{"collection":{"slug":"observability","name":"Observability & Monitoring"},"digest":{"id":"cwd_4yb0V0l49wuWTN-MK_Mqr","weekStart":"2026-05-18","title":"Agent observability matures as Sentry and Datadog ship session replay and LLM debugging tools","intro":"Sentry launched Conversations in open beta, providing readable agent session logs, while Datadog published guides on AI guardrail placement and LLM-driven experiment tracking, and shipped fixes for Bedrock instrumentation across multiple APM versions.","body":"### Agent observability goes mainstream\n\nTwo of the biggest names in observability made agent observability more practical this week. **Sentry** opened [Conversations](/release/rel_ys9cB7RU2pz9AJDUZ2m9T-sentry-conversations-now-available-in-open-beta) to all users in open beta, giving developers a readable timeline of past agent sessions — messages and tool calls together, not scattered across separate logs. The feature fills a gap that anyone building on LLM frameworks has felt: understanding *why* an agent did what it did without stitching together traces and chat logs by hand.\n\nMeanwhile, **Datadog** published a guide comparing [guardrail placement strategies](/release/rel_bJtG1kpE5ZpdvT3UC0MHB-datadog-publishes-guide-to-ai-agent-guardrail-placement-strategies) across Amazon Bedrock Agents and self-orchestrated agents using AI Guard — timely given the indirect prompt injection demo used in the post. A separate guide demonstrated using [LLM Observability Experiments](/release/rel_jgTG1zQYL06IFi4Ol1qvv-datadog-publishes-guide-to-improving-sql-query-optimization-agent-accuracy-with) to track an AI agent's autonomous iteration cycle: one agent ran 23 experiments to improve a SQL query optimizer from 54% to 86% accuracy, tracking every hypothesis and failure along the way.\n\n### Datadog tightens Bedrock and Langchain instrumentation\n\nSeveral Datadog APM releases addressed gaps in LLM framework coverage. Python APM v4.8.5 fixed [Bedrock Converse guardContent blocks](/release/rel_nuj-UrhrbnnJAmsnFdXgK-datadog-apm-v4-8-5-fixes-bedrock-trace-data-loss-and-deprecates-inferred-spans) that had been dropping user input from traces when text was wrapped in guard responses, and deprecated `DD_TRACE_INFERRED_SPANS_ENABLED` in favor of `DD_TRACE_INFERRED_PROXY_SERVICES_ENABLED`. Python APM v4.8.6 fixed [Langchain Bedrock inference profile](/release/rel_j0n5nuJnp5eBBWAyDzGhk-datadog-apm-v4-8-6-fixes-langchain-bedrock-span-model-attribution) attribution — spans were being tagged with the profile's ARN instead of the actual LLM model. On the JavaScript side, APM v5.104.0 fixed an [unfinished CONNECT span](/release/rel_EDS43itzo7mZaNjhMydBI-apm-v5-104-0-fixes-undici-connect-span-and-metrics-delta-temporality) in undici instrumentation and resolved a delta temporality issue with OTLP counter exports.\n\nThe Datadog Agent v7.79.0 upgraded JMXFetch to 0.52.0, adding mappings for Generational Shenandoah GC, but macOS users need to note a breaking change: the Agent now installs as a [system-wide LaunchDaemon](/release/rel_JOtDcXtEQanOW5rVMIZLW-datadog-agent-7-79-0-upgrades-jmxfetch-and-switches-macos-to-system-wide) under a dedicated `_dd-agent` user instead of a per-user LaunchAgent, and existing installations must reinstall.