Sources

OTel: agent-observability blog (Mar 2025), GenAI follow-up (2026), semantic-conventions-genai repo state (Jun-Jul 2026); FinOps Foundation: FinOps for AI (updated Feb 17 2026), State of FinOps 2026, FOCUS 1.3; Tokenomics Foundation (CIO coverage); Harness State of AI in FinOps 2026 [vendor]; Datadog State of AI Engineering 2026 [vendor telemetry]; Entra sponsor roles (Microsoft Learn) [vendor]; Workday ASOR (Feb 2025 announce; GA Feb 2026) [vendor]; Agent 365 billing policies (Nov 18 2025) [vendor]; Agentforce Flex Credits (May 15 2025) + Digital Wallet [vendor]; consolidation: Dynatrace-Arize (Aug 13 2026), ClickHouse-Langfuse (Jan 16 2026), Cisco-Galileo (Apr-May 2026), CoreWeave-W&B (2025), OpenAI-Statsig (Sep 2 2025), LangChain Series B (Oct 20 2025), Braintrust Series B (Feb 17 2026); judge reliability: arXiv 2606.19544 (Jun 2026), arXiv 2508.06225 (Aug 2025), GoDaddy calibration (Nov 24 2025); rollout discipline: arXiv 2606.08867 (100M-user framework, Jun 2026), canary pattern catalogs; Bedrock Guardrails CloudWatch metrics [vendor]; AgentCore harness caps and pricing [vendor]; LiteLLM budget docs [vendor OSS]; incidents: OpenClaw $1.3M (self-reported, May 2026, medium confidence), $47K A2A loop (vendor blog, LOW confidence); outcome pricing: Intercom, Zendesk [vendor]; vendor-embedded agent visibility: Microsoft Graph Copilot usage APIs GA (Oct 2025) with documented gaps, Anthropic Admin Usage & Cost + Enterprise Analytics APIs [vendor]. Excluded: fabricated Langfuse funding story (content-farm; on the do-not-cite list). Re-verify by Q4 2026: OTel stability; Dynatrace-Arize close and Phoenix fate; Workday budgeting depth; vendor GA states.

Added 2026-10-14 (accessed 12 to 14 October 2026): the $1.3 million, 603-billion-token bill was a deliberate fleet of about 100 coding-agent instances over 30 days, not a loop, per the team's own account as reported by Tom's Hardware https://www.tomshardware.com/tech-industry/artificial-intelligence/openclaw-creator-burns-through-1-3-million-in-openai-api-tokens-in-a-single-month [secondary]; the $47K agent loop analysis, which notes the weekly figures in the original account do not add up https://dev.to/waxell/the-47000-agent-loop-why-token-budget-alerts-arent-budget-enforcement-389i [secondary].