pith. machine review for the scientific record. sign in

arxiv: 2605.11027 · v1 · submitted 2026-05-10 · 💻 cs.SE · cs.AI

Recognition: no theorem link

From Code-Centric to Intent-Centric Software Engineering: A Reflexive Thematic Analysis of Generative AI, Agentic Systems, and Engineering Accountability

Authors on Pith no claims yet

Pith reviewed 2026-05-13 01:22 UTC · model grok-4.3

classification 💻 cs.SE cs.AI
keywords generative AIagentic systemssoftware engineeringreflexive thematic analysisintent-centric developmentengineering accountabilitysocio-technical systems
0
0 comments X

The pith

Generative AI is shifting software engineering from isolated code authorship to specifying intent and governing human-agent systems.

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

The paper performs a reflexive thematic analysis on peer-reviewed literature, technical benchmarks, public talks, product announcements, and online discourse to map how generative AI and agentic systems alter software engineering practice. It establishes that lower costs for generating plausible code elevate the importance of intent specification, context curation, architecture knowledge, verification, security, provenance, governance, and accountable human judgment. A sympathetic reader would care because the change reframes daily engineering work from writing code to supervising socio-technical systems of humans, agents, tools, and evidence gates. If the transition holds, speed-focused adoption risks hidden technical debt and accountability gaps, whereas bounded autonomy can sustain quality, security, maintainability, and trust.

Core claim

The analysis shows that GenAI lowers the cost of producing plausible code while increasing the importance of intent specification, context curation, architecture knowledge, verification, security, provenance, governance, and accountable human judgment. The findings indicate that software engineering is becoming less about isolated code authorship and more about supervising, validating, and governing socio-technical systems of humans, agents, tools, and evidence gates. This matters because speed-focused adoption can accumulate hidden technical debt and accountability gaps, whereas bounded autonomy can preserve quality, security, maintainability, and trust.

What carries the argument

Reflexive thematic analysis of a corpus combining peer-reviewed software engineering and AI literature, technical benchmarks, public talks, essays, product announcements, and X discourse, organized through a corpus register, codebook, coding matrix, and traceability tables to surface themes of the shift from code-centric to intent-centric work.

If this is right

  • Engineers will need stronger skills in intent specification, context curation, and architecture knowledge.
  • Verification, security, provenance, governance, and accountable judgment will grow in importance over raw code production.
  • Unchecked speed in adopting agentic tools can create hidden technical debt and responsibility gaps.
  • Structured use of bounded autonomy can maintain software quality, security, maintainability, and trust.

Where Pith is reading between the lines

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

  • Development teams may need explicit protocols for evidence gates and human sign-off to manage agent outputs at scale.
  • The same intent-over-code pattern could appear in adjacent fields such as data pipeline engineering or embedded systems design.
  • Longitudinal tracking of job postings and project metrics could test whether oversight time is measurably replacing coding time.

Load-bearing premise

The chosen corpus of peer-reviewed literature, public talks, product announcements, and online discourse accurately and representatively captures the profession's near-term transition without significant selection or interpretive bias.

What would settle it

A broad industry survey or time-allocation study showing that practicing engineers still devote most of their effort to direct code authorship rather than intent specification, verification, and oversight of agents would contradict the claimed transition.

read the original abstract

Generative artificial intelligence (GenAI) and agentic systems are moving software engineering from code-centric production toward intent-centric human-agent work in which natural language, repository context, tools, tests, and governance shape delivery. Prior studies examine code generation, AI pair programming, and software engineering agents, but less is known about how public technical discourse and peer-reviewed evidence together frame the profession's near-term transition. This study addresses that gap through a reflexive thematic analysis (RTA) dominant and interpretative phenomenological analysis (IPA) informed public-discourse and document analysis. The corpus combines peer-reviewed software engineering and AI literature, technical benchmarks, public talks and interviews, essays, product-facing technical announcements, and X-originated discourse from prominent AI and software engineering voices. Sources were organized through a corpus register, codebook, coding matrix, theme-to-source traceability table, DOI/reference audit, and reproducibility protocol. The analysis shows that GenAI lowers the cost of producing plausible code while increasing the importance of intent specification, context curation, architecture knowledge, verification, security, provenance, governance, and accountable human judgment. The findings indicate that software engineering is becoming less about isolated code authorship and more about supervising, validating, and governing socio-technical systems of humans, agents, tools, and evidence gates. This matters because speed-focused adoption can accumulate hidden technical debt and accountability gaps, whereas bounded autonomy can preserve quality, security, maintainability, and trust.

