Agentic Software Engineering/Overview Conceptual model Curriculum Learning outcomes Assessment Tooling reference Glossary

Assessment design

Grading the discipline

In an agentic course the artefact a student delivers is co-produced with an agent, so the grading scheme grades the discipline rather than the lines of code.

Grading philosophy

The course grades what the student contributes that the agent cannot: framing, specification, context engineering, verification design, code review against the MRP standard, multi-agent coordination, security and governance analysis, and the practitioner’s reading of the field.

First, the audit trail is gradable. The student’s chat transcript, commit history, context files, lessons-learned notes, and verification records are submitted alongside the final artefact. Without these, the artefact is opaque and cannot be evaluated.

Second, the spiral is gradable, and not only its endpoint. The mid-project commit-point decisions, the recovery from drift, and the revisions to the specification that the build forced are the points where the student’s judgment shows.

Third, verification is gradable as a deliverable, and not only as a state to be reached. A student who produced thorough verification gates that caught real failure modes scores higher than a student whose work happens to be correct but whose gates would have caught nothing.

Graded artefacts and weights

Continuous coursework

50%

Collected as four portfolios, each submitted at the end of its module cluster.

15%

Framing and specification portfolio

Modules 6 & 10 → LO3, LO4

The framing document for the student’s individual project together with a complete specification that approximates Knuth’s five criteria.

10%

Context engineering and safety portfolio

Modules 11 & 12 → LO5, LO6

The project’s CLAUDE.md, the lessons-learned file maintained across at least three sessions, a Git practice log, and a brief retrospective on one trajectory-management intervention.

15%

Verification and review portfolio

Modules 13 & 16 → LO7

A verification gate design, evidence that the gates caught at least one real failure mode, and an MRP-compliant code review of agent-generated or legacy code.

10%

Multi-agent workflow design

Modules 14 & 15 → LO8

A short design document that decomposes a given task into a multi-agent workflow, justifies the cost of explicit orchestration, names the patterns applied, and reflects on what failed.

The project

50%

Runs through the semester. The student builds a small software product across at least three full turns of the evolutionary spiral, with audit trail.

20%

Audit trail of the project’s turns

Modules 4 & 10 → LO2, LO6, LO10

Graded for legibility, completeness, and the visibility of judgment: original intent, what changed between turns and why, the commit-point decision at each turn, and where drift threatened.

10%

MRP-compliant pull request

Modules 13 & 16 → LO7

The merge-ready deliverable for the final turn. Must satisfy all five MRP criteria with evidence.

Security and governance memo

Module 17 → LO9

Prompt-injection surface, permission boundaries, OWASP-relevant risks, and the governance questions a deploying organisation would have to answer.

15%

Reflective analysis

Modules 6 & 10 → LO10

What the student learned by executing the project, focused on what changed in the specification because of the build — demonstrating they noticed the co-evolution rather than treating it as scope creep.

Rubric structure

Four general criteria

Process traceability

Whether an assessor can reconstruct what the student did and why. Audit trails that are absent, opaque, or evidently retrofitted score zero, regardless of the quality of the final artefact.

Specification quality

Whether the student’s framing and specifications hold under interrogation. A specification that addresses the five-criteria approximation, names its known pitfalls, and includes the rationale behind its constraints scores higher than one that lists features.

Verification credibility

Whether the student’s verification work would catch the failure modes the project actually faces. Gates that are present but catch nothing (verification theatre) are explicitly penalised.

MRP-readiness

For the final deliverable, whether all five MRP criteria are satisfied with evidence. An artefact that satisfies four out of five is not merge-ready and is treated as such in grading.

Academic integrity

The integrity standard for an ASE course is different from the standard for a programming course. Agent use is expected. What the student must do is declare it.

Every submitted artefact carries a brief delegation note: which parts the student wrote themselves, which parts the agent produced under direction, and what the audit trail records. The audit trail itself is the primary check.

The integrity violations the course recognises are three: hidden delegation, modification of the audit trail to conceal what was delegated, and submission of artefacts that cannot be reconstructed from the audit trail. Using an agent to produce the artefact is, by itself, none of these.

