What changed is that TCS moved from a traditional, manually‑maintained test‑automation stack to a fully agentic test orchestrator powered by Amazon Bedrock AgentCore. The new platform ingests Jira tickets, business requirement documents, and GitLab changes, then generates, refines, and executes API and UI tests while preserving per‑tenant isolation. Practitioners care because the approach removes repetitive translation work, leverages model choice per task, and offloads scaling and isolation to a managed runtime.
Why Conventional Automation Was Insufficient
Existing test frameworks accelerated execution but left the upstream effort untouched: engineers still had to author scenarios, translate them into scripts, maintain locators, and debug failures. When a user story changed, scripts quickly became stale, and a single generative‑AI prompt could not enforce the multi‑stage validation required for regulated BFSI workloads. The need for deterministic routing, retry logic, error classification, and repeatable evidence collection drove the design of a more granular, agent‑based solution.
Agentic Test Orchestrator Architecture
The orchestrator decomposes testing into specialized agents, each with its own model, tools, and validation rules. A deterministic supervisor routes work between agents, while an execution workflow enforces retries, persistence, and reporting. Key AWS services provide the surrounding platform:
Amazon EKShosts the tenant web UI behindAmazon CloudFrontwith authentication viaAmazon Cognito.- Agents run in the
Amazon Bedrock AgentCoreruntime, which scales on demand and isolates each session in a microVM. - Generated tests execute on
Amazon ECScontainers. - Results and logs are stored in
Amazon S3,Amazon DynamoDB, and captured byAmazon CloudWatch. - Per‑tenant security is enforced with
AWS WAF,AWS KMS, andAWS Secrets Manager.
The agents themselves are built with the Strands Agents SDK, allowing TCS to focus on testing logic rather than agent infrastructure.
Implementation and Operational Flow
Input artifacts flow from Jira, BRDs, and GitLab into the tenant application. The supervisor creates a session in the AgentCore runtime, which spawns isolated microVMs for each tenant. Specialized agents generate test scenarios, produce code for API or browser UI, and iteratively refine output with a human tester. Once approved, the execution agent dispatches the test to ECS, monitors execution, and records outcomes in S3 and DynamoDB. CloudWatch aggregates logs for observability, while WAF and Cognito enforce network and identity boundaries per tenant.
Related CloudNinjas coverage: AWS.
What This Means For Practitioners
Adopting an agentic orchestrator means you can offload model selection, scaling, and isolation to a managed service, reducing operational overhead. Engineers should evaluate the fit of per‑task models, design deterministic supervision logic, and ensure that tenant data never leaves isolated microVMs. Monitoring should focus on session health, retry patterns, and evidence integrity. Security teams must verify that the combination of WAF, Cognito, KMS, and Secrets Manager satisfies isolation requirements without adding custom layers.
