pith. sign in

arxiv: 2606.01152 · v1 · pith:AO4622UUnew · submitted 2026-05-31 · 💻 cs.CY · cs.AI· cs.SE

ASE-26: a curriculum for agentic software engineering as a discipline

Mikael Gorsky This is my paper

Pith reviewed 2026-06-28 16:36 UTC · model grok-4.3

classification 💻 cs.CY cs.AIcs.SE
keywords agentic software engineeringundergraduate curriculumAI agent directionsoftware engineering educationpractitioner skillsevolutionary spiralco-evolution of intent and build
0
0 comments X

The pith

Structured undergraduate curricula in agentic software engineering close the practitioner skills gap by teaching direction of agents rather than code writing.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper establishes that software engineering work has shifted toward directing agents, backed by data showing 75-79 percent automation exposure and employment declines in exposed roles. Academic literature identifies the missing element as structured practitioner discipline rather than better models. ASE-26 supplies a 21-module undergraduate curriculum organized around the evolutionary spiral of co-evolving intent and build, along with methods for grading agent-co-produced work. A sympathetic reader would care because the position frames education as the direct mechanism to supply the exact skills industry lacks. The curriculum is constructed to remain useful even after current model capabilities change.

Core claim

The practitioner skills the industry currently lacks are precisely the skills the discipline of agentic software engineering names, and structured undergraduate curricula in agentic software engineering are the principal mechanism by which the gap closes. The paper sets out the discipline framing, the evolutionary spiral as the operational form of the co-evolution of intent and build, the twenty-one-module structure, the pedagogical commitments for grading agent-co-produced work, what graduates leave with, and how the taught discipline is designed to outlast specific model capabilities.

What carries the argument

The evolutionary spiral, the operational form of the co-evolution of intent and build that structures the agentic software engineering discipline for teaching.

If this is right

  • Graduates possess the ability to direct agents in ways that address the automation rates reported in industry data.
  • The twenty-one-module structure provides a systematic way to teach the discipline at undergraduate level.
  • Grading methods accommodate work co-produced with agents while preserving academic integrity.
  • The curriculum design ensures the taught discipline remains relevant beyond the capabilities of any particular current models.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The same curriculum structure could be adapted for agentic work in adjacent fields such as data science or systems administration.
  • Industry-university partnerships focused on agent direction skills could speed adoption of the proposed training.
  • Longitudinal tracking of ASE-26 graduates' task-level output would supply direct evidence on whether the curriculum narrows the documented gap.
  • The emphasis on enduring discipline over model specifics suggests parallel curricula may be needed wherever agents take over routine production tasks.

Load-bearing premise

The academic literature converges on the finding that the missing capability is structured practitioner discipline rather than better models.

What would settle it

A study tracking productivity and employment outcomes for graduates of ASE-26-style programs versus those without such training, while holding model access constant, that shows no measurable closure of the skills gap.

read the original abstract

The work of a professional software engineer has begun to consist, increasingly, of directing agents rather than writing code, and the empirical evidence for the shift is now several years deep. Anthropic's Economic Index puts automation at 79 per cent of Claude Code interactions [2]; Handa and colleagues at Anthropic find AI exposure for Computer Programmer tasks at approximately 75 per cent of the role's distinct activities [3]; Brynjolfsson and colleagues at Stanford's Digital Economy Lab report a 13 per cent relative decline in employment for workers aged 22 to 25 in occupations most exposed to AI [4]. The shift is also unfinished, and the academic literature on agentic software engineering converges on the finding that the missing capability is not better models but structured practitioner discipline. This paper presents ASE-26, a comprehensive undergraduate curriculum for agentic software engineering as a discipline, deposited as a citable reference on Zenodo under CC BY-ND 4.0 [12]. The paper sets out the discipline framing the curriculum rests on, the conceptual contributions it makes (most importantly, the evolutionary spiral as the operational form of the co-evolution of intent and build), the twenty-one-module structure that organises the discipline for teaching, the pedagogical commitments that follow from grading work co-produced with an agent, what graduates leave with, and how the discipline as taught is designed to outlast the specific capabilities of today's models. The position the paper takes is that the practitioner skills the industry currently lacks are precisely the skills the discipline names, and that structured undergraduate curricula in agentic software engineering are the principal mechanism by which the gap closes.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

