Datadog Bits AI alternative: the infrastructure context layer

Datadog Bits AI became generally available in December 2025 and has been tested across 2,000+ customer environments. For teams already invested in Datadog (logs, metrics, traces, APM, synthetics), it delivers cross-signal correlation in natural language without onboarding a new vendor. The integration is excellent because it has to be: Bits AI lives inside the same UI engineers were already using.

Two structural limits define the box. First, Bits AI sees only the data Datadog sees: any signal that is not in your Datadog ingest is invisible. Tools like CloudTrail, GitHub, Vault, ArgoCD, or another logging stack are not in scope unless you also pay to ingest them into Datadog. Second, billing is per-investigation on top of existing Datadog costs, which becomes a real number at scale. There is no independent infrastructure graph; the model of the production stack is whatever you have already configured in Datadog's APM service map.

Cloudflare's November 2025 cascading outage is the canonical case for why a vendor-bound model is risky. Monitoring detected the failures within minutes. Tracing them back to the original internal change took hours, because the dependency chain crossed signals and tools no single observability vendor saw end-to-end.

Incident-time questions are causal and temporal, not statistical. "Which deploy caused this?" and "what changed across cloud, Kubernetes, and code between 14:00 and 15:00?" require an indexed model of the production stack itself, not a model derived from telemetry. Telemetry tells you something is anomalous. It does not tell you which IAM update, Helm rollout, Terraform apply, or commit produced the anomaly.

That model is what Anyshift builds. The platform connects to AWS, GCP, Azure, Kubernetes, and your git provider directly, with no in-cluster agents, no telemetry-pipeline ingestion, and no per-investigation billing. The graph spans every infrastructure layer regardless of whether you ship its data to Datadog, Splunk, Honeycomb, Grafana, or your own backend, which is the same architectural argument we walk through here.

The methodology behind Annie, Anyshift's investigation agent, is documented in Agentic Context Engineering, a paper authored with researchers at Stanford and SambaNova Systems and accepted at ICLR 2026. The same technique has been live in production since October 2025, where it has cut root-cause-analysis time by 30% on real customer incidents.

Most Anyshift customers stay on Datadog for observability and add Anyshift for investigation. Bits AI keeps doing what it does well, summarizing anomalies and answering questions inside the Datadog UI. Anyshift adds a vendor-agnostic infrastructure graph that spans signals Datadog never sees, and posts a root-cause report into Slack the moment an incident fires.

The net effect is more decisive RCA without expanding Datadog's already-large footprint, no extra per-investigation Bits AI charges, and visibility into the cross-tool causes (IAM, secrets rotation, IaC drift, schema migrations) that telemetry-bound AIs miss by construction. Teams running both report MTTR reductions of 85% or more on the incidents where the cause is upstream of any observability backend.