Agentic Coding Workshops for Universities, Lecturers and Students

Why universities cannot ignore coding agents

Students already use tools like Claude Code, OpenAI Codex, Cursor or GitHub Copilot, whether a course allows them or not. Graduates will meet them in their first job, where the expectation is shifting from writing every line to directing an agent, reviewing its work and owning the outcome. A department that bans the tools entirely risks teaching a workflow that no longer exists. One that allows them without guidance risks students who can produce working code they cannot explain.

What belongs in the curriculum

Agentic coding is not a replacement for fundamentals. Students still need data structures, algorithms, debugging and system design, arguably more than before, because reviewing code requires understanding it. What changes is the layer on top:

• Task specification: breaking a problem into steps an agent can execute, writing acceptance criteria first.
• Context engineering: instruction files such as AGENTS.md or CLAUDE.md, project conventions and giving the agent the right files.
• Verification: writing tests before or alongside generated code, reading diffs critically, spotting plausible but wrong solutions.
• Security and licensing: hallucinated dependencies, secrets in prompts, licence questions around generated code.
• Professional responsibility: the person who merges the code is accountable for it.

A practical sequencing many lecturers choose: fundamentals courses with limited or no agent use, then project courses where agents are explicitly allowed and the review process becomes part of the grade.

Keeping assessment honest

Take-home programming assignments lose their value as a measure of individual skill once an agent can solve them in minutes. Detection tools for AI-generated code are not reliable enough to base grading decisions on. More robust approaches shift what is assessed:

1. Oral or live components: students explain and modify their own code in a short session.
2. Process evidence: commit history, agent session logs and a short reflection on what the agent got wrong.
3. Review tasks: students receive generated code with seeded defects and must find and fix them.
4. Clear declarations: each assignment states which tools are allowed and how their use must be documented.

Whatever the policy, it should be written down per course and aligned with the university's examination regulations.

Hands-on formats that work

Lecturers should go first. Teaching staff who have spent a day steering an agent through a real codebase write better assignments and more realistic policies. Our article on a workshop agenda for half-day, one-day and multi-day formats covers the building blocks.

Practical points for universities

• Check how student accounts are provided: personal sign-ups, institutional licences or API access through the university.
• Clarify data protection before students paste research data or personal data into cloud tools.
• Provide a fallback for students who cannot or will not use a specific vendor's service.
• Agree which repositories and data the agents may access in lab environments.

How jocapps can help

Our training brand neooutput runs hands-on workshops, team trainings and 1:1 consulting on agentic coding and AI tools, in English or German, on-site or remote. For universities this can mean a lecturer workshop, a student session within an existing course or a project week, each ending with working results rather than slides. Details are on our agentic coding workshops page, or book a free 30-minute intro call via the meeting planner.