3 major / 2 minor

Summary. The paper proposes ASE-26, a 21-module undergraduate curriculum for agentic software engineering as a discipline. It frames the shift toward directing AI agents (citing automation rates of ~79% and ~75% task exposure plus employment declines from [2,3,4]) as requiring structured practitioner discipline rather than better models, asserts that the academic literature converges on this diagnosis, and presents the curriculum—organized around an evolutionary spiral of intent-build co-evolution, specific grading commitments for agent-co-produced work, and design for longevity beyond current models—as the principal mechanism to close the resulting skills gap. The curriculum is deposited on Zenodo under CC BY-ND 4.0.

Significance. If the curriculum were shown to be effective, it would provide a concrete, citable undergraduate framework for training in agent-directed software engineering and could help standardize practitioner skills in an emerging area. The explicit deposition of the full curriculum as a reusable artifact and the articulation of an evolutionary-spiral operational model are positive contributions to accessibility and conceptual clarity. However, the absence of any empirical validation, outcome measures, or independent grounding for the curriculum's design substantially limits the significance of the proposal as presented.

major comments (3)
  1. [Abstract] Abstract: The claim that 'the academic literature on agentic software engineering converges on the finding that the missing capability is not better models but structured practitioner discipline' is unsupported by the only citations supplied ([2] Anthropic Economic Index, [3] Handa et al., [4] Brynjolfsson et al.). These works quantify automation percentages, task exposure, and employment trends but contain no analysis or conclusion regarding the relative importance of model improvements versus practitioner discipline or curricula. This premise is load-bearing for the central argument that the proposed curriculum is the principal mechanism to close the skills gap.
  2. [Abstract] Abstract: The assertion that 'the practitioner skills the industry currently lacks are precisely the skills the discipline names' is circular; the paper supplies no independent survey, meta-analysis, or additional references establishing that the 21-module structure or evolutionary spiral directly maps onto verified industry deficiencies beyond the three external statistics already cited.
  3. [Abstract] Abstract and curriculum description: No section provides empirical validation, pilot data, outcome metrics, or controlled comparison demonstrating that the 21-module structure, grading commitments for agent-co-produced work, or evolutionary spiral improves practitioner capability or closes any measured skills gap.
minor comments (2)
  1. The paper would benefit from an explicit literature-review subsection or table summarizing additional agentic-SE sources to support (or qualify) the convergence claim.
  2. Notation for the evolutionary spiral and the 21 modules could be clarified with a single summary diagram or table early in the manuscript.

Simulated Author's Rebuttal

3 responses · 1 unresolved

We thank the referee for the detailed and constructive report. The comments correctly identify overstatements in the abstract and the absence of empirical validation. We will revise the manuscript to address the first two points by tempering unsupported claims while preserving the curriculum proposal and evolutionary spiral as conceptual contributions. The third point reflects the scope of this work as a design proposal.

read point-by-point responses

  1. Referee: [Abstract] Abstract: The claim that 'the academic literature on agentic software engineering converges on the finding that the missing capability is not better models but structured practitioner discipline' is unsupported by the only citations supplied ([2] Anthropic Economic Index, [3] Handa et al., [4] Brynjolfsson et al.). These works quantify automation percentages, task exposure, and employment trends but contain no analysis or conclusion regarding the relative importance of model improvements versus practitioner discipline or curricula. This premise is load-bearing for the central argument that the proposed curriculum is the principal mechanism to close the skills gap.

