01
AI spend is invisible: token and provider cost do not attach to intents, teams, or SLO burn.
Case 21
Cost and Token Observability
Tokens as an operational meter: ingest usage, attribute owners, budget burn, and alert before finance becomes your first observability backend.
01
How it works
Watch the meter — before the invoice.
Cost and Token Observability by Andrey Lesnikov — AI spend as a live ops signal with owners, budgets, and caps.
-
01
Meter every event
Gateway and provider usage become telemetry — not a monthly surprise CSV.
-
02
Attribute an owner
Team, service, and env tags make spend discussable with a human who can change it.
-
03
Set the budget
Thresholds turn raw tokens into a fuel story ops can steer.
-
04
Alert and cap
Burn alerts and soft caps beat discovering the problem in a finance meeting.
02
Evidence
From problem to proof.
How the system is framed, operated, and proven — not a slide deck summary.
02
You need request routing and policy first — Cloud-Native AI Gateway and AI Infra Control Plane own those contracts.
03
Cloud-Native AI Gateway and AI Infra Control Plane emit the signals this case makes actionable. OpenTelemetry Observability Mesh correlates cost with traces.
Constraints
Token attribution, cloud tags, model pricing, request volume, budget alerts, and developer-readable reports.
03
Related
Nearby systems.
AI Infra Control Plane
Private AI platforms need shared governance across identity, policy, audit, cost, and SLO signals — not scattered service configs
Private AI with explicit platform contracts
Cloud-Native AI Gateway
AI usage needs routing, policy, budget awareness, and provider resilience
One boundary: rate, budget, fallback, OTel
OpenTelemetry Observability Mesh
Metrics, logs, and traces often exist separately, making incidents slower and ownership unclear
Request path explainable end-to-end
Available for meaningful infrastructure conversations