From black box to glass box

The Dynatrace 3rd-generation platform continues to evolve, helping teams see, explain, and trust the AI shaping their business.

Key insights

• Confidence where it counts: Real time, contextual data turns AI from a black box into decisions you can see, explain, and trust.
• Outcomes, not features: Teams reduce risk, control cost, and improve answer quality by grounding AI in better data and context.
• Role-based paths: Leaders, platform and SRE, AI engineering and ML ops, application developers, and security each have clear ways to act.
• Dynatrace 3rd generation foundation: Grail unifies telemetry, Smartscape maps live dependencies, and Davis AI turns insight into explainable actions.

The shift is underway

Your AI helped ship a feature customers love. Then a bad answer slips through, support tickets rise, and no one can explain why. Was it a model change, a prompt tweak, or missing context from an upstream service? When AI behaves like a black box, you cannot manage risk, cost, or trust.

You’re not alone. Budgets and expectations reflect the push to make AI observable and accountable. In our 2025 State of Observability data, 70% of organizations increased observability spend in the last year, and 75% expect to increase it again. Leaders often see the biggest returns from optimizing model configurations, detecting anomalies in model outputs, and automating remediation. Most believe that AI decisions still require a human check because proof matters.

At Dynatrace, we believe the difference is the data. Grail keeps observability, security, and business telemetry together in real-time, connected context. Smartscape maintains a real-time map of services, dependencies, and releases. Davis AI reasons over this context to explain cause and effect and suggest what to do next. Together they make AI explainable and governable, so teams can move faster without losing trust.

Here’s how your teams can build confidence in AI

Choose models with real context

Run evaluations on real workloads, not synthetic tests. Compare latency, cost per answer, and relevancy side by side, then use prompt traces to see why outputs differ. Your team picks the right AI model for the job with evidence, not guesswork. Use canary rollouts to validate the winning configuration on a small slice of traffic, then scale with confidence while a policy records who approved the change.

• This showcases: AI model evaluation and versioning across providers, prompt tracing and debugging, cost and performance insights tied to real services.

Explain every decision on demand

Capture inputs, prompts, context, and model versions so you can show how a decision was made. Export an auditable bundle, attach it to your review, and keep evidence with the workflow. Approvals move faster because proof is built in. Evidence travels with the workflow so leaders can audit decisions without meetings.

• This showcases: Exportable audit evidence, lineage across prompts and context, review-ready artifacts for governance.

Spot and prevent drift

Watch for quality changes after releases or upstream shifts in agentic AI systems that span multiple models. Because data lives together in Grail and dependencies are mapped in Smartscape, Davis AI detects drift, explains the likely root cause, and recommends next steps. Your team rolls back or tunes with confidence and documents the outcome.

• This showcases: End-to-end drift detection and explanation across services and models, causal analysis tied to releases, and actionable root cause context.

Move fast with guardrails

Set policy rules that pause risky paths for humans and promote safe improvements automatically. When a policy is triggered, the flow pauses for approval. When targets are met, changes move forward on their own. Speed and accountability rise together.

• This showcases: Policy-driven automation, human-in-the-loop controls, measurable targets for quality and cost.

One place to work, together

Platform and SRE, AI engineering and ML ops, developers, and security see the same facts and act in the same space. A built-in experience brings multi-cloud and multi-model views together with alerts, traces, and reviews. Teams focus on outcomes because data and context are already aligned, and agentic tracing makes complex multi-LLM paths explainable.

• This showcases: Unified, role-aware experience in the Dynatrace 3rd generation platform, powered by Grail for data, Smartscape for live topology and dependencies, and Davis AI for reasoning.

How it works

Dynatrace brings key, differentiated capabilities together, so AI becomes observable, governable, and improvable. Grail stores and relates telemetry with context as a single source of truth. Smartscape maps real-time topology and dependencies. Davis AI analyzes that knowledge to explain issues, correlate cause and effect, and drive or recommend actions. These explainable insights show up in built-in experiences so teams can act quickly in one place, with the same context everyone trusts. Together, they give you a glass-box view of AI across your environment.

Use cases you can try today

Platform Engineering and SRE

Use progressive delivery to your advantage and run a canary release with two model versions on a real service. Compare key metrics and data points, such as latency and error rates, then set an automated rollback if quality drifts beyond the threshold.

AI engineering and ML ops

A/B test two providers for a summarization workload. Measure token usage, cost per answer, and relevancy. Use prompt debugging to tune instructions, then promote the winning configuration.

Application developers

Trace a problematic response from UI to model call. Inspect the prompt, context, and dependencies, then commit a configuration change to fix a latency regression.

Security and compliance

Export an audit bundle for a high-risk workflow. Attach it to your review ticket and record a human approval step. Automate export on schedule.

The market signal

Leaders are investing to make AI observable and governable. From our 2025 State of Observability report, 70% increased observability budgets last year, with 75% of those surveyed planning to increase again next year. Teams expect the biggest return from optimizing model configurations, detecting anomalies in model outputs, and automated remediation. Ninety-eight percent already use AI to support security compliance.

What this means for your organization

For leaders

• Improve decision quality with model evaluations grounded in a real workload context.
• Reduce risk with audit trails that satisfy compliance.
• Control spend by measuring cost next to performance.

For Platform Engineering and SRE teams

• Standardize model rollouts with versioning and repeatable A/B tests.
• Tie prompt traces to services, releases, and alerts to cut MTTR.
• Automate safe rollbacks or configuration changes with policy guardrails.

For AI engineering and ML ops

• Compare models and versions on latency, cost per answer, and relevancy.
• Use prompt debugging to tune instructions and ground responses in context.
• Track drift and trigger-governed actions when quality drops.

For application developers

• See how AI components behave in production next to service health.
• Debug prompts without leaving the release context.
• Ship changes with data on performance and cost impact.

For security and compliance

• Export signed audit trails that show inputs, outputs, models, and context.
• Prove who did what, when, and why for policy and regulatory reviews.
• Route risky outputs for human approval and record the decision.

Gain confidence where it counts

Learn how to get started