TCS Builds Agentic Test Orchestrator Using Amazon Bedrock AgentCore for Regulated Workloads

Tata Consultancy Services (TCS) built an Agentic Test Orchestrator on AWS designed to automate software quality engineering workflows for Banking, Financial Services, and Insurance (BFSI) clients. To avoid relying on single prompts or traditional automation that still requires manual scripting, TCS decomposed testing into modular, specialized tasks executed by dedicated agents. The orchestrator runs on Amazon Bedrock AgentCore runtime using the Strands Agents SDK, leveraging isolated microVMs to enforce session-level tenant separation without requiring custom agent infrastructure. The system accepts diverse inputs, including Jira user stories, business requirement documents, Swagger specifications, and GitLab source folders. Project materials are indexed in tenant-isolated Amazon Bedrock Knowledge Bases using vector embeddings to ground generation. Specialized agents handle distinct phases: a generation agent creates Gherkin scenarios and Java or Python code, a chat agent handles iterative refinements, a dynamic task agent extracts structured stories from BRDs, an incremental agent extends tests across Jira chains, and an execution agent manages test runs. Testing workloads are routed dynamically on Amazon ECS—using AWS Fargate for API tests and Amazon EC2 with virtual displays for Selenium UI testing. An automated execution loop built with LangGraph applies bounded code corrections for verified errors while rejecting anomalous file alterations. TCS operates the platform in production using a pooled-compute, siloed-data architecture on Amazon EKS with tenant-specific encryption keys and storage. In an internal benchmark across approximately 100 scenarios, TCS stated that average engineering effort per scenario decreased from roughly 2 hours (about 200 total hours) to 14 minutes (about 24 total hours), representing an 88 percent reduction in effort, though the company noted actual results vary by workload and complexity.

Key takeaways

• TCS created an Agentic Test Orchestrator hosted on Amazon Bedrock AgentCore runtime to automate testing workflows across multi-tenant BFSI environments.
• The architecture divides responsibilities among dedicated agents for scenario drafting, code generation, conversational refinement, and execution management.
• Test runs execute on Amazon ECS with API workloads assigned to AWS Fargate and Selenium UI suites routed to EC2 with virtual display streaming.
• In an internal benchmark of 100 test scenarios, TCS reported an 88 percent reduction in manual effort, shifting from 200 hours to approximately 24 hours.