30 Oct 2026 · 1 min read
Agentic Testing: building autonomous QA Councils with Claude
A meta-architecture where specialized AI agents collaborate to generate, execute and self-heal test suites.
#ai#qa#claude#performance
A QA Council is a group of specialized agents, each owning one part of the testing lifecycle, coordinated by Claude Code.
The council
| Agent | Responsibility |
|---|---|
| Analyzer | Reads a microservice codebase and proposes candidate test scenarios |
| Mocker | Finds upstream dependencies and generates stubs (WireMock) |
| Developer | Writes the load / API tests (NBomber, k6) |
| SLO agent | Connects to Grafana MCP to set and validate Service Level Objectives |
The loop
analyze → mock → generate → execute → compare → self-heal
Each run is isolated with Testcontainers and Toxiproxy, so results don't depend on shared environments.
Reporting
A reporting agent compares artifacts between service versions (e.g. v1.49 vs v1.50) and produces a PDF summary for non-technical stakeholders: improvements, regressions and points of concern.
Takeaways
- Small, specialized agents beat one giant prompt
- Custom skills make agent behaviour repeatable
- Keep humans in the loop for judgement calls