Debugging goes live: Datadog and Sentry ship new runtime tools

Runtime debugging without the redeploy

The biggest story this week is Datadog's Live Debugger, which lets teams place logpoints on running services to capture variable values, method arguments, and execution context — no restart or redeploy required. It integrates with Bits AI to analyze source code, suggest fixes, and even place non-breaking logpoints from production telemetry. For anyone who has ever burned hours adding temporary logging to reproduce a flaky issue, this closes the loop between observation and action.

Sentry also pushed runtime improvements. The Sentry JavaScript 10.72.0 release stops all AI integrations (OpenAI, Anthropic, Google GenAI, LangChain, LangGraph) from double-reporting errors that the framework already propagates to your code — the span still gets marked as errored, but your own error handling decides whether to send an event. That removes a whole class of noise for LLM-heavy apps. The same release adds rpcTracePropagationBindings for Cloudflare Workers RPC trace propagation, with the older enableRpcTracePropagation now deprecated.

On the Python side, Sentry Python 2.68.1 clarifies enable_logs behavior before its removal in the next major: auto-collected logs still send when the flag is True, but you can now control per-integration capture via capture_sentry_logs. It also fixes gen_ai tool-call gating across Anthropic, OpenAI, and LangChain to key on outputs rather than inputs, preventing unwanted span creation.

Logs, feature flags, and agent-driven workflows

Sentry made logs a default in JavaScript 10.71.0, flipping enableLogs to true so Sentry.logger.* APIs work out of the box — though nothing is captured unless you call them or add a log-forwarding integration. The release also deprecates scope.clear() and fixes a UTF-8 truncation issue in React Router SSR streams. Meanwhile, Sentry's 11.0.0-alpha.2 previews the migration path from 10.x with bundle size tables included.

Datadog’s Browser SDK v7.10.0 brings a Feature Flags tab to all sites with per-site overrides, attaches debug IDs to log events, and lets session replay clear its string table. A companion blog post explains a build-time allowlist approach for RUM action names: the mask-unless-allowlisted privacy level paired with a build plugin extracts static strings from compiled artifacts, keeping readable names while masking runtime-generated or sensitive values.

Dash0’s GitLab connector is notable for the AI-agent trend: Agent0 can now read diffs, issues, and pipeline results behind your telemetry, then open merge requests with deep links back to the triggering metrics. It operates through each member’s own GitLab account, so it only reaches what you can reach.

Infrastructure and forward-looking changes

Datadog’s Agent 7.82.3 fixes an upgrade regression in system-probe and stops the Dynamic Instrumentation proxy from dropping debugger payloads when logs_enabled is unset. The dd-trace-py v4.14.0 makes agentless OpenFeature configuration the default (load straight from Datadog over HTTPS), adds AI Guard redaction, and supports LLM Observability experiment queries plus remove_tag/remove_metric span methods. The dd-trace-js v5.124.0 similarly adds AI Guard sensitive-data redaction and agentless log submission for Bunyan and Pino in test optimization.

One to plan around: Sentry will switch its TLS certificate authority from DigiCert to Google Trust Services and Let’s Encrypt in February 2027. If you pin Sentry’s certificates, you’ll need to remove pinning or add the new CAs; users on older Android or Java platforms must update to trust ISRG Root X1.

Several smaller releases round out the week: Sentry CLI 3.7.0 adds dSYM support to IPA uploads, and a series of internal snapshot builds (3.7.1-snapshot, 3.7.1-snapshot.20260828, 3.6.3-snapshot.20260827, 3.6.3-snapshot) landed for internal testing. Axiom’s Go SDK v0.36.1 is a dependency bump with OpenTelemetry v1.46.0.

Datadog also published two cost-focused posts — how it saves over $1M monthly by optimizing AI usage and a federal FY27 renewal guide — plus an AI agent security retrospective and a Golden Paths piece for non-human platform users. Worth reading if you’re building agentic workflows, but they’re strategic rather than tactical.