\n\n### Sentry tightens security and Dash0 improves database debugging\n\nSentry CLI 3.4.3 disabled Xcode `Info.plist` preprocessing by default across `releases propose-version`, `send-event`, and `react-native xcode` commands — a [security hardening move](/release/rel_ZHpxdbU1RyqDTGXsG9W4A-sentry-cli-3-4-3-disables-xcode-info-plist-preprocessing-by-default-for-security) to prevent project-controlled compiler settings from reaching `cc` during release auto-discovery. The same change shipped on the 2.x line as version 2.58.6. Sentry also fixed a [data scrubbing regression](/release/rel_-hEZ0X5Gxwup0y4r_mm3F-sentry-fixes-data-scrubbing-regression-for-query-and-url-span-attributes) that prevented `http.query`, `url.query`, and `url.full` span attributes from being scrubbed.\n\nDash0 added a [dedicated database query widget](/release/rel_BP3Jl_FwDBf3Lvv-Ot7hF-database-query-widget-and-query-parameter-support) to the span sidebar, syntax-highlighted and with prepared statement parameters filled in, so you no longer have to hunt through the attributes tab for slow query debugging. The Dash0 Operator 0.141.0 added a `captureSqlQueryParameters` flag for finer-grained control.\n\n### Smaller releases worth knowing about\n\nDatadog published several guides covering practical ops workflows: [budget forecasting](/release/rel_obpaU9VfScnGYu_y46ltV-datadog-announces-budget-forecasting-for-cloud-cost-management) for Cloud Cost Management, a [monitor audit framework](/release/rel_mPC_RsJnvtQjBSuvctzJN-datadog-publishes-guide-to-auditing-and-cleaning-up-monitors) for reducing alert fatigue, [natural language queries](/release/rel_fyJ-LhG0IHikNQvqpUGsp-datadog-publishes-guide-to-exploring-metrics-with-natural-language-queries) for exploring metrics in plain English, [explain plan correlation](/release/rel_oG1Er_UsKF9U_xutsdKbo-datadog-database-monitoring-correlates-postgresql-explain-plan-nodes-to-sql) for PostgreSQL in Database Monitoring, and a guide to [reducing CVE noise with OpenVEX](/release/rel_xEhua0sHX6tfy0I4IMnKw-datadog-publishes-guide-to-reducing-cve-noise-with-openvex-assessments) assessments.","releaseIds":["rel_ys9cB7RU2pz9AJDUZ2m9T","rel_bJtG1kpE5ZpdvT3UC0MHB","rel_jgTG1zQYL06IFi4Ol1qvv","rel_nuj-UrhrbnnJAmsnFdXgK","rel_j0n5nuJnp5eBBWAyDzGhk","rel_EDS43itzo7mZaNjhMydBI","rel_JOtDcXtEQanOW5rVMIZLW","rel_ZHpxdbU1RyqDTGXsG9W4A","rel_-hEZ0X5Gxwup0y4r_mm3F","rel_BP3Jl_FwDBf3Lvv-Ot7hF","rel_obpaU9VfScnGYu_y46ltV","rel_mPC_RsJnvtQjBSuvctzJN","rel_fyJ-LhG0IHikNQvqpUGsp","rel_oG1Er_UsKF9U_xutsdKbo","rel_xEhua0sHX6tfy0I4IMnKw"],"releaseCount":27,"generatedAt":"2026-07-11T17:09:08.517Z","releases":[{"id":"rel_ys9cB7RU2pz9AJDUZ2m9T","title":"Sentry Conversations now available in open beta","path":"/release/rel_ys9cB7RU2pz9AJDUZ2m9T-conversations-feature-enters-open-beta","org":{"slug":"sentry","name":"Sentry"},"importance":null},{"id":"rel_bJtG1kpE5ZpdvT3UC0MHB","title":"Datadog publishes guide to AI agent guardrail placement strategies","path":"/release/rel_bJtG1kpE5ZpdvT3UC0MHB-ai-agent-guardrail-placement-guide-for-bedrock-and-self-orchestrated-agents","org":{"slug":"datadog","name":"Datadog"},"importance":null},{"id":"rel_jgTG1zQYL06IFi4Ol1qvv","title":"Datadog publishes guide to improving SQL query optimization agent accuracy with LLM Observability","path":"/release/rel_jgTG1zQYL06IFi4Ol1qvv-sql-query-optimization-agent-accuracy-improved-59-using-llm-observability","org":{"slug":"datadog","name":"Datadog"},"importance":null},{"id":"rel_nuj-UrhrbnnJAmsnFdXgK","title":"Datadog