Token costs, prompt traces, and audit logs: enterprise AI observability today
Submitted to Observability Summit Europe 2026 on . Rejected .
- Level
- Intermediate
- Length
- Session Presentation (25-minutes; 1-2 speakers)
- Track
- End-User Case Studies
Abstract, as submitted
As enterprise platform teams take on AI workloads, the observability stack they had for, now, "old" applications doesn't quite cover what they need for agents and MCP servers. These are things like: token cost narrowed to applications and users, prompt/response tracing, error budgets that include LLM latency tails, and all the enterprise-y annoyances like audit logging for governance and compliance. This session reports on what developers and platform teams are actually instrumenting today, drawn from conversations with platform engineers at enterprises deploying Spring AI, MCP gateways, and agent runtimes. It will go over what works well, what wasn't needed, and a toolkit for starting to put observability in place focused on larger organizations.
This is the text that went into the CFP form, kept as submitted - not a later rewrite of it. Talks get retitled and reworked between submission and stage, so what was actually delivered may differ.