Datadog, Inc.

Datadog details how agent spend depends on loop turns, tool calls, and re-sent context, and how Agent Observability ties estimated cost to applications and traces, flagging savings like uncached prefixes and repeated full histories.

Datadog engineers show how trace analysis of their own agent exposed cost drivers, such as search tools returning more content than needed that is then resent on every turn, and investigated tool-call failures before preparing a code fix.

Datadog shares how its Bits Investigation agent went from $25 and 11 minutes per alert to about $4 and 2 minutes, with a 44% quality gain on an internal benchmark. Covers the hypothesis-driven multi-agent design behind the gains.

Datadog argues agent workloads should start on serverless, since sessions spend most of their time idle between steps, and suggests using production data to decide which workloads need dedicated capacity.

Datadog unveils Scout 1.0, an RL-trained model for production incident investigation behind its AI SRE. It reports matching frontier models at several times lower estimated inference cost. Available in closed preview with a technical report.

Datadog reproduces two self-replicating prompt injection attacks from OpenAI's report against a test agent and shows how evaluating tool outputs and tool calls catches the planted instruction and its copying into replies or skill files.

Datadog's Agent Observability adds Online Experiments, which compare agent variants such as prompt, model, or tool changes on live production traffic. Results combine eval quality with latency, cost, token usage, and error metrics, with traces for debugging.