Engineering Quality for the AI Era
From Test Automation to Intelligent, Agentic Quality Engineering
Key takeaways
- Traditional automation remains necessary, but test execution alone cannot establish trust in AI-enabled and agentic systems.
- Context — retrieval quality, prompt/context construction, tool permissions and memory behavior — is part of the product and must be engineered and evaluated.
- Evaluation must be multi-dimensional, combining deterministic assertions with semantic, risk-based, behavioral and human review.
- Agentic systems require quality controls around decisions and actions, not only outputs.
- The target state is continuous quality intelligence, where delivery and production evidence feed back into risk, evaluations and engineering decisions.
In the paper's own terms
The problem
Software delivery is moving from systems that execute predefined logic toward systems that interpret intent and participate in decisions. The paper frames the resulting shift as quality moving from verifying transactions to governing outcomes.
What the paper proposes
The IQE Labs Methodology: six constructs, each with one job — the IQE Maturity Model (Assess), IQE Transformation Lifecycle (Transform), IQE Loop (Operate), Agentic QE Control Plane (Control), Quality Evidence Architecture (Evidence) and the 30/60/90 Adoption Path (Scale) — following the canonical relationship Assess → Transform → Operate → Control → Evidence → Scale.
Who should read it
The paper describes itself as "a field guide for Quality Engineering leaders" and a practitioner-focused methodology architecture written for QE leaders.
How this paper labels its claims
“The future of Quality Engineering is not more testing at the end.”
