Live debugging in production, and agentic answers in Slack

Debugging without a deploy

The week's biggest shift is that production debugging no longer requires shipping new code. Datadog's Live Debugger lets you place logpoints on running services and capture variable values, method arguments, and execution context without restarting or redeploying anything — and Bits AI can read the linked source, place non-breaking logpoints, pull variable snapshots, and suggest fixes from live production data. It's the same instinct behind Datadog's Bitbucket Cloud source code integration, which collapses the hop between an error or slow method and the repository, file, and commit behind it. Datadog also explained how Bits proves a query rewrite is actually faster, using synthetic data and measured execution time rather than vibes.

Dash0 took the agentic angle in the other direction: every alert in Slack now carries a Fix this in thread button that asks Agent0 to explain the failing check, why it fired, and whether it needs fixing, with the answer landing in the thread where the team is already talking. A new GitLab connector lets Agent0 read diffs, issues, and pipeline results during an investigation and open merge requests with deep links back to the telemetry that prompted them, acting through each member's own account. And the CLI now deletes assets based on Git history via an experimental apply --since option, so removing a check from a file removes it from Dash0 too.

Logs get safer and quieter by default

Sentry flipped a meaningful default: logs are now enabled by default in the JavaScript SDK, though nothing is captured unless you call the Sentry.logger.* APIs or wire up a forwarding integration, and enableLogs: false opts out. The Python SDK is converging on the same model from the other side — auto-collected logs keep flowing when enable_logs=True, but the option is deprecated and per-integration control moves to capture_sentry_logs, since the logger API no longer depends on it. JavaScript also stopped AI integrations from double-reporting errors that OpenAI, Anthropic, Google GenAI, LangChain, or LangGraph hand back to your code: the span is still marked errored, but your own error handling decides whether Sentry hears about it. An 11.0.0 alpha with a work-in-progress migration guide is available for anyone who wants to look ahead.

Sensitive data got similar attention. Datadog added build-time privacy allowlists for RUM action names, preserving readable names for static text while masking runtime-generated values, and the Browser SDK shipped a Feature Flags tab plus debug IDs on log events. Across the tracers, dd-trace-js and its v6 line added AI Guard redaction, agentless Bunyan and Pino log submission for test optimization, and a new recordException API, while dd-trace-py made agentless OpenFeature configuration delivery the default.

Platform hygiene and the long tail

Two operational notes worth calendar entries. Sentry will swap its TLS certificate authorities to Google Trust Services and Let's Encrypt in February 2027 — transparent for most, but anyone pinning Sentry certificates needs to act. The Datadog Agent patched a Dynamic Instrumentation proxy that was silently dropping debugger data and removing workload-specific configs, plus a system-probe upgrade regression. Sentry CLI added dSYM support to IPA uploads, and the rest of its week was snapshot builds. Datadog's remaining output was mostly reading rather than shipping: AI agent security lessons from monitoring its own agents, adapting Golden Paths for non-human platform users, cost evidence for federal FY27 renewals, its own $1M/month AI cost optimization, and a Tines and Observability Pipelines workflow for threat response.