Stacks/SRE / incident response

Datadog Bits Investigation

Datadog's Bits Investigation is a production incident-investigation agent evaluated on labeled real incidents from hundreds of Datadog teams. The architecture is a hypothesis-and-test loop over live telemetry, designed so the agent follows causal evidence instead of summarizing a pile of tool results. The company reports the agent has helped teams decrease time to resolution by up to 95%.

Use case

An AI SRE agent that automatically investigates incidents and monitor alerts, reasons over live telemetry, and produces audit-ready root-cause analyses

Category

SRE / incident response

Team size

not stated

Orchestrator / framework

Bits Investigation follows a human-SRE style loop: it formulates hypotheses about the root cause, validates or rejects them with targeted telemetry queries, and repeats, breaking complex hypotheses into sub-hypotheses and digging deeper when evidence supports a lead. The authors contrast this with an earlier approach that issued many tool calls and then asked an LLM to summarize, which let noisy signals override a correct finding.

Memory

not stated

Evals

Datadog collected and labeled real incidents from hundreds of internal teams and built a benchmark of test scenarios. The agent is scored against each scenario's ground-truth label, and that benchmark is used to regularly measure and improve performance.

Observability

The agent queries the customer's Datadog telemetry — logs, traces, and metrics — as its investigation tools. The published post is about the agent itself, not a separate third-party observability product.

Monthly cost

not stated

Models

not stated

Observability tools

Datadog

Other tools

not stated

Source

https://www.datadoghq.com/blog/building-bits-ai-sre/

Daniel Shan and Tristan Ratchford · Jan 11, 2026