AI Coding Agents in HBO-ICT: What Should Programs Teach Students?
A curriculum-level view of the capabilities HBO-ICT students need for responsible AI-assisted software development, beyond prompt writing.
TraceYield Insights
12 min read · Updated September 14, 2026
TraceYield
AI engineering evidence
AI Coding Agents in HBO-ICT: What Should Programs Teach Students?
Prompting is not a curriculum
Students need more than a collection of prompt patterns. They need independent programming fundamentals, problem framing, context specification, code reading, debugging, verification, architecture reasoning, security awareness, and the ability to recognize when an agent is wrong or unsuitable.
The professional value of a coding agent depends on the human who can direct, evaluate, constrain, and take responsibility for the work. Curriculum should develop that capability across levels rather than treat a tool interface as the subject.
Teach the foundations that make assistance useful
Students who understand data structures, control flow, testing, interfaces, version control, and basic software design can evaluate generated code more effectively. Fundamentals are not made obsolete by agents; they become the basis for asking precise questions and noticing plausible errors.
This does not mean returning to an exclusively manual curriculum. It means ensuring that assisted practice does not remove the conceptual work students need to do independently at the stage where it is being learned.
Add agent supervision capabilities
Students should learn to provide relevant context, define constraints, inspect plans, compare approaches, review changes, run tests, investigate failures, and manage scope. They should understand tool limitations, permissions, external data handling, and the difference between a suggestion and verified behavior.
These capabilities can be taught through structured exercises: critique an agent’s patch, repair a flawed generated test, compare two designs, or ask students to explain what evidence would justify acceptance.
Make responsible use technical and professional
Privacy, security, licensing, accessibility, bias, reliability, and accountability belong in programming education because they affect software decisions. A student should know that entering code or data into a tool can create obligations beyond the local assignment.
Responsible AI use is not a separate ethics lecture only. It can be embedded in code review, architecture, testing, project planning, and assessment. That makes the responsibility concrete.
Teach when not to use an agent
Professional judgment includes choosing a smaller tool, reading documentation directly, debugging without generated output, or pausing to understand a concept before asking for implementation. Students need experiences where assistance is limited so that they can recognize their own understanding and gaps.
A curriculum that permits AI everywhere without designing these contrasts risks confusing completion with learning. A curriculum that bans it everywhere misses the professional context students will enter.
Assess the capabilities explicitly
If a programme values AI literacy, it should assess it through decisions, verification, explanation, and responsible practice. Do not make “good prompting” the sole outcome. The programme should be able to show how students progress from using a tool to supervising an AI-assisted development process.
That progression belongs across the curriculum, not in one optional workshop.
A possible progression across years
Early study can emphasize reading code, tracing behavior, and controlled debugging. Middle study can add context specification, test design, code review, and comparison of agent suggestions. Later study can include architecture, security, professional tool governance, and authentic project delivery. Each stage should retain enough unassisted work to keep the foundational capability visible.
The exact sequence belongs to the programme. The principle is that AI literacy develops alongside programming competence, not as a substitute for it.
Assess supervision as a professional skill
Students can be asked to review a generated patch, identify risks, improve a prompt or task description, explain a rejected approach, and defend a verification plan. These activities assess supervision and judgment rather than prompt cleverness.
A curriculum that makes those expectations explicit will prepare students for a wider range of tools than one tied to a single assistant.
Place tool competence inside software competence
A student who can prompt an agent but cannot read a stack trace, inspect a dependency, or reason about an interface is not prepared for responsible development. Curriculum design should pair agent practice with the underlying programming and engineering concepts needed to evaluate the result.
This pairing can be explicit in course outcomes: students might be required to review a generated patch for correctness, security, maintainability, and fit with the project context.
Teach professional boundaries
Students should practice deciding what information is safe to share, when a human review is required, how to disclose material use, and how to respond when a tool produces a confident but unsupported answer. These decisions are part of employability, not optional etiquette.
A curriculum that teaches the boundary conditions will remain useful when a new tool replaces the current set of brand names.
Curriculum outcomes should name judgment
Outcomes can say that students evaluate generated code against requirements, design and execute verification, recognize security and privacy risks, explain tool limitations, and adapt a solution responsibly. These are more durable than an outcome that simply says students can write prompts.
They also give assessment teams something concrete to look for across courses.
Agent completion does not always mean engineering completion.
References
- Npuls: Vision on assessment, examination and AI
- UNESCO: Guidance for generative AI in education and research
- ACM Task Force on Generative AI and Programming Assessment
- Designing assessments in the generative AI era for ICT education
- Anthropic Claude Code setup
- OpenAI Codex CLI overview
- Cursor Agent overview
Pilot Program
Understand the WHY behind your engineering AI usage.
TraceYield evaluates trajectory evidence instead of stopping at spend totals. Join the private pilot to review AI coding usage with engineering context, security controls, and developer trust.
Join the TraceYield private pilot