Honours Dissertation — Metacognitive Calibration

The question

I've written before about working with AI without becoming its passenger — the rituals of reading diffs and interrogating the agent instead of accepting what it types. My Honours dissertation at Belgium Campus takes that instinct seriously as a research question: do developers understand AI-generated code as well as they think they do? The industry keeps measuring how much faster AI makes us; far less attention goes to whether our confidence in understanding the resulting code tracks our actual understanding. That gap — metacognitive calibration — is what the work maps.

What "metacognitive calibration" means

Calibration is the fit between how confident you are and how right you are: a well-calibrated developer who says "I'm ninety percent sure I understand this function" is right about nine times in ten. Metacognition is the layer above that — noticing what you know and what you don't. Put together, the question becomes something you can measure. When a developer reads code an AI wrote, does their sense of understanding track the understanding they can actually demonstrate — explaining it, predicting what it does on an odd input, spotting where it's wrong?

Psychology has studied this gap for decades in other domains: the illusion of explanatory depth, the way fluent text feels understood, the way confidence rises faster than accuracy. The dissertation asks what the software-engineering literature found when it looked at the same gap with AI-generated code in front of the reader.

Why a scoping review

A scoping review maps what evidence exists rather than testing one hypothesis: which studies have measured developers' confidence against their comprehension of AI-written code, how they measured it, what they found, and where nobody has looked yet. It is the right shape for a question the field only recently started asking — an experiment would answer one narrow version of it, a map shows what is known and what is missing. It also keeps the outcome honest by construction: the result is a charted landscape and a list of gaps, not a headline.

Where it stands

The proposal and literature review are done; the charting and synthesis are the work of the coming months, and the final report is due in December 2026. Because the review is still moving, this page stays deliberately high-level: the title and the question are public, the research questions and findings are not until the work is submitted. When it is, this page will say what it found — including if the answer turns out to be "the evidence is thinner than the confidence."

Why I'm the one asking

My own working habit is the motivation. Most of the code I ship is AI-typed under my review, and the honest question is whether that review is as good as it feels from the inside. The essay that started it lays out the instinct; the dissertation is the disciplined version of the same question. Writing happens where everything I do happens — in a terminal, now on macOS, working the same as it did on Linux.