§ 08 · Domain mastery

Observability & SRE

Signal that survives 3 AM. Postmortems that end in code changes, not blame.

SLOs and error budgets your product team helps set, dashboards a CFO can read, on-call rotations with clear ownership, and blameless postmortems that actually change the codebase. Cost and reliability treated as the same conversation, not siloed OKRs.

Talk to a partner →All services

§ I

Pillars

SLOs tied to user journeys

Availability of endpoints is not availability of your product. SLOs written against what the user is trying to do, with error budgets that inform release cadence.

OpenTelemetry as the standard

One vendor-neutral instrumentation across languages. When you change backends, you're not rewriting instrumentation for three months.

On-call that respects sleep

Rotations sized for the incident load, runbooks that pass a real drill, and a paging discipline that treats false positives as bugs.

Blameless postmortems, actioned

The template is the easy part. The habit that turns action items into shipped code is what separates practice from theater.

Cost + reliability as one lens

The cheapest way to make a system reliable is often to remove things. Cost signals in the reliability conversation, and vice versa.

Chaos and drills, not hope

Game days for the top failure modes, DR drills that actually flip traffic, load tests before Black Friday — not after.

§ II

Domain vocabulary

The words that separate insiders from readers.

Error budget	The allowed unreliability in an SLO. Teams that treat it as a resource ship faster than those that don't.
SLI vs. SLO	The metric vs. the target. Confusing them makes 'we hit our SLO' meaningless.
Burn rate alert	Alerting on trajectory to breach, not on breach itself. The difference between paging early and paging useless.
MTTR	Mean time to recover. Widely quoted, easily gamed. Insiders pair it with mean time between incidents.
Postmortem culture	Not the doc — the follow-through. Teams that ship the action items outperform teams that write beautiful docs.
Golden signal	Latency, traffic, errors, saturation. Naming which one you're alerting on is a mark of clarity.
OTel collector	The vendor-neutral pipeline for telemetry. Where you win back leverage from any single observability vendor.
Blast radius	Shared with cloud vocab; the SRE version is about traffic routing and cell isolation.

§ III

Open problems we help with

01
SLOs for AI-heavy features
'Correctness' isn't a percentile of a p95. What replaces it as a shipping gate is unresolved.
12 mo
02
Vendor-neutral observability past the demo
OpenTelemetry works. The operational cost of running your own pipeline vs. paying a vendor is where teams still hesitate.
18 mo
03
Cost-aware reliability decisions
Adding a region for 9s vs. cutting spend by removing them. Framing the tradeoff is often harder than the engineering.
ongoing
04
On-call recruitment and retention
The staffing story more than the technology story. Rotations that don't burn people out are still an open problem.
ongoing
05
Postmortem-to-fix throughput
Action items decay. The systems that keep them alive — and closed — are what separate mature teams from performative ones.
ongoing

§ IV

What we ship

01SLOs tied to user journeys with error budgets.
02OpenTelemetry instrumentation across your services.
03Runbooks tested in game days.
04On-call rotations sized to the real load.
05A postmortem template plus the habit to close its action items.
06A cost + reliability dashboard the whole org reads.

NEXT STEP

How does this look in your case?

Talk to a partner →

Observability & SRE

Pillars

SLOs tied to user journeys

OpenTelemetry as the standard

On-call that respects sleep

Blameless postmortems, actioned

Cost + reliability as one lens

Chaos and drills, not hope

Domain vocabulary

Open problems we help with

SLOs for AI-heavy features

Vendor-neutral observability past the demo

Cost-aware reliability decisions

On-call recruitment and retention

Postmortem-to-fix throughput

What we ship