    Authors: We accept this critique. The citations establish the scale of the shift to agent-directed work but do not address the relative importance of models versus practitioner discipline. We will revise the abstract to frame the diagnosis as the position advanced in this paper rather than an established convergence in the literature, removing the unsupported assertion while retaining the motivation for proposing the curriculum. revision: yes

  2. Referee: [Abstract] Abstract: The assertion that 'the practitioner skills the industry currently lacks are precisely the skills the discipline names' is circular; the paper supplies no independent survey, meta-analysis, or additional references establishing that the 21-module structure or evolutionary spiral directly maps onto verified industry deficiencies beyond the three external statistics already cited.

    Authors: We agree the phrasing is circular. The 21-module structure is derived from our analysis of skills needed to direct agents, but no independent mapping or survey is supplied. We will revise the abstract and relevant sections to present the modules as a proposed structure addressing the documented shift, without claiming precise equivalence to externally verified deficiencies. revision: yes

  3. Referee: [Abstract] Abstract and curriculum description: No section provides empirical validation, pilot data, outcome metrics, or controlled comparison demonstrating that the 21-module structure, grading commitments for agent-co-produced work, or evolutionary spiral improves practitioner capability or closes any measured skills gap.

    Authors: This manuscript proposes a curriculum design and does not include empirical validation, which would require implementation and longitudinal data collection beyond the scope of the current work. We will add a limitations section acknowledging this and outlining potential future evaluation methods for the evolutionary spiral and grading commitments. revision: partial

standing simulated objections not resolved
  • No section provides empirical validation, pilot data, outcome metrics, or controlled comparison demonstrating that the 21-module structure, grading commitments for agent-co-produced work, or evolutionary spiral improves practitioner capability or closes any measured skills gap.

Circularity Check

0 steps flagged

No circularity in derivation chain

full rationale

The paper advances a curriculum proposal grounded in external citations for automation trends and employment effects. No self-citation load-bearing step, self-definitional reduction, fitted-input prediction, or ansatz smuggling appears in the abstract or described structure. The claim that literature converges on practitioner discipline over model improvements is asserted without meta-analysis, but this is an evidentiary gap rather than a circular reduction of the curriculum's 21-module or evolutionary-spiral elements to their own inputs by construction. The derivation remains independent of the target result.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

The paper is a curriculum proposal and introduces no free parameters, axioms, or invented entities in a scientific sense; all content is descriptive design.

pith-pipeline@v0.9.1-grok · 5825 in / 1050 out tokens · 29830 ms · 2026-06-28T16:36:54.210369+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

16 extracted references · 1 canonical work pages

  1. [1]

    speed vs. trust

    The work has changed; the curriculum has not Open the chat window of any modern coding agent, whether Claude Code, Codex or Cursor, and watch a working developer ask for a feature: they write a paragraph, the agent reads it, asks two clarifying questions and then produces six hundred lines of code, a handful of tests, a commit message and a draft pull req...

    2026

  2. [2]

    demands a radical reimagining of the foundational pillars of SE

    Agentic software engineering as a discipline The opening move of ASE-26 is to insist that agentic software engineering is a discipline, and the word matters: a discipline has principles, named artefacts, recognised failure modes and standards a practitioner can be held to, and it carries from one product to the next because what it teaches is the structur...

    2026

  3. [3]

    entering as additional dimensions that shape how the turn is structured when complexity warrants. Two judgment skills make or break the rhythm, and the curriculum names both: the commit point is the decision at the end of each turn about what to lock, what goes back into the specification, and what remains open, while drift detection is the skill of recog...

    1993

  4. [4]

    bread in the AI sandwich

    What graduates leave with A graduate of ASE-26 is able to do twelve things the course names explicitly, and although the list lives in full in the curriculum’s learning-outcomes section [12], the summary form covers the conceptual, operational and evaluative capabilities the discipline requires. On the conceptual side, the graduate can articulate agentic ...

