Conference Talk
Closing the loop: Using Test Gap Analysis and Coding Agents to ship with confidence
Closing the loop: Using Test Gap Analysis and Coding Agents to ship with confidence

Coding agents have made writing code and tests dramatically faster. As agents churn out changes and whole test suites faster than any team can review them, our already limited picture of the codebase gets even blurrier. We lose track of what parts of last week's work are actually exercised by tests and which only look like they are. That is quality debt, and it accumulates.

More generated tests and raw coverage percentages don't help much here. A suite can sit at a comfortable 80% test coverage while many lines an agent edits during a sprint go untested, because an aggregate number says nothing about code that was modified. Test Gap Analysis approaches the problem from the other side: it cross-references what changed in your code with what your tests actually executed, and surfaces the intersection of new or modified code that no test has ever run. Instead of a number you get a map of concrete risk.

In the talk, I will go through some of the observed effects AI agents have on software testing and product quality. I will show how test gaps can help by becoming a feedback loop for agentic coding. Once they are available as structured, real-time data, they can become a signal your agents can act on.

I'll walk you through an example setup: capturing coverage from your Robot Framework test runs, computing gaps against the diff, and exposing test gaps to a coding agent to implement additional tests via, for example, (Robot)MCP. This allows the agent to write tests for the specific untested changes it introduced, instead of just generating plausible-looking tests and hoping they matter.

We will also look at the limits. With Test Gap Analysis we can see which changed code was executed, not whether the code is bug free. It puts the focus on where we are missing coverage as tests cannot find bugs in code they never execute. It needs coverage from real test runs, so coverage instrumentation is an essential part of the setup. And an agent that closes gaps will cheerfully write tests that satisfy the metric without deepening anyone's understanding, so human review still matters.

Problem solving

Invited Talks

Your location, our expertise

We are happy to come visit you in your office for an internal conference or a workshop. Our list of topics includes quality analyses, quality control, but also test control or introducing peer reviews. You are also welcome to pick a topic of your choice

60-90 minutes talk
About a software quality topic
In your office or remote
Our travel expenses only
Date by arrangement