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

2 major / 1 minor

Summary. The paper conducts a reflexive thematic analysis (RTA) informed by interpretative phenomenological analysis on a corpus of peer-reviewed SE/AI literature, benchmarks, public talks, essays, product announcements, and X discourse. It claims that GenAI and agentic systems are shifting software engineering from isolated code authorship to intent-centric supervision of socio-technical systems, where engineers focus on natural-language intent specification, context curation, verification, security, provenance, governance, and accountable human judgment. The analysis highlights risks of hidden technical debt from speed-focused adoption and advocates bounded autonomy to preserve quality and trust.

Significance. If the interpretive claims hold, the work provides a timely synthesis of how public technical discourse frames the profession's near-term evolution under GenAI, with direct implications for engineering education, tool design, accountability frameworks, and risk mitigation in AI-augmented workflows. Credit is due for the structured qualitative apparatus (corpus register, codebook, coding matrix, theme-to-source traceability table, DOI audit, and reproducibility protocol), which supports transparency and traceability in a mixed-methods discourse analysis.

major comments (2)
  1. Corpus construction (Methods): The corpus explicitly incorporates X-originated discourse from prominent voices and product-facing announcements. This selection risks systematic over-representation of promotional or hype-aligned narratives, which directly undermines the load-bearing claim that the sources 'accurately and representatively capture the profession's near-term transition.' The reproducibility protocol and traceability table address downstream coding but do not provide evidence that the initial register mitigates visibility bias, selection effects, or under-representation of dissenting or day-to-day practice perspectives.
  2. Analysis transparency (Methods): While the paper states that sources were organized via a codebook, coding matrix, and theme-to-source traceability table, these artifacts are not presented or exemplified. Combined with limited detail on how reflexivity was enacted and how IPA elements shaped theme development from the RTA-dominant process, it is difficult to evaluate the rigor of the interpretive leap from corpus to the central claim that SE is becoming 'less about isolated code authorship and more about supervising... socio-technical systems.'
minor comments (1)
  1. Abstract: Several sentences are long and compound; splitting them would improve readability without altering meaning.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for these constructive comments on methodological rigor. We address each major point below and will incorporate revisions to strengthen transparency and address potential biases.

read point-by-point responses
  1. Referee: Corpus construction (Methods): The corpus explicitly incorporates X-originated discourse from prominent voices and product-facing announcements. This selection risks systematic over-representation of promotional or hype-aligned narratives, which directly undermines the load-bearing claim that the sources 'accurately and representatively capture the profession's near-term transition.' The reproducibility protocol and traceability table address downstream coding but do not provide evidence that the initial register mitigates visibility bias, selection effects, or under-representation of dissenting or day-to-day practice perspectives.

    Authors: We acknowledge the risk of visibility and selection bias from including X discourse and product announcements, which may skew toward prominent or promotional voices. Our intent was to capture the public technical discourse influencing the profession, balanced by triangulation with peer-reviewed literature, benchmarks, and essays. The corpus register includes source justifications and a DOI audit. In revision, we will expand the Methods section with an explicit rationale for source selection, a discussion of potential biases (including under-representation of dissenting views), and how the multi-source design mitigates them. We will also add a limitations subsection on corpus construction and consider adding counter-examples from day-to-day practitioner sources where identifiable. revision: yes

  2. Referee: Analysis transparency (Methods): While the paper states that sources were organized via a codebook, coding matrix, and theme-to-source traceability table, these artifacts are not presented or exemplified. Combined with limited detail on how reflexivity was enacted and how IPA elements shaped theme development from the RTA-dominant process, it is difficult to evaluate the rigor of the interpretive leap from corpus to the central claim that SE is becoming 'less about isolated code authorship and more about supervising... socio-technical systems.'