    2026

  5. [5]

    What lasts and what doesn’t The curriculum is built to be revised, and this is, by design, the most important thing the curriculum teaches about itself. The closing module is structured around an explicit inventory of which principles taught in the course are likely durable and which are tied to current capabilities: the durable list includes co-evolution...

    1962

  6. [6]

    Which economic tasks are performed with AI? Evidence from millions of Claude conversations,

    K. Handa et al. (Anthropic), “Which economic tasks are performed with AI? Evidence from millions of Claude conversations,” arXiv preprint arXiv:2503.04761,

    arXiv

  7. [7]

    Agentic software engineering: Foundational pillars and a research roadmap,

    A. E. Hassan, H. Li, B. Adams (Queen’s University), D. Lin, D. Qiu (Huawei Canada), T.-H. Chen (Concordia University), and Y. Kashiwa (Nara Institute of Science and Technology), “Agentic software engineering: Foundational pillars and a research roadmap,” arXiv preprint arXiv:2509.06216,

    Pith/arXiv arXiv

  8. [8]

    Toward agentic software engineering beyond code: Vision, values, vocabulary,

    R. Hoda et al. (Monash University), “Toward agentic software engineering beyond code: Vision, values, vocabulary,” in Proc. 47th Int. Conf. Software Engineering (ICSE), 2026, arXiv:2510.19692. 12

    arXiv 2026

  9. [9]

    From correctness to collaboration: A human-centered taxonomy of AI agent behavior in software engineering,

    T. Dong, S. Shi, H. Sampath, and A. Macvean (Google), “From correctness to collaboration: A human-centered taxonomy of AI agent behavior in software engineering,” in Extended Abstracts of the 2026 CHI Conference on Human Factors in Computing Systems (CHI EA ’26),

    2026

  10. [10]

    ASE-26: A conceptual framework, curriculum, and teaching foundation for agentic software engineering,

    M. Gorsky (Holon Institute of Technology), “ASE-26: A conceptual framework, curriculum, and teaching foundation for agentic software engineering,” Version 1.0, Zenodo, 2026, doi: 10.5281/zenodo.20468021. Available: https://doi.org/10.5281/zenodo.20468021

    work page doi:10.5281/zenodo.20468021 2026

  11. [11]

    Formalising design exploration as co-evolution,

    (Republished from the 1999 Cross & Dorst conference paper; see also M. L. Maher, J. Poon, and S. Boulanger (University of Sydney), “Formalising design exploration as co-evolution,” in Advances in Formal Design Methods for CAD, Springer, 1996.)

    1999

  12. [12]

    On the impact of AGENTS.md files,

    (Anonymous authors), “On the impact of AGENTS.md files,” arXiv preprint arXiv:2601.20404,

    arXiv

  13. [13]

    Evaluating AGENTS.md: Are context files helpful?,

    (Anonymous authors), “Evaluating AGENTS.md: Are context files helpful?,” arXiv preprint arXiv:2602.11988,

    Pith/arXiv arXiv

  14. [14]

    GPTs are GPTs: An early look at the labor market impact potential of large language models,

    T. Eloundou, S. Manning, P. Mishkin, and D. Rock (OpenAI), “GPTs are GPTs: An early look at the labor market impact potential of large language models,” arXiv preprint arXiv:2303.10130,

    arXiv

  15. [15]

    D. E. Knuth (Stanford University), The Art of Computer Programming, Volume 1: Fundamental Algorithms, 3rd ed. Addison-Wesley, 1997 (orig. 1968). G. Polya (Stanford University), How to Solve It: A New Aspect of Mathematical Method. Princeton University Press,

    1997

  16. [16]

    OWASP Top 10 for Large Language Model Applications,

    OWASP, “OWASP Top 10 for Large Language Model Applications,” 2025, and “OWASP Top 10 for Agentic Applications,” Dec

    2025