{"id":2601,"date":"2026-06-03T15:44:01","date_gmt":"2026-06-03T22:44:01","guid":{"rendered":"https:\/\/devblogs.microsoft.com\/foundry\/?p=2601"},"modified":"2026-06-03T15:52:59","modified_gmt":"2026-06-03T22:52:59","slug":"build-2026-from-observability-to-roi-for-ai-agents-on-any-framework","status":"publish","type":"post","link":"https:\/\/devblogs.microsoft.com\/foundry\/build-2026-from-observability-to-roi-for-ai-agents-on-any-framework\/","title":{"rendered":"Build 2026: From observability to ROI for AI agents on any framework\u00a0"},"content":{"rendered":"

9 min read \u00b7 June 3, 2026 \u00b7 Sebastian Kohlmeier<\/span><\/i>\u00a0<\/span><\/p>\n

\u00a0<\/span><\/p>\n

Shipping an AI agent is the easy part. Keeping it\u00a0accurate, safe, and accountable in production is where teams get stuck. Agents are\u00a0non-deterministic. Their behavior shifts as models update, tools change, and traffic patterns\u00a0evolve\u00a0and most of that drift happens silently, long after the demo.\u00a0<\/span>End-to-end observability\u00a0<\/span><\/b>covering the full development lifecycle is how you close that gap: See every step an agent takes, evaluate quality and safety against criteria you define,\u00a0optimize\u00a0what\u00a0isn\u2019t\u00a0working, and prove the business value of what is.<\/span>\u00a0<\/span><\/p>\n

This spring we hit a major milestone \u2014\u00a0<\/span>tracing and evaluations in Microsoft Foundry reached general availability\u00a0<\/span><\/b>with hosted agents coming soon.\u00a0Every team building on Foundry can rely on them in production today. At Build 2026,\u00a0we\u2019re\u00a0extending that foundation to\u00a0<\/span>any agent framework, any deployment target, and the full Agent DevOps loop<\/span><\/b>\u00a0\u2014 from the first inference call to the ROI dashboard your CFO will ask about.<\/span>\u00a0<\/span><\/p>\n

This post walks through the key capabilities\u00a0we\u2019re\u00a0landing in BRK252 \u2014 From observability to ROI for AI agents on any framework: I<\/span>nteroperability, context-specific rubric evaluators that are multi-turn enabled, code-first observability, optimization, and business ROI.<\/span><\/b>\u00a0<\/span><\/p>\n

\"Agent<\/a><\/span><\/p>\n

\u00a0<\/span>What\u2019s\u00a0new at Build 2026<\/span><\/b>\u00a0<\/span><\/h2>\n

All capabilities are part of Microsoft Foundry observability. Preview status reflects state at Build.<\/span>\u00a0<\/span><\/p>\n

\"What<\/a><\/p>\n

\u00a0<\/span>Why observability is the foundation for trustworthy agents<\/span><\/b>\u00a0<\/span><\/h2>\n

Traditional software is deterministic: Same input, same output, same code path. Agents\u00a0aren\u2019t. The same prompt can take three different tool routes today and a fourth one tomorrow when the model or prompt is updated. That makes the standard reliability stack \u2014 logs, metrics, error rates \u2014 insufficient on its own. You also need to know what the agent\u00a0<\/span>decided<\/span><\/i>, whether that decision was\u00a0<\/span>good<\/span><\/i>, and whether\u00a0it\u2019s\u00a0<\/span>getting better or worse<\/span><\/i>\u00a0over time.<\/span>\u00a0<\/span><\/p>\n