Evaluating Dynatrace for autonomous operations? Ask Where Its Intelligence Can Run

Dynatrace is an observability platform whose AI layer, Dynatrace Intelligence (formerly Davis AI), reasons over the telemetry Dynatrace itself has ingested into Grail and mapped in Smartscape. NeuBird AI is a standalone Production Ops Agent that reasons across 50+ tools without requiring your telemetry to land in any one platform first. It prevents incidents with scheduled sweeps, resolves them with a verifiable evidence chain, and runs as SaaS, private VPC, on-prem, hybrid, or fully air-gapped.

No. Dynatrace has rebranded Davis AI as Dynatrace Intelligence, and Davis CoPilot is now Dynatrace Assist. The documentation still lives at the older davis-ai URL, so both names circulate, but Dynatrace Intelligence is the current product name. If a vendor comparison you are reading still says Davis AI throughout, it was written against an older version of the platform and is worth re-checking on other details too.

Its reasoning is grounded in Grail, the Dynatrace data lakehouse, and Smartscape, the dependency graph Dynatrace builds from its own instrumentation. Dynatrace ingests OpenTelemetry and supports 715+ technologies, so a great deal of data can be brought in, but the AI reasons over what has been ingested and modeled: anything outside Grail is outside the analysis. That is a real architectural choice with real benefits, and it is also why getting full value means routing more of your estate through Dynatrace. NeuBird AI reads your tools where they already are, including Dynatrace itself, with no re-instrumentation and no requirement to centralize telemetry first.

Yes, within its own automation stack. Dynatrace markets Dynatrace Intelligence as an agentic operations system with built-in SRE, Developer, and Security agents, and its agentic workflows can propose and run approved actions under policy guardrails, with Request Approval tasks keeping a human in the loop. Verify current maturity during evaluation, since parts of the agentic surface have carried Preview labels and scope changes often. The more important question for most buyers is where those actions can reach: the automation is native to the Dynatrace platform, while NeuBird AI acts through whichever tools already run your environment.

This is the sharpest difference. Dynatrace Managed is their on-premises deployment and remains supported, but it does not support Grail-based innovations, which by Dynatrace’s own account includes AppEngine, AutomationEngine, Workflows, and native agentic and generative AI, because those depend on the SaaS processing architecture. Dynatrace publishes a SaaS Upgrade Assistant to move Managed customers to SaaS. In practice, an on-premises Dynatrace customer gets classic problem detection, not the agentic operations system Dynatrace now markets. NeuBird AI runs the same agent with the same capabilities on-prem, hybrid, and fully air-gapped as it does in SaaS.

The concern behind it is legitimate: an agent that guesses confidently is worse than no agent. Where the framing gets slippery is treating determinism as the only route to trust. What an operator actually needs is the ability to check the work. NeuBird AI ships every verdict with a verifiable evidence chain: the signals consulted, the correlations drawn, and the reasoning that produced the conclusion, so an engineer can audit a specific investigation rather than trusting the category of math behind it. Determinism also has a cost, since a system that reasons only over a graph it built itself is bounded by that graph, and novel incidents in messy estates rarely respect those boundaries. Ask both vendors to show their reasoning on your own incident and judge the output.

Dynatrace uses a Dynatrace Platform Subscription: one annual commitment drawn down against a published rate card, with Full-Stack Monitoring at $58 per month per 8 GiB host, Infrastructure Monitoring at $29 per host per month, log ingest at $0.20 per GiB, and separate lines for traces, metrics, events, RUM, synthetics, and Application Security. The rate card is genuinely transparent and unlimited seats are included. The structural point is that the bill scales with hosts instrumented and data ingested, so getting more AI value means ingesting more, and the AI sits on top of that spend. NeuBird AI charges credits tied to work performed: one credit per investigation run, always-on triage included, unlimited alerts, no seat fees, no ingest fees, and no observability bill required underneath.

No, and most teams should not. Dynatrace is one of the strongest instrumentation and topology products available, and NeuBird AI reads it through the Dynatrace Environment API v2 with a read-only token, pulling problems, events, entities, Smartscape topology, traces, and logs. The common pattern is Dynatrace for deep full-stack telemetry and NeuBird AI as the agent that reasons across Dynatrace plus the 50+ other tools in your estate. See the Dynatrace integration for how the two fit together.

Dynatrace does not publish a single headline accuracy figure for Dynatrace Intelligence; its argument is that deterministic causal AI is accurate by construction because it traces the dependency graph rather than inferring correlation. NeuBird AI publishes 94% investigation accuracy with root cause in 2 to 5 minutes. Neither number should be taken on faith. Both are best tested the same way: run each system against an incident you have already solved, and see whether the evidence it produces would have convinced you at the time.