Reliability Economics

Archives
Log in
Subscribe
July 14, 2026

AI SREs hit production, now the hard part starts

The Ratio #9

Issue №9July 14, 2026

The Ratio

A weekly newsletter on reliability economics

The Number

3.9x

What happens to incident volume once reactive work crosses a quarter of a team's capacity

Teams spending more than 25% of capacity on reactive work report 120+ incidents a year. They hit that volume at 3.9x the rate of lower-reactive teams.

The Ratio Take:Prevention

The easy read is "reactive work causes incidents." Wrong. It's one fire seen from two windows. Incidents demand reactive capacity. Reactive capacity starves prevention. Starved prevention breeds more incidents. A feedback loop, not a cause-and-effect chain.

The 25% line isn't a workload threshold. It's where a team stops choosing its work and the incidents start choosing for it.

Below the line, you fund prevention. Above it, prevention is whatever survives the firefighting. Think of it like a hospital ER running at 25% over capacity: the staff isn't doing less medicine, they're doing less of the right medicine. Every minute triaging a new arrival is a minute not spent on the patient who needed surgery an hour ago. The building doesn't collapse. The outcomes just quietly get worse until someone dies.

Nothing survives above that line. Across the 99 services in the benchmark, zero teams above 25% reactive reported a declining incident trend year over year. Zero. Past 25% reactive, you don't have a reliability problem. You have a team that no longer decides what it works on.

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

This Week in Reliability

AI SRE Reaches Production

AI-driven SRE tooling is moving from promise to production deployment, forcing teams to confront the trust gap between autonomous remediation and operational accountability. The conversation is shifting from 'can AI do SRE work?' to 'how do we govern it in regulated environments and prevent it from creating new failure modes?'

Deep Reads

The 4-body problem of SRE: Why autonomous operations depend on context

CNCF Blog  ·  Primary evidence — identifies the deployment blocker

Senior SREs at a Bengaluru event identified a trust gap as the primary blocker to autonomous operations adoption. The piece argues context — business logic, organizational structure, tribal knowledge, and operational constraints — is the missing ingredient AI systems need before teams will trust them with production remediation.

The Ratio Take:The Ratio

This is the deployment wall every AI SRE vendor will hit: teams don't distrust the ML, they distrust the system's understanding of their specific operational reality. Until autonomous tools can ingest and reason over the four-body context (code, infra, process, politics), they'll stay in advisory mode while engineers do the actual work. Trust isn't a model accuracy problem — it's a context problem.

Can You Trust AI for SRE in Regulated Industries?

StackGen  ·  Regulatory constraint forcing maturity

StackGen addresses the compliance and auditability requirements for deploying AI SRE agents in finance, healthcare, and other regulated sectors. The article outlines traceability, explainability, and human-in-the-loop requirements that these industries demand before allowing autonomous remediation.

The Ratio Take:Prevention

Regulated industries are the forcing function that will determine whether AI SRE becomes infrastructure or stays a demo. If you can't produce an audit trail showing why the AI made a change and who approved the guardrails, you can't deploy it where the stakes are highest — and that's where the ROI is biggest. Compliance is the design constraint AI SRE needed.

How Aiden Automates Incident Remediation After an RCA

StackGen  ·  Vendor response — conservative deployment model

StackGen's Aiden agent automates post-incident remediation by executing fixes after root cause analysis, rather than during the live incident.

The Ratio Take:FirefightingPost-RCA automation is the low-risk entry point for autonomous SRE — the fire is out, humans have diagnosed it, and the AI just executes the boring toil. Automate the fix after you understand it.

Top 15 AI SRE Tools in 2026: The Complete Comparison

Sherlocks.ai  ·  Market maturity signal

Comparative analysis of 15 AI SRE platforms evaluated on causal reasoning, auto-remediation capability, Kubernetes support, and pricing transparency.

The Ratio Take:The RatioThe market has consolidated enough for buyers to demand apples-to-apples comparisons — a sign the category is maturing past the 'trust us, it's AI' phase. Buyers now have enough vendors to comparison-shop.

Why sandboxing your agent is not enough

CNCF Blog  ·  Security constraint on autonomous deployment

Discusses the agent-sandbox project and argues that sandboxing alone doesn't address the full security and reliability risk surface of AI agents operating in production environments.

The Ratio Take:PreventionContainment is necessary but insufficient — if the agent has enough context to be useful, it has enough access to cause harm, sandboxed or not. Sandboxing solves blast radius, not bad decisions.

The organizational iceberg: the invisible data breaking your AI agents

The New Stack  ·  Adjacent — data prerequisite for autonomy

Argues that AI agents fail in enterprises because they lack access to unstructured organizational knowledge — meeting notes, Slack context, tribal knowledge — that humans use to make decisions.

The Ratio Take:The RatioThis is the other half of the context problem: even if the agent has infra telemetry, it's missing the 'why we made this architectural choice in 2019' data that prevents it from confidently acting. Agents need the tribal knowledge, not just the logs.

ilert introduces dedicated incident management

ilert  ·  Human-AI workflow delineation

ilert ships dedicated incident management separate from status updates, adding incident channels, incident commander roles, severity levels, real-time timeline, and post-mortems as internal coordination features.

The Ratio Take:FirefightingHuman incident coordination tooling is maturing because AI is entering the space — vendors are drawing clearer lines between 'what the machine does' and 'what the IC coordinates.' Better human tooling is the AI deployment prerequisite.

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

The Crowd Favorite

  1. Roxanne — The Police — Flapping alert, infinite repeat. On-call learns to sleep through it. Silence the alarm or fix the signal. Never neither.
  2. Purple Rain — Prince — One power failure ends the show. Find your single point of failure in a game day, not during a revenue event.
  3. Kashmir - Remaster — Led Zeppelin — No timeout, no endpoint, eight minutes. A request without a deadline holds a thread until the pool is gone.
  4. Space Oddity - 2015 Remaster — David Bowie — Ground control loses the signal and still has to act. No observability means every incident timeline is reconstructed from memory.
  5. Like a Rolling Stone — Bob Dylan — Six minutes of unanswered questions. A postmortem without an assigned corrective action is the longest way to change nothing.

The Ratio Take:Prevention

Reduces future firefighting

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

The Challenger — Book Review

Becoming SRE — David N. Blank-Edelman (O'Reilly, 2024)

Three sections: the groundwork of SRE culture, how an individual becomes an SRE, how an organization builds and matures the function. The part that earns its place is the catalog of why SRE adoptions fail. Title-flipping sysadmins into "SREs." Dumping tier-3 support into the team. On-call with no engineering time. Funding the headcount but not the work.

Read them together and they're all one failure: trying to buy reliability at a discount.

The implication is blunt. Reliability isn't a job title you assign. It's a cost you fund. A relabeled ops team is not an SRE function. It's the same cost center with a new sign on the door, waiting to break on schedule.

Read if you're standing up SRE, or your "SRE team" is relabeled ops and you can't figure out why nothing changed. Skip if you're past the founding questions. This is a starting-line book, not a frontier one.

The Ratio Take:Prevention

Reliability is a price, not a prize

Should this replace the favorite?👍 Promote it👎 Not 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 👎 get replaced. Topics that get 👍 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 ship, but are they fighting or preventing fires? Older → AI SRE agents: faster war rooms or fewer fires?
Powered by Buttondown, the easiest way to start and grow your newsletter.