    Authors: We agree that exemplification of the analytical artifacts and greater detail on reflexivity and IPA integration would improve evaluability. The codebook, coding matrix, and traceability table exist but were omitted from the main text for brevity; the reproducibility protocol references them. In the revised manuscript, we will include representative excerpts from the codebook and coding matrix in the Methods section, along with a sample from the theme-to-source table. We will also expand the description of reflexivity (e.g., iterative team positionality discussions and bracketing of assumptions) and clarify how IPA-informed elements (such as attention to experiential framing in discourse) informed theme refinement within the RTA process, making the interpretive steps more traceable. revision: yes

Circularity Check

0 steps flagged

No circularity: claims derived interpretively from external corpus

full rationale

The paper performs a reflexive thematic analysis (RTA) informed by interpretative phenomenological analysis on a mixed external corpus of peer-reviewed SE/AI literature, benchmarks, public talks, product announcements, essays, and X discourse. The central claim—that SE is shifting from code-centric authorship to intent-centric supervision of socio-technical systems—is explicitly framed as an output of this analysis, supported by a corpus register, codebook, coding matrix, theme-to-source traceability table, DOI audit, and reproducibility protocol. No equations, fitted parameters, predictions, or self-citations appear as load-bearing steps in the provided derivation. The findings do not reduce to the inputs by construction; they are presented as traceable interpretations of the collected sources. This matches the default expectation for non-circular qualitative work and aligns with the reader's assessment of minimal circularity.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

This is a qualitative interpretive study relying on established methods of reflexive thematic analysis and document analysis; no numerical free parameters, new entities, or ad-hoc axioms are introduced beyond standard assumptions of qualitative research.

axioms (1)
  • domain assumption Reflexive thematic analysis yields valid insights into professional and technical discourse when applied to a curated corpus of literature and public sources.
    Invoked in the description of the RTA-dominant and IPA-informed approach.

pith-pipeline@v0.9.0 · 5566 in / 1217 out tokens · 30182 ms · 2026-05-13T01:22:34.884644+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

34 extracted references · 34 canonical work pages

  1. [1]

    Alves, Nicolli S. R. and Mendes, Thiago S. and Mendon. Identification and Management of Technical Debt: A Systematic Mapping Study , journal =. 2016 , doi =

  2. [2]

    Proceedings of the 35th International Conference on Software Engineering , pages =

    Bacchelli, Alberto and Bird, Christian , title =. Proceedings of the 35th International Conference on Software Engineering , pages =. 2013 , doi =

  3. [3]

    , title =

    Bowen, Glenn A. , title =. Qualitative Research Journal , volume =. 2009 , doi =

  4. [4]

    Qualitative Research in Psychology , volume =

    Braun, Virginia and Clarke, Victoria , title =. Qualitative Research in Psychology , volume =. 2006 , doi =

  5. [5]

    Qualitative Research in Sport, Exercise and Health , volume =

    Braun, Virginia and Clarke, Victoria , title =. Qualitative Research in Sport, Exercise and Health , volume =. 2019 , doi =

  6. [6]

    Counselling and Psychotherapy Research , volume =

    Braun, Virginia and Clarke, Victoria , title =. Counselling and Psychotherapy Research , volume =. 2020 , doi =

  7. [7]

    Palliative Medicine , volume =

    Braun, Virginia and Clarke, Victoria , title =. Palliative Medicine , volume =. 2024 , doi =

  8. [8]

    , title =

    Brooks, Frederick P. , title =. Computer , volume =. 1987 , doi =

  9. [9]

    Brown, Tom B. and Mann, Benjamin and Ryder, Nick and Subbiah, Melanie and Kaplan, Jared and Dhariwal, Prafulla and Neelakantan, Arvind and Shyam, Pranav and Sastry, Girish and Askell, Amanda and others , title =. Advances in Neural Information Processing Systems , volume =. 2020 , doi =

  10. [10]

    2021 , eprint =

    Chen, Mark and Tworek, Jerry and Jun, Heewoