Find your way to be trained or even get certified in TMAP.
Start typing keywords to search the site. Press enter to submit.
Thinking of GenAI, most people have limited use cases such as Large Language Models (LLMs) and Image generators in mind. There are many more implementations of GenAI that may be less obvious but will have at least as much impact. Below figure gives a brief overview of the various use cases of GenAI. Many other possibilities are available or will be added in the near future.
For quality engineering and testing, the top of mind GenAI tools to apply are LLMs.Most people are only familiar with one way to use these models:
To work with a GenAI tool you need to provide a so-called prompt. Creating good prompts is an important skill. A good prompt is structured using several elements. We use the Crafting AI prompts framework to create prompts. An overview of this framework is described in this section and a more detailed description is found on the website www.craftingaiprompts.org.
Besides prompting there are two more ways of using AI for quality engineering and testing:
Creating prompts is a labor-intensive and error-prone approach for repetitive tasks. For frequent tasks, GenAI-based tools are created that have a user-interface and (behind the screens) makeuse of prefab prompts. An example of such GenAI based tool is Sogeti’s “GenAI Amplifier”.
GenAI based “agents” are autonomously operating bots that can perform specific predefined tasks without human intervention. These AI agents can operate individually but can also work together in a team of AI agents.Groups of collaborating GenAI agents may be organized in such a way that the tasks performed by the agents are aligned, and the group of agents, together, achieves goals that would not be possible by applying individual agents.
Agentic AI is a type of artificial intelligence that can operate autonomously rather than waiting for step-by-step instructions. AI agents can operate individually, but often work together in a team (or “swarm”) of AI agents.
Agentic AI presents a wide array of capabilities and holds significant promise for the future. On the one hand, its potential appears virtually limitless; on the other, it introduces a range of risks. For example, the agents lack true insight into the underlying goals and user needs. Also, the high operational speed of these agents makes it difficult for the “expert-in-the-lead” to oversee the process and to detect and address issues in a timely manner.
Agentic AI is well described in the paper Business, meet agentic AI: Confidence in autonomous and agentic systems [Engels 2025]:
The reason agentic and autonomous systems are transformative is because they allow users of technology to shift from defining the solution to simply stating their problem.
An AI Agent is any AI entity that works on behalf of another (mostly human) entity. It can accomplish high- level objectives and complete specific tasks, often using specialist capabilities. It has a degree of autonomy and authority to take actions that influence their environment.
The first wave of GenAI in software delivery has largely taken the form of assistance. Co-pilots summarize requirements, review code, generate test cases, and accelerate documentation. They respond to prompts and enhance individual productivity. Agentic AI represents the next step. It shifts from assistance to agency.
An agent does not merely generate output upon request. It operates with a defined objective. It can interpret signals, plan actions, trigger workflows, collaborate with other agents, and adapt its behavior within predefined governance boundaries. In a quality engineering (QE) context, this transition is structural. AI is no longer positioned as a support tool for testers and developers; it becomes an active participant in the control system of IT delivery.
The implications extend across the entire Software Development Life Cycle. In the requirements elaboration, agents continuously analyze specifications, detect ambiguity, and quantify business risk. During architecture and design, they assess whether proposed solutions align with performance, scalability, security, and regulatory expectations. In development, they evaluate the impact of changes and highlight structural weaknesses that may compromise quality. In testing, they dynamically adjust coverage based on risk signals rather than static regression lists. In release management, they aggregate technical and business indicators to provide transparent readiness assessments. In production, they monitor telemetry and user behavior, translating operational incidents into upstream learning.
Taken together, this creates a closed-loop quality system. Quality engineering becomes a continuously adaptive capability embedded throughout the lifecycle.
The real power of agentic AI, however, does not lie in isolated agents. It emerges through coordinated workflows. Agents interact through shared risk models, lifecycle artifacts, and event-driven triggers such as code commits, requirement changes, or production anomalies. Orchestration mechanisms define overarching goals, such as optimizing regression risk or assessing release readiness, and delegate subtasks to specialized agents. Outputs are consolidated, inconsistencies are resolved, and recommendations are generated in a structured and traceable manner.
This orchestration is governed by an increasingly important principle: policy as code.
In an agentic environment, governance cannot remain informal or document-based. Decision criteria, risk thresholds, compliance rules, and escalation paths must be codified in machine-readable form. Policies that once existed in manuals, quality gates, or steering committees are translated into executable rules embedded in workflows.
Examples include:
By expressing governance as code, organizations ensure that agentic autonomy operates within explicit, enforceable boundaries. Autonomy without policy introduces systemic risk; autonomy governed by policy enables scalable control.
In this new model, the role of the expert evolves significantly. Agentic QE does not diminish accountability; it sharpens it. Experts design the policies that agents execute. They define decision boundaries, encode escalation thresholds, and determine where autonomous execution is permitted. In high-impact or regulated scenarios, experts remain directly involved. In lower-risk contexts, they supervise autonomous workflows and intervene when policy conditions are breached.
The expert shifts from executor to system architect and risk governor. The focus moves from performing quality activities to designing and refining the rules, models, and controls that guide agent behavior.
This elevation of QE introduces a fundamental governance requirement. Every decision must be explainable. Every action must be logged. Every policy enforcement must be traceable. In regulated industries, such transparency is not optional; it is foundational to trust.
For senior leadership, the strategic implications are considerable. Agentic AI positions Quality Engineering as an intelligent, policy-driven control layer across digital delivery. Release decisions become continuously informed rather than event-based. Feedback cycles compress. Risk visibility improves. Organizational roles shift toward orchestration, governance design, and systemic oversight.
Agentic AI will not eliminate Quality Engineering. It will sharpen it. What was often perceived as mainly a testing function becomes a lifecycle-wide, policy-encoded quality intelligence system; integrated, adaptive, and strategically indispensable to digital performance.
AmpQE overview
