Corinna is a software architect at Digitec Galaxus. Her original background is in mathematics: she joined the software industry in 2017 after a PhD in algebra. Her two main passions have always been mapping complex domains onto code that humans can easily understand, and helping groups of people work well together towards a common goal. Noticing that both come down to basic principles of human cognition, emotion, motivation and communication, she started studying psychology at FernUniversität in Hagen in 2025.
What is the main outcome of programming: code, or understanding?
In 1985, Peter Naur argued that programming is fundamentally about building a theory: the ability to explain a system, reason about it, and connect the code to the real-world problem it solves. Agentic coding challenges this goal. As coding agents become more capable, developers can produce more code while being less involved in producing it.
In my own team, I began noticing signs of what Margaret-Anne Storey calls “cognitive debt”: an erosion of shared understanding of the codebase. As a developer, technical leader, and part-time psychology student, I started looking at this problem through the lens of research on human learning, cognition, and LLM interactions.
In this talk, I will share what I learned and how it changed the way I think about working with coding agents. We will look at how concepts such as desirable difficulties and learning from errors can shape LLM workflows, how collaboration can keep developers intellectually engaged, and how teams can create an environment that supports learning and preserves shared understanding.
The goal is not to abandon AI, but to use it without outsourcing our understanding of the systems we build.
Searching for speaker images...