GoodData.AI launches Agentic Serving Plane for enterprise data
The new governed layer allows companies to serve consistent data, semantics, and security policies to both human dashboards and autonomous AI agents.
The short version
- The Agentic Serving Plane sits between enterprise data platforms and the applications or agents that consume them.
- It manages high-concurrency workloads where a single agent task can trigger hundreds of short-lived data operations.
- The system enforces row-level and field-level permissions to keep agent actions within explicit authorization boundaries.
- GoodData.AI is adding native local inference to run AI capabilities closer to governed data without third-party providers.
- The platform uses open technologies like Apache Iceberg and Apache Arrow to serve data from customer-controlled environments.
Why it matters
As companies move from human-led queries to autonomous agents, they need a way to ensure AI doesn't hallucinate or access unauthorized data. This tool provides a single control layer so that 'truth' remains consistent across every platform.
Bridging the gap for AI agents
GoodData.AI has launched the Agentic Serving Plane, a new category of enterprise data infrastructure. This governed execution layer is specifically designed to handle the unique data workloads created by autonomous AI agents. It sits between a company's core data platforms and the various tools that use that information.
Traditional enterprise data systems were built primarily for human users who perform occasional, predictable queries. In contrast, AI agents operate with a different logic, often generating hundreds of short-lived data operations to complete a single business task. The Agentic Serving Plane manages this continuous, high-concurrency demand without disrupting the underlying data platforms.
The platform aims to solve the problem of fragmented data stacks. As cloud warehouses and lakehouse platforms grow more complex, companies need a common layer to keep business definitions consistent. This new infrastructure ensures that regardless of where the data lives, the rules for accessing and interpreting it remain the same.
How the serving plane functions
The Agentic Serving Plane works by providing workload serving, business semantics, and contextual grounding. It uses open technologies like Apache Iceberg and Apache Arrow to connect with customer-controlled data environments. This allows companies to add governance and execution capabilities to their existing setups without moving their data to a new location.
One of the primary functions is managing complex sequences of queries. For example, if an agent is investigating a drop in revenue, it might jump between regions, product lines, and individual accounts. The serving plane manages this entire sequence as a governed workload, applying the correct business definitions and access policies to every single step of the process.
The architecture is built on technology GoodData.AI has used for years to help software companies embed analytics into their own products. By using this foundation, the Agentic Serving Plane can isolate machine workloads from human ones. It ensures that agent traffic does not slow down production warehouses or interfere with the performance of standard dashboards.
Governance and security for machines
A major challenge for enterprises is ensuring that AI agents stay within authorized boundaries. The Agentic Serving Plane provides delegated access control, which enforces permissions at the field and row levels. This means an agent can only see the specific data it is allowed to access, just like a human employee would be restricted by company policy.
The platform also focuses on governed business semantics. Teams can define metrics and business concepts once and reuse them across dashboards, applications, and AI experiences. This prevents a situation where an AI agent uses a different definition of 'profit' than the one used in the company's official financial reports. It ensures that agents use the same certified metrics as established reporting workflows.
To maintain oversight, the system includes AI observability and auditability features. These tools record how responses are generated, providing security and compliance teams with a clear view of the execution path. If a question falls outside of the defined semantic or policy boundaries, the system is designed to state that it cannot answer rather than producing a plausible but incorrect number.
Future of local inference
GoodData.AI plans to expand the Agentic Serving Plane by adding native inference capabilities. This rollout is expected to begin in the coming months and will allow certain AI capabilities to run closer to the governed enterprise data. By doing this, companies can avoid sending sensitive data to third-party model providers in the middle of a data path.
This move toward local inference is intended to reduce latency and lower the costs associated with external inference. It also keeps AI execution more closely aligned with the customer's specific business definitions and permissions. By keeping the processing closer to the data, the system maintains a tighter grip on how information is interpreted and used.
The broader strategy, as described by CEO Roman Stanek, is to build for a world where agents are constantly asking questions. The Agentic Serving Plane provides the necessary infrastructure to support these autonomous actions while maintaining human-level control. It allows enterprises to shorten development cycles by not having to rebuild the data foundation every time they launch a new AI application.
What to watch
- The rollout of local inference capabilities in the coming months.
- How other enterprise data platforms respond to the 'Agentic Serving Plane' category.
Try it
Check if your current AI tools use the same 'source of truth' for metrics as your human-facing dashboards.
Sources and links
First reported by Yahoo Finance UK. Written by the SI Agent News Desk from the reporting above.Written with AI from the source above. Every number is checked against the source. Spot a mistake? Corrections.
Liked it? Say why
Add a line for other readers.
No account needed. We never show your email. Comments with links, ads or abuse are held and not shown. Your comment shows after a quick check.