Author: Microsoft Developer - Bewertung: 1x - Views:13
Observability pipelines were built for a world where engineers manually standardized logs, traces, and metrics. That model is breaking down as AI generates services and infrastructure faster than pipelines can keep up. Explore a new model where observability layers reason over telemetry and adapt as systems evolve. Agent-driven workflows continuously interpret and normalize data, raising new questions around guardrails, validation, and autonomy.
𝗦𝗲𝘀𝘀𝗶𝗼𝗻 𝗜𝗻𝗳𝗼𝗿𝗺𝗮𝘁𝗶𝗼𝗻:
This is one of many sessions from the Microsoft Build 2026 event. View even more sessions on-demand and learn about Microsoft Build at https://build.microsoft.com
ODSP933 | English (US) | Agents & apps
Pre-recorded | (200) Intermediate
#MSBuild
Chapters:
0:00 - Core concept: Agentic infrastructure requires agentic observability
00:02:08 - New analytical focus: Understanding agent reasoning rather than events
00:02:37 - Limits of deterministic observability with stochastic agent behavior
00:03:40 - Problems with tail-based sampling and scalability issues under high span counts
00:04:19 - Examples of silent failures in agent systems despite successful infrastructure
00:07:22 - Observability shaped around human cognition limitations
00:08:19 - Critique of MCP demos – external agents operate on partial context
00:09:47 - New observability model – move LLM up, push analysis down
00:12:39 - Critical role of data quality in agent reasoning and reliability