Read
Mental models first
Explain, trace, and name purpose. Writing from a blank file is not the default move.
Dissertation prototype
Introductory programming in the generative AI era
CodeSense supports the development of foundational programming skills. CodeSense provides small, focused challenges that help you understand code, spot mistakes, evaluate solutions, and develop confidence with fundamental programming concepts.
Choose a topic 0 of 48 activities attempted in this browser.
Judge
AI as an object of critique
Accept or reject a suggestion. Find the bug. The tutor never finishes the item for you.
Shape
Improve and specify
Rewrite weak generated code, and practise saying what a solution must do — including edge cases.
Activity types
Each type is short, single-concept, and mapped to a Bloom-inspired outcome.
Read — build a mental model
L1 Knowledge
Identify Terminology
Pick the correct technical term for a piece of code.
Shared vocabulary is the foundation for everything else: students cannot explain, debug, or critique generated code (or specify what they want) without words such as parameter, index, key and iteration. Distractors are near-miss terms that novices genuinely confuse, such as parameter versus argument.
Read — build a mental model
L1 Knowledge
Define a Concept
State what a core programming idea is, in your own words.
Putting a definition into one's own words shows whether a term is understood or merely recognised. Short, low-pressure definitions give an early, honest baseline before students read or evaluate code that depends on the idea.
Read — build a mental model
L1 Knowledge
Match Terms to Definitions
Pair each technical term with the definition that fits it.
Recall of vocabulary is the base of the taxonomy, and matching tests several related terms at once, including the pairs novices mix up. The terms are anchored to a short code fragment so each definition can be checked against something concrete, and one extra definition fits no term so the last pair cannot be answered by elimination.
Read — build a mental model
L2 Comprehension
Explain Code
Say in plain English what a short fragment does.
Before students generate or accept code, they need a verbal model of behaviour. Explanation items make reading a first-class activity rather than a prelude to writing.
Read — build a mental model
L2 Comprehension
Program Tracing
Predict output by simulating execution.
Tracing builds an operational mental model: what is true after each statement. Distractors are drawn from documented novice misconceptions rather than random wrong answers.
Read — build a mental model
L2 Comprehension
Program Purpose
Name the problem the fragment is solving.
Students often describe syntax ('it uses a for loop') instead of purpose ('it finds the largest value'). Purpose items separate mechanism from intention — a distinction needed when judging AI-generated code that may be locally plausible but globally wrong.
Shape — improve or specify
L3 Application
Prompt Construction
State requirements clearly enough to specify a solution.
The learning goal is precise communication of behaviour — inputs, outputs, and edge cases — not prompt-engineering tricks. If a requirement is underspecified, generated code will guess.
Shape — improve or specify
L3 Application
Complete the Code
Choose the fragment that completes a partly written program.
Application without a blank page: students apply a concept by deciding what the missing piece must be, which requires reading the surrounding code and the stated requirement. Distractors encode off-by-one, quoting and operator errors that generated code often contains.
Shape — improve or specify
L3 Application
Modify the Code
Change existing code to meet a new requirement.
Most real programming, and most work with AI output, is changing code that already exists. Students must find the part of the code a new requirement touches and change only that, which is a different skill from writing a solution from scratch.
Judge — analyse and evaluate
L4 Analysis
Debugging
Find the fault and say why it matters.
Debugging is analysis: locating the mismatch between intended and actual behaviour. In an AI-rich setting this is the same skill used to inspect generated code.
Judge — analyse and evaluate
L4 Analysis
Compare Solutions
Decide how two solutions to the same problem differ.
Generated code often arrives as alternatives. Comparing them means tracing both and finding the exact point where they diverge: analysis rather than recall.
Judge — analyse and evaluate
L4 Analysis
Analyse a Misconception
Diagnose why a student's prediction is wrong.
Diagnosing why an answer is wrong needs a model of how a novice reasons. The same analytic skill is used to review explanations written by AI tools, which can be fluent and still wrong.
Shape — improve or specify
L5 Synthesis
Code Improvement
Improve a working but weak generated solution.
Synthesis here means revising an existing artefact: names, structure, edge cases. Students start from AI-like output so the task is judgement plus rewriting, not blank-page authorship.
Shape — improve or specify
L5 Synthesis
Extend the Code
Add functionality to working code, combining ideas.
Synthesis: combining concepts already met (variables, loops, conditions, lists, functions) to build something new on a working base, and reasoning about where the new code belongs.
Judge — analyse and evaluate
L6 Evaluation
AI Suggestion Evaluation
Accept or reject a generated line, with a reason.
The pedagogically interesting moment is not generating code but deciding whether to trust it. Students practise evaluation: correctness, fit to the requirement, and a stated justification.
Judge — analyse and evaluate
L6 Evaluation
Review Code Quality
Judge working code and pick the most justified review comment.
Evaluating quality is different from checking correctness. Learners practise separating 'wrong' from 'could be better', and rejecting confident but unjustified criticism or praise, which is a core skill when reviewing generated code.