Welcome to station five. We traveled through the entire spectrum of agentic testing. Now it is time to step into a critical phase of this course: the architect's perspective.
When organizations fail with AI, it is almost always because they treat it as a tool to spam code. If your strategy is to let engineers generate scripts that do not end, you will run into a maintenance nightmare. As an SDET, your job is to shift the paradigm using a strategy I call the 80/20 automation split.
We divide all work into two strict buckets: the execution bucket and the architecture bucket.
The 80% execution bucket is the repetitive work of automation, such as mapping fields, writing basic assertions, and translating documented cases into tests. We offload 100% of that execution to our agentic workforce.
The 20% architecture bucket is reserved for the human engineer. AI should not design a framework. It should not connect secure tool infrastructure, and it should not write the core boundaries.
By offloading the boring parts of scripting, you free up your mental bandwidth to focus on system design, framework scaling, and team mentoring. Your value is found in architecting the ecosystem.
To orchestrate this assembly line, you must understand the mechanical difference between a skill and a sub-agent.
A skill file is a static knowledge manual. It is written in plain markdown, like a page object skill file. It defines the laws of our framework, but it cannot run code, it cannot open a browser, and it cannot talk to external tools.
Now, when it comes to a sub-agent, it is an active, autonomous worker. It is configured with a targeted system prompt, integrated with tool sets like the Spira MCP or the Playwright CLI, and handed our skill file as its mandatory operational guide.
The skill file is the brain. The sub-agent is the muscle. And when you connect them together, you unlock the power of an agentic assembly line — which we're going to look at next.
Quiz
The quiz for this chapter can be found in 5.3.