AICODEAgentic Programming · 67723 · 2026-7

67723 · Semester A (Fall 2026) · 2 credits · elective

Agentic Programming

Working with an AI coding agent is a skill of its own, separate from knowing how to code. This course teaches it directly: how to direct an agent, how to check its work, how to sandbox it, how to run several at once — and where all of that breaks.

You will learn to
  • Direct an agent through real, non-trivial software tasks.
  • Manage its context deliberately — subagents, plans, skills, hooks, memory files.
  • Spot prompt injection and the other risks specific to agent-based development.
  • Isolate an agent's execution in containers or virtual machines; use git worktrees to separate parallel tasks.
  • Use tests, type checks, linters and CI to check an agent's work.
  • Judge, with numbers, whether an agent is actually helping on a given task.
  • Compose several agents into one workflow.
Format

Watch the videos before class; use the lab for hands-on work.

There is no lecture. Each lab is two academic hours of 45 minutes, with a break between them: ninety minutes of building, in pairs, with a lab instructor present.

Your grade
20%Weekly quizzes on Moodle, taken without an agent. Each closes at 23:55 Israel time on its due date.
40%Labs, graded in the room on the day.
40%Final project, built in a group.

There is no written exam.

Attendance

Required. Labs are graded in the room.

Missing a lab for a justified reason — illness, reserve duty, and the like — means emailing your lab instructor: you are exempted, or you do the lab on your own and email the results.

Using AI

Everywhere except the quizzes.

Use an agent as much as you like in labs and the project — that is the point. Quizzes are the one place you work unaided.

Prerequisites

Intro2CS, and the C/C++ Programming Workshop.

You must be comfortable reading and writing code in at least one language. No AI or machine-learning background is assumed — this course does not teach you to program.

Cost

A paid coding-agent subscription, from the first lab to the end of the semester.

Budget roughly USD 100 per month. The labs are taught with Claude Code, so the plan we support is Claude Max 5×; a smaller plan is unlikely to be enough. Subscribe before your first lab, in the week of Tue 13/10.

Prof. Aviv Zohar · School of Computer Science & Engineering, Hebrew University of Jerusalem