GoodData.AI launches AI Observability for enterprise agent tracing

GoodData.AI has announced AI Observability, a new capability extending its agentic analytics platform that gives enterprise teams layered visibility into how their AI systems behave, what they cost, and where they fail. The product launched on 23 September 2026 and is available in both GoodData.AI's managed cloud and on customer-controlled infrastructure.

The company says the core problem it is addressing is a gap between conventional monitoring tools and root-cause diagnosis. Existing dashboards, it argues, can surface changes in AI quality or token cost, but rarely expose the execution logic that produced a given response. AI Observability is positioned to close that gap by linking aggregate usage signals to the step-by-step traces behind individual interactions.

What the product does

The capability operates across three layers. At the organisational level, teams can track query volumes, active users, adoption rates across agents and skills, quality signals, and token consumption. At the interaction level, engineers and product teams can inspect the full execution path of a single response: which skills were activated, what knowledge and memory were retrieved, how model calls chained together, where failures occurred, and how long each step took. A third layer applies pattern analysis across conversation history to surface recurring issues and recommend changes to knowledge bases, semantic models or configuration settings.

Rosta Striz, Principal Product Manager at GoodData.AI, described the intent clearly: "Enterprise teams need more than a dashboard telling them that AI quality or cost changed. They need to understand what produced an answer: which skills ran, what knowledge and memory were used, where execution failed, and what they should change next. AI Observability connects those layers so teams can improve AI with evidence rather than intuition."

CEO and founder Roman Stanek added that observability "turns every interaction into evidence: evidence about adoption, quality, cost and the context behind the answer."

On the architecture side, the company says AI Observability runs as a managed workspace using interaction data the AI platform already generates, avoiding the need for a separate observability pipeline. Organisations can start with prebuilt dashboards and extend them on the same underlying data set.

Market context

The enterprise AI observability category is attracting growing attention as organisations move beyond pilots and into production deployments at scale. Several dedicated vendors, including established APM players extending their platforms as well as AI-native startups, are pursuing similar approaches to interaction tracing and quality monitoring. Hyperscalers are also building native observability features into their managed AI services, which creates competitive pressure on specialist platforms to differentiate on depth of integration and governance.

GoodData.AI's positioning is centred on its existing semantic layer and governance framework, which it calls the Agentic Serving Plane. By embedding observability inside that governed environment rather than treating it as a bolt-on logging tool, the company is arguing for a tighter coupling between data definitions, access controls and audit trails. That architecture may appeal particularly to regulated industries where AI audit history carries compliance weight.

Governance is a meaningful angle here. Requirements under the EU AI Act, which obliges providers and deployers of general-purpose AI systems to maintain logs of model usage and outputs, are phasing in across 2026 and 2027. Audit-trail capabilities of the kind GoodData.AI is describing map directly onto those obligations, as well as onto sector-specific frameworks such as SOC 2 and ISO 27001. The company has not published formal compliance certifications for the new module, which is a detail enterprise procurement teams will want confirmed.

GoodData.AI did not disclose pricing, customer names or early adoption metrics in the release. Named milestones and benchmark comparisons will be watched as the product matures.