Re-submission and feedback

The course allows re-submission of one continuous-coursework portfolio. The rationale is that the spiral is the discipline’s central rhythm, and the assessment process should model it. A student who learns from feedback and re-submits with a corrected audit trail is exercising the discipline.

The final project is graded once, with reflective analysis allowing the student to articulate what they would have changed.

Mikael Alemu Gorsky

Mikael Alemu is an educator and researcher, and the author of two programs: Agentic Software Engineering, on building software with AI agents, and Building AI-Native Agentic Systems, on building software that thinks.

He teaches at the Holon Institute of Technology, near Tel Aviv, where Agentic Software Engineering runs as a credit-bearing course. He is an educator and researcher.

Nine published works, 76 citations. A 350-page textbook under contract with a major academic publisher.

Teaching and programs

Agentic Software Engineering — program, preprint and textbook

The discipline of structured, auditable human-agent workflows for building software. The human frames, specifies and judges. The agent executes. Nineteen modules in four parts, built on a running project called Tribunal, a web application in which agents argue opposing sides of a case and a judge agent decides. Taught for credit at the Holon Institute of Technology.

Agentic Software Engineering curriculum

Building AI-Native Agentic Systems — program, paper and book in writing

How to build systems that hold a language model as a working component, and treat that component as what it is: stochastic, slow and metered. Fourteen modules in four parts, about seventy hours. The running project is the Observatory, a news agency that watches sources, selects what matters and publishes on a cadence.

Research and analytics

Publications — journals and proceedings

Nine works, 76 citations. Research on artificial intelligence in education, with Ilya Levin and Alexei Semenov.

Artificial Intelligence as Game Changer in Cybersecurity: What We Learned in 2025-2026, and how this is relevant for Africa — Preprint, June 2026. By Mikael Alemu Gorsky. arXiv.
ASE-26: a curriculum for agentic software engineering as a discipline — Preprint, May 2026. By Mikael Gorsky. arXiv.
Cognifying Education: Mapping AI's transformative role in emotional, creative, and collaborative learning — Conference Paper, September 2025. By Mikael Gorsky, Ilya Levin. 13th Higher Education Institutions Conference.
Papert's Vision Realized: Constructionism and Generative AI — Article, June 2025. By Ilya Levin, Alexei Semenov, Mikael Gorsky. Constructionism Conference Proceedings.
Smart Learning in the 21st Century: Advancing Constructionism Across Three Digital Epochs — Preprint, January 2025. By Ilya Levin, Alexei Semenov, Mikael Gorsky. arXiv.

The AI Pravda — LinkedIn newsletter

Critical analysis of artificial intelligence and its effect on work and society. 5,500+ subscribers. The complete archive of 103 issues (2023–2026) is published in full at mgorsky.net/theaipravda.

Subscribe to The AI Pravda on LinkedIn

Pro bono

AI for seniors — free workshop

Helping older adults use everyday AI tools. Delivered to Russian-speaking communities in Israel.

For older adults, artificial intelligence is about preserving quality of life, maintaining autonomy, and sustaining the feeling of independence that defines dignified aging. For seniors who have emigrated, AI becomes a bridge: it can translate documents, explain official letters, help compose emails in the local language, and guide users through government websites. The workshop has been delivered to Russian-speaking communities in Israel, where participants — many of them in their 70s and 80s — discovered that AI could help them read Hebrew documents and communicate with Israeli institutions.

Startup competitions — unpaid time

Judging and mentoring early-stage ventures. Helping teams clarify their value proposition, assess technical feasibility, and prepare for the realities of scaling an AI product.

AC/VC LinkedIn group — community

A group for developers and students working with coding agents. The community shares practical insights, code examples, tool comparisons, and honest assessments of what works in production.

Join the AC/VC LinkedIn group

Recent

31 May 2026 — Agentic Software Engineering preprint on arXiv
2026 — paper accepted, Higher Education Institutions Conference