TL;DR
- Introducing Test Data Agent, a new agentic infrastructure capability that generates realistic data so enterprises can validate AI agents before production
- Model evaluations alone can't confirm an agent makes the right call under real conditions
- Built for continuous agent improvement, not a one-time check
- Runs entirely inside an organization's own environment
- Particular depth for complex SAP environments, including S/4HANA migration
- Already in early access with tier-1 banks, limited availability from August 18
Building an AI agent is becoming the easy part. Proving it can be trusted with a real business process is not.
That's the gap we're addressing with Test Data Agent, a new agentic infrastructure capability. Before an AI agent earns a place in production, it needs to be tested against something real: realistic data, business context, and system states that reflect how the enterprise actually operates. Test Data Agent generates exactly that. It's framework-agnostic and PII-free by design, with particular depth for complex SAP estates, finance, procurement, supply chain, operational workflows, and ECC-to-S/4HANA transformation programs.
The gap between building an agent and trusting it
Most enterprises can now build AI agents faster than they can prove those agents are actually ready to operate across mission-critical workflows.
Model evaluations and demonstration datasets can confirm that an agent produces a plausible response. What they can't confirm is whether that agent will make the correct decision once it meets the real conditions of an enterprise environment: the missing records, unusual transactions, conflicting instructions, access restrictions, and cross-system dependencies that don't show up in a clean demo.
That gap is hard to close because the environments needed to close it are hard to come by. Enterprise test environments are frequently stale, incomplete, or disconnected from production complexity, and privacy, security, sovereignty, and regulatory requirements limit how much sensitive production data can actually be used in development and testing.
We built Test Data Agent to be that missing validation layer. Starting from a business scenario or testing objective, it identifies and provisions the data, relationships, and system states required to test how an agent performs under conditions that actually resemble production.
What it does
Test Data Agent lets teams:
- Identify the business entities, records, relationships, and system states a test requires
- 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 testing and evaluation frameworks can call it directly
- Operate within on-premises, private-cloud, and hybrid environments under an organization's existing security controls
Evaluation frameworks are effective at measuring an agent's performance. What they need underneath them is a realistic enough world for that performance actually to mean something. We built Test Data Agent to provide the production-representative data and enterprise context that makes agent evaluation rigorous rather than superficial, helping teams determine whether an agent delivers the correct business outcome, not just a plausible-sounding one.
"Building an agent is becoming easier. Proving that it can be trusted with a real business process is not," said Nicolai Baldin, our Founder and CEO. "Evaluation frameworks can measure how an agent performs, but they still need a realistic world in which that performance becomes meaningful. Test Data Agent is built to create that world safely, before an agent is allowed to act on live systems."
From one-time testing to continuous agent improvement
Rather than a one-time pre-production check, we built Test Data Agent to support a continuous agent-improvement loop. Define a business outcome, create the environment, define edge cases, run the agent through the team's existing framework, evaluate the result, improve the agent, then rerun the same scenario suite before promoting a new version.
That directly supports a few common use cases. Pre-production validation, gating agents before they get access to live systems. Regression testing, recreating consistent environments across runs so a regression can be attributed to the actual change rather than to shifting data. Continuous optimization, feeding curated datasets into prompt-optimization, fine-tuning, and reinforcement-learning pipelines. Model and framework comparison, running competing agent configurations against identical enterprise scenarios. And release governance, generating repeatable evidence that an agent has passed defined business, security, and operational conditions before deployment.
"Enterprise agents will not improve through production traces alone," Baldin said. "Teams need safe environments where they can replay those patterns, introduce controlled variations, and determine whether a change genuinely improves the agent. We aim to turn enterprise context into a repeatable validation asset."
Built to keep enterprise data where it belongs
We built Test Data Agent with highly sensitive, regulated data estates in mind. It stays inside an organization's own environment, whether on-premises, private-cloud, or hybrid, generating, masking, and provisioning data under existing identity, networking, security, and governance controls rather than pulling raw production data out to an external service.
That's what makes Test Data Agent a framework-agnostic validation layer for agent-platform providers and systems integrators. Whatever orchestration, evaluation, and governance tools a team already relies on stay in place, Test Data Agent just supplies the production-faithful context those tools need underneath.
Purpose-built for SAP
The validation challenge becomes more complex in SAP environments, where a single business process can affect multiple related tables, organization-specific configurations, authorization rules, and integrations with other enterprise applications. Properly testing an agent here takes more than generating individual records. Document chains, process states, and cross-system dependencies also need to hold together.
That covers a range of SAP use cases:
- Pre-production validation of agents operating in SAP environments
- ECC-to-S/4HANA migration validation
- Regression and business-process testing
- Application modernization and release assurance
- Privacy-safe SAP development and testing environments
- Continuous validation as agent models, instructions, and tools evolve
Take an invoice-processing agent, for example. Test Data Agent can generate realistic combinations of suppliers, purchase orders, invoices, payment terms, and currency conditions, allowing the agent to be evaluated against both routine transactions and difficult exceptions that determine whether it's ready for live financial systems.
"The difficult part of testing an SAP agent isn't demonstrating that it can navigate a workflow," Baldin said. "It's recreating the relationships, exceptions, and business conditions that determine whether the resulting action is actually correct. That's the infrastructure we're building."
An open layer for the agent ecosystem
With Test Data Agent, we've built open infrastructure for the enterprise agent ecosystem rather than another standalone testing tool. Agent-platform providers, agentic testing companies, systems integrators, and internal engineering teams can connect it directly into their existing development lifecycle.
The resulting workflow looks like this: define the business scenario, create the production-faithful environment, execute the agent, evaluate the outcome, improve the agent, validate again before release. Agent platforms focus on building, orchestrating, observing, and improving agents, while we provide the environments and enterprise context needed to do so reliably.
Availability
Test Data Agent is already in early access with tier-1 global bank design partners, running inside their own environments. Limited availability opens to existing clients and ecosystem partners from August 18, 2026, ahead of general availability later in Q3 2026.
Want to see what production-faithful agent validation looks like for your organization? Get in touch to evaluate Test Data Agent for enterprise agent validation, SAP transformation, continuous agent improvement, and other complex testing use cases.
SAP, SAP S/4HANA, and other SAP products and services mentioned are trademarks or registered trademarks of SAP SE or its affiliates in Germany and other countries. This announcement does not imply endorsement by or affiliation with SAP SE unless separately stated.
