BetaWeekly digests are a beta — we're trying something new. Feedback welcome.
Agentless tracing, SLOs, and custom policy rules arrive across observability tooling
September 14–20, 2026
Datadog's dd-trace-py gained a single-switch agentless mode and OpenTelemetry consistent-probability sampling, while Dash0 launched service level objectives with OpenSLO support. Datadog also let teams define their own IaC security rules in Rego.
Sampling and shipping without an Agent
The week's biggest shift landed in the Python tracer. dd-trace-py v4.15.0 introduced DD_AGENTLESS_ENABLED, one flag that sends telemetry, traces, Remote Configuration, Dynamic Instrumentation, crash reports, Test Optimization, and LLM Observability data straight to Datadog's intake instead of routing through a local Agent. It requires an API key and becomes the default for per-product agentless settings — a meaningful architectural simplification for serverless and short-lived workloads where running an Agent was always awkward. Alongside it came consistent probability sampling for distributed traces in mixed Datadog/OpenTelemetry environments, a discard field on DD_TRACE_SAMPLING_RULES, and new CPython GC collection and stop-the-world pause metrics.
The same sampling work appeared in the 4.15.0rc2 pre-release, which added OTLP-exported spans carrying OpenTelemetry consistent probability state when OTEL_TRACES_EXPORTER=otlp, plus a heap profiler Mem domain defaulting to true on Python 3.12+, Kafka span links, and fixes for crashtracking intake routing and a memory leak. If you run Python services in mixed-vendor tracing setups, this pair of releases is where to look first.
Two bug-fix follow-ups matter if you're already on the 4.x line: v4.14.1 fixed a ModuleNotFoundError that silently disabled all vLLM tracing and metrics on vLLM 0.14.0+, plus a startup-blocking AttributeError on google-adk 2.7.0+. And 4.15.1 stopped emitting gen_ai.* attributes on APM spans when LLM Observability is off — a small but welcome reduction in unwanted span payload.
Objectives, budgets, and policy as code
Dash0's Service Level Objectives launch is the week's other headline: point an SLO at any signal you already send, and you get remaining error budget and burn rate, with Agent0 able to draft definitions and one click converting an SLO into an alert. Definitions can live in the product or in a repository, applied via CLI or Terraform, serialized as OpenSLO — a sensible bet on an open format over a proprietary schema.
Policy-as-code showed up in two more places. Datadog's IaC Security custom Rego rules let security teams encode organization-specific requirements and run them across Ansible, CloudFormation, Dockerfile, Kubernetes, Terraform, and GitHub Actions, with an AI rule creator drafting from natural language and versioning via compare/restore. Separately, Observability Pipelines gained Microsoft Sentinel Packs that map firewall, VPN, and NDR logs from Palo Alto, Fortinet, Cisco ASA, Meraki, and ExtraHop into Sentinel table schemas before ingest — letting teams keep raw logs in cheap storage while sending only high-value events to the SIEM.
Error capture and incident response get closer to where work happens
Sentry's JavaScript SDKs continued their v11 ramp with 11.0.0-rc.0, and the stable 10.75.0 added Effect v4 error capture: Sentry.effectLayer now registers an ErrorReporter, so failures flowing through Effect.withErrorReporting, ErrorReporter.report, or HTTP/RPC boundaries are captured automatically with ignore/severity/attribute annotations honored. On the backend side, Sentry 26.9.0 continued detector-class refactoring, and Sentry Python 2.69.2 fixed a nasty infinite serializing loop in the Bottle integration and made functions_to_trace work with streaming spans.
Datadog pushed incident work into chat: Bits Chat in Slack lets responders start an investigation with @Datadog, dig into telemetry, trigger remediation like paging or a fix PR, and close out with a postmortem notebook — all without leaving the channel. The Agent 7.83.2 release carried an important correctness fix: the Data Plane pre-flight config was built from resolved config and could send metrics and your API key to datadoghq.com instead of your configured site. Worth upgrading for that alone, plus corrected Network Path filter parsing, configurable GPU static-metrics cadence, and OTEL_-prefixed tracer env vars in Single Step Instrumentation.
Smaller but useful: RUM Browser SDK v7.13.0 defaults service to the applicationId and captures wasm module build IDs on error events, and Axiom Go v0.37.1 fixed annotation list filter encoding. The remainder of the week was snapshot churn — Sentry CLI's 3.8.0 fixed one real thing, routing snapshot image uploads to the server-selected objectstore usecase, while the flurry of 3.7.1/3.8.1 snapshot builds and the Dash0 Operator 0.155.1 collector-crash fix are routine maintenance.
Releases covered17
- Datadog dd-trace-py v4.15.0 adds DD_AGENTLESS_ENABLED and OTel sampling support
- Datadog dd-trace-py 4.15.0rc2 adds discard sampling rule field and OTLP consistent probability state
- Datadog dd-trace-py v4.14.1 fixes vLLM 0.14.0 tracing and google-adk startup
- Datadog dd-trace-py 4.15.1 stops emitting gen_ai attributes when LLM Observability is off
- Datadog IaC Security adds custom Rego rules and AI rule creator
- Datadog Observability Pipelines adds Microsoft Sentinel Packs for security log mapping
- Datadog Bits Chat brings incident investigation and remediation into Slack
- Datadog Agent 7.83.2 fixes Data Plane pre-flight sending metrics to wrong site
- Datadog RUM Browser SDK v7.13.0 defaults service to applicationId and adds wasm debug IDs