APM v4.8.5 fixes Bedrock trace data loss and deprecates inferred spans setting","path":"/release/rel_nuj-UrhrbnnJAmsnFdXgK-bedrock-guardcontent-no-longer-drops-user-input-inferred-spans-setting","org":{"slug":"datadog","name":"Datadog"},"importance":null},{"id":"rel_j0n5nuJnp5eBBWAyDzGhk","title":"Datadog APM v4.8.6 fixes Langchain Bedrock span model attribution","path":"/release/rel_j0n5nuJnp5eBBWAyDzGhk-langchain-bedrock-spans-now-use-correct-model-inference-profile-arn-bug-fixed","org":{"slug":"datadog","name":"Datadog"},"importance":null},{"id":"rel_EDS43itzo7mZaNjhMydBI","title":"APM v5.104.0 fixes undici CONNECT span and metrics delta temporality","path":"/release/rel_EDS43itzo7mZaNjhMydBI-undici-connect-span-no-longer-unfinished-metrics-delta-temporality-fixed","org":{"slug":"datadog","name":"Datadog"},"importance":null},{"id":"rel_JOtDcXtEQanOW5rVMIZLW","title":"Datadog Agent 7.79.0 upgrades JMXFetch and switches macOS to system-wide LaunchDaemon","path":"/release/rel_JOtDcXtEQanOW5rVMIZLW-macos-agent-now-system-wide-launchdaemon-jmxfetch-upgraded","org":{"slug":"datadog","name":"Datadog"},"importance":null},{"id":"rel_ZHpxdbU1RyqDTGXsG9W4A","title":"Sentry CLI 3.4.3 disables Xcode Info.plist preprocessing by default for security","path":"/release/rel_ZHpxdbU1RyqDTGXsG9W4A-xcode-info-plist-preprocessing-disabled-by-default","org":{"slug":"sentry","name":"Sentry"},"importance":null},{"id":"rel_-hEZ0X5Gxwup0y4r_mm3F","title":"Sentry fixes data scrubbing regression for query and URL span attributes","path":"/release/rel_-hEZ0X5Gxwup0y4r_mm3F-query-and-url-span-attributes-now-scrub-properly","org":{"slug":"sentry","name":"Sentry"},"importance":null},{"id":"rel_BP3Jl_FwDBf3Lvv-Ot7hF","title":"Database Query Widget and Query Parameter Support","path":"/release/rel_BP3Jl_FwDBf3Lvv-Ot7hF-database-query-widget-and-query-parameter-support","org":{"slug":"dash0","name":"Dash0"},"importance":null},{"id":"rel_obpaU9VfScnGYu_y46ltV","title":"Datadog announces budget forecasting for Cloud Cost Management","path":"/release/rel_obpaU9VfScnGYu_y46ltV-budget-forecasting-for-cloud-spend-cost-alerts-and-reports","org":{"slug":"datadog","name":"Datadog"},"importance":null},{"id":"rel_mPC_RsJnvtQjBSuvctzJN","title":"Datadog publishes guide to auditing and cleaning up monitors","path":"/release/rel_mPC_RsJnvtQjBSuvctzJN-guide-to-auditing-monitors-and-reducing-alert-fatigue","org":{"slug":"datadog","name":"Datadog"},"importance":null},{"id":"rel_fyJ-LhG0IHikNQvqpUGsp","title":"Datadog publishes guide to exploring metrics with Natural Language Queries","path":"/release/rel_fyJ-LhG0IHikNQvqpUGsp-metrics-natural-language-queries-guide-published","org":{"slug":"datadog","name":"Datadog"},"importance":null},{"id":"rel_oG1Er_UsKF9U_xutsdKbo","title":"Datadog Database Monitoring correlates PostgreSQL explain plan nodes to SQL clauses","path":"/release/rel_oG1Er_UsKF9U_xutsdKbo-explain-plan-nodes-now-map-to-sql-clauses-in-postgresql-queries","org":{"slug":"datadog","name":"Datadog"},"importance":null},{"id":"rel_xEhua0sHX6tfy0I4IMnKw","title":"Datadog publishes guide to reducing CVE noise with OpenVEX assessments","path":"/release/rel_xEhua0sHX6tfy0I4IMnKw-openvex-assessments-help-prioritize-exploitable-vulnerabilities","org":{"slug":"datadog","name":"Datadog"},"importance":null}]}}