Reliability Economics

Archives
Log in
Subscribe
October 13, 2026

SRE rewrites its playbook, what changed in 10 years

The Ratio #22

Issue 22October 13, 2026

The Ratio

A weekly newsletter on reliability economics

 

The Number

7.04 vs 1.92 (AI vs no-AI)

Organizations using AI for reliability operations have a median reliability ratio 3.7x higher than those that don't, meaning their unplanned failure costs dwarf prevention spend by a wider margin.

AI adopters have a median reliability ratio of 7.04. Non-adopters sit at 1.92. The four most common AI capabilities, triage, log summarization, runbooks, anomaly detection, all activate after something breaks.

Firefighting

Teams using AI for reliability aren't preventing more failures. They're running more expensive failure patterns. The AI gets deployed as a faster ambulance, not a safer road. This is the equivalent of an ER that buys better defibrillators instead of funding cardiology, the saves look impressive but the underlying disease rate hasn't moved.

Teams using AI for reliability have failure costs 3.7x higher relative to prevention spend than teams that don't.
Did this land?
More like thisSwitch it up
 

This Week in Reliability

SRE's Second Edition

Google published the second edition of the SRE book after ten years, codifying how modern reliability engineering has evolved from static game days to continuous resilience, from dedicated SRE teams to embedded practices, and from manual capacity planning to cloud-native auto-scaling. The timing matters: as AI agents reshape development velocity, the industry is re-examining what 'reliable enough' means and how to budget for it.

 

Deep Reads

SRE book second edition

Google Cloud Blog  ·  Primary evidence — the authoritative update

Google released the second edition of the SRE book, updating three core areas: reliability testing now emphasizes continuous resilience checks and live canary verification over scheduled game days; capacity management shifted from manual provisioning to dynamic auto-scaling and GPU/TPU optimization; organizational models expanded beyond dedicated SRE teams to embedded and consulting patterns.

Neutral

This is the industry admitting that the 2016 playbook doesn't match 2026 infrastructure — cloud-native tooling, AI workloads, and agentic deployments demand continuous verification instead of quarterly chaos exercises. The shift from 'SRE as a team' to 'SRE as a practice' reflects economic reality: most companies can't afford dedicated reliability teams, so the discipline has to distribute or die.

SRE went from a team you hire to a practice you embed.

A Blueprint for a Better Society: Operationalising Team Topologies in the UK Public Sector

Team Topologies  ·  Organizational evolution parallel

The UK public sector (NHS, UKHSA, Home Office) applied Team Topologies to pandemic vaccination booking, testing systems, and immigration platforms, contrasting pre-pandemic failures with breakthroughs achieved through stream-aligned teams and platform abstraction. The case study documents how organizational structure determines system reliability in high-stakes, resource-constrained environments.

Prevention

Public sector deployments are the hardest reliability test case: legacy constraints, regulatory overhead, life-or-death stakes, and zero room for ZIRP-era over-engineering. If Team Topologies worked there, the model has teeth — and it reinforces the SRE second edition's point that you can't bolt reliability onto bad structure. The NHS didn't need more SREs; it needed better boundaries.

Reliability failed until they fixed the org chart first.

What Is Site Reliability Engineering (SRE)?

Traversal  ·  Economic foundation — why SRE stuck

Traversal's SRE explainer cites Uptime Institute data showing 57% of major outages cost over $100,000, and frames error budgets as the shared language for reliability tradeoffs between engineering and business stakeholders.

Neutral

The error budget framing is the single best artifact from SRE 1.0 — it turns 'how reliable should we be?' from a religious argument into a spreadsheet.

Error budgets made reliability a budget line item.

What is Site Reliability Engineering? - SRE Explained

AWS  ·  Vendor adoption signal

AWS published an SRE explainer covering core concepts (error budgets, SLOs, toil reduction) and how SRE practices integrate with AWS services like CloudWatch, Auto Scaling, and Systems Manager.

Prevention

AWS entering the SRE education space signals that cloud providers now see reliability practice as table stakes — they'll sell you the primitives, but you still have to do the thinking.

Cloud vendors now teach SRE because they can't automate judgment.

Why AI development creates a reliability blind spot for humans, and what to do about it

Gremlin Blog  ·  The AI-era gap

Gremlin argues AI coding tools create reliability blind spots because developers don't understand generated code as deeply as hand-written code, and proposes chaos engineering as the verification layer.

Prevention

This is the 2026 addendum to the SRE book Google should have included — when agents write the code, humans lose contextual understanding of failure modes, and the old 'review-then-deploy' gate doesn't catch what you can't see.

AI-generated code breaks the mental model SRE assumed you had.

Announcing Gremlin Foresight AI

Gremlin Blog  ·  Vendor response — tooling catches up

Gremlin launched Foresight AI, which analyzes systems for failure modes, recommends fixes, and verifies resilience at AI speed, positioning it as the chaos engineering response to agentic development velocity.

Prevention

If the first edition of SRE was 'automate toil,' the second edition is 'automate verification' — Gremlin is betting that AI-generated infrastructure needs AI-speed reliability checks or the feedback loop collapses.

Chaos engineering finally moves at agent deployment speed.

If your team is happy, are you doing a good job?

Charity Majors  ·  Cultural evolution — the org side of SRE

Charity Majors and Dr. Cat Hicks discuss ZIRP-era management anti-patterns, brilliant jerks, and how psychological safety lost the plot when teams mistook comfort for effectiveness.

Neutral

The cultural half of SRE 2.0 — distributed reliability practices only work if teams can admit failure without fear, but 'psychological safety' turned into 'never give hard feedback,' which is how you end up with happy teams shipping unreliable systems.

Psychological safety died when comfort replaced accountability.
Did these land?
More like thisSwitch it up
 

The Crowd Favorite

  1. Alive — Daft Punk — A liveness probe that never fails is just a warmth check. Unhealthy pods stay in rotation until the SLI craters.
  2. Riders on the Storm — The Doors — Chaos injection without scope isn't engineering. Bound the blast radius before the experiment, or the incident teaches the lesson for you.
  3. Sound and Vision - 2017 Remaster — David Bowie — Head-based sampling throws away slow requests. Tail-latency SLO breaches vanish from traces before anyone opens the post-mortem.
  4. Money — Pink Floyd — Toil you don't quantify is invisible to budget owners. Measure it in engineer-hours or it never funds its own automation.
  5. Yesterday - Remastered 2009 — The Beatles — A post-mortem started 72 hours after containment is archaeology. Begin the timeline during the incident, not after.

Prevention

Five failure modes, zero round numbers.
The playlist — keep or kill?
Keep the crownDethrone it
 

The Challenger — Contrarian Take

Popular belief: The 7.04 AI-team ratio proves AI is the reliability lever.

It doesn't.

Teams adopting AI-assisted SRE in 2024 are disproportionately the ones that already had SLO maturity, structured error budgets, and dedicated headcount. The tool follows the process. The process drives the ratio.

In professional sports, GPS load-monitoring correlates with fewer injuries. But those teams also have sports scientists and periodization programs. Strip those away and the sensor is inert.

You might be wrong: cohorts with no prior SLO maturity that still show ratio gains post-adoption would isolate AI as causal. That data doesn't exist yet.

Neutral

The ratio tracks maturity, not the tool.
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 SRE agents are acting in prod — prevention or speedup? Older → AI made us faster. Now it's breaking us.
Powered by Buttondown, the easiest way to start and grow your newsletter.