TCS cuts BFSI test effort 88% with Agentic Test Orchestrator

TCS's starting point was not slow test execution but the fragmented path from business requirements to trustworthy test evidence. Traditional automation improved execution speed, yet teams still had to create scenarios, convert them to scripts, maintain locators, and diagnose failures—and scripts could become stale when a linked user story changed. TCS concluded that a single generative AI prompt could not provide the control a multi-stage engineering workflow needs, so it decomposed testing into specialized agents with their own tools, validation rules, and context, coordinated by a deterministic supervisor.

The choice of Amazon Bedrock AgentCore followed three needs: a choice of models so each agent uses the one that fits its task, a runtime that scales with demand across many tenants without provisioning or tuning agent servers, and session-level isolation through a dedicated microVM per session. In production, TCS runs the orchestrator as a multi-tenant SaaS on AWS, with a shared control plane and per-tenant application planes that separate identity, data, and keys. All of this supports the four requirements TCS set: grounded context, coverage of both API and UI testing, strict tenant isolation, and a repeatable path from requirement ingestion to test evidence and reporting.