cycle.

This isn't a forecast. The benchmark shows it now.

The 4.26 teams started compounding early. The 1.64 teams are paying the interest.

Neutral

The ratio gap compounds — early movers pull away
Should this replace the favorite?
Promote itNot ready

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.

Don't miss what's next. Subscribe to Reliability Economics:
← Newer AI agents fail 60% of tasks—who pays for the cleanup? Older → AI SRE agents meet production—and production wins
Powered by Buttondown, the easiest way to start and grow your newsletter.