Synthesized builds Test Data Agent to validate AI agents with production-like data
Synthesized has announced its Test Data Agent, a new agentic infrastructure capability being developed to create and provision the realistic data, business context, and system states enterprises need to validate AI agents safely before production deployment.
The Test Data Agent integrates with agent development, evaluation, testing, and orchestration frameworks, providing production-faithful environments for determining whether an agent can reliably complete real business processes, not merely perform well in controlled demonstrations.
This latest innovation from Synthesized, advances the company’s mission to provide AI-native test infrastructure for enterprises operating complex and highly regulated technology environments. The Test Data Agent targets purpose-built support for complex SAP estates, helping organizations validate agents across critical finance, procurement, supply-chain, and operational workflows, as well as SAP ECC-to-S/4HANA transformation and testing programs.
Closing the enterprise agent-validation gap
Enterprises can build AI agents faster than they can prove those agents are ready to operate across mission-critical workflows.
Model evaluations and demonstration datasets can show whether an agent produces a plausible response. They cannot establish whether it will make the correct decision when confronted with the data, permissions, application states, business rules, and cross-system dependencies found in a real enterprise environment.
That distinction matters. An agent that performs well in a controlled test may still fail when it encounters missing records, unusual transactions, conflicting instructions, access restrictions or complex dependencies across multiple systems.
Yet recreating these conditions is difficult. Enterprise test environments are often stale, incomplete or disconnected from production complexity, while privacy, security, sovereignty and regulatory requirements limit the use of sensitive production data in development and testing.
Synthesized is developing the Test Data Agent as the production-faithful validation layer for enterprise AI agents. Starting with a business scenario or testing objective, it identifies and provisions the data, relationships, and system states required to test how an agent performs under realistic operating conditions.
The Test Data Agent lets teams:
- Identify the business entities, records, relationships, and system states required for a test.
- Generate, mask or subset production-representative data.
- Preserve referential integrity, statistical characteristics and business rules across interconnected systems.
- Create repeatable happy-path, exception, failure and adversarial scenarios.
- Refresh validation environments as applications and enterprise data evolve.
- Trigger provisioning through REST APIs and CI/CD pipelines so downstream testing and evaluation frameworks can call it.
- Operate within on-premises, private-cloud and hybrid environments under an enterprise’s existing security controls.
Evaluation frameworks measure how an agent performs. The Synthesized Test Data Agent provides development and testing platforms with the production-representative data and realistic enterprise context needed to rigorously validate agent behavior and accuracy under real operating conditions. By making testing meaningful and repeatable, it helps organizations determine whether an agent can deliver the correct business outcome, rather than merely produce a plausible response, and build the confidence required before deploying it into mission-critical workflows.
“Building an agent is becoming easier. Proving that it can be trusted with a real business process is not,” said Nicolai Baldin, CEO of Synthesized. “Evaluation frameworks can measure how an agent performs, but they still need a realistic world in which that performance becomes meaningful. The Test Data Agent is built to create that world safely, before an agent is allowed to act on live systems.”
From build-and-deploy to continuous agent improvement
Test Data Agent helps organizations move beyond one-time pre-production testing and establish a continuous agent-improvement loop. Teams can define a business outcome, generate the required environment, create edge cases, execute the agent through their chosen framework, evaluate the result, improve the agent, and rerun the same scenario suite before promoting a new version.
Common use cases include:
- Pre-production validation: Gate enterprise and hosted agents before they receive access to live systems.
- Agent regression testing: Recreate consistent environments across runs so regressions can be attributed to the change rather than the data.
- Continuous optimization: Supply curated datasets and realistic environments to prompt-optimization, fine-tuning, and reinforcement-learning pipelines.
- Model and framework comparison: Run competing models or agent configurations against the same enterprise scenarios.
- Release governance: Create repeatable evidence that an agent has passed defined business, security, and operational conditions before deployment.
Rather than scoring an agent’s output, Synthesized will create the underlying data and environment against which the agent can be evaluated. The Test Data Agent therefore complements existing agent platforms, testing systems, observability tools and model-optimization pipelines.
“Enterprise agents will not improve through production traces alone,” Baldin continued. “Teams need safe environments where they can replay those patterns, introduce controlled variations and determine whether a change genuinely improves the agent. Our aim is to turn enterprise context into a repeatable validation asset.”
Keeping enterprise data within trusted boundaries
The Test Data Agent is designed for enterprises with highly sensitive and regulated data estates. It runs within customer-controlled on-premises, private-cloud and hybrid environments, generating, masking and provisioning the necessary data under existing identity, networking, security and governance controls, rather than requiring organizations to centralize or export raw production data to an external testing service.
This approach is intended to preserve the structure and complexity of enterprise data for agent validation while reducing exposure of customer, employee, financial and commercially sensitive information.
For agent-platform providers and systems integrators, the Test Data Agent acts as a framework-agnostic validation environment layer. They can retain their preferred agent orchestration, evaluation and governance technologies while using Synthesized to create the production-faithful context beneath them.
Purpose-built support for SAP environments
The validation challenge is particularly acute in complex SAP environments, where a single business process can involve multiple related tables, organization-specific configurations, authorization rules, and integrations with other enterprise applications. Testing an AI agent in these environments requires more than generating individual records.
The relevant document chains, process states, and cross-system dependencies must also remain consistent. Pre-production validation with Test Data Agent helps ensure that an agent can correctly handle restrictive authorizations, complex data relationships, and real-world process states before it gains access to live systems.
Specific SAP-focused use cases include:
- Pre-production validation of AI agents operating in SAP environments
- SAP ECC to SAP S/4HANA migration validation.
- Regression and business-process testing.
- Application modernization and release assurance.
- Privacy-safe SAP development and testing environments.
- Testing across SAP and connected databases or enterprise applications.
- Continuous validation as agent models, instructions and tools change.
For example, an enterprise validating an invoice-processing agent could use the Test Data Agent to create realistic combinations of suppliers, purchase orders, invoices, payment terms and currency conditions. The agent could then be evaluated against normal transactions and difficult exceptions before receiving access to live financial systems.
“The difficult part of testing an SAP agent is not demonstrating that it can navigate a workflow,” said Baldin. “It is recreating the relationships, exceptions and business conditions that determine whether the resulting action is actually correct. That is the infrastructure Synthesized is building.”
An open infrastructure layer for the agent ecosystem
Synthesized is positioning the Test Data Agent as an open infrastructure component for the enterprise agent ecosystem.
Agent-platform providers, agentic testing companies, systems integrators and internal engineering teams can connect the Test Data Agent to their existing development lifecycle rather than introducing another isolated testing framework.
The resulting workflow is:
Define the business scenario > create the production-faithful environment > execute the agent > evaluate the outcome > improve the agent > validate again before release.
This approach allows agent platforms to focus on building, orchestrating, observing, and improving agents while Synthesized supplies the environments and enterprise context required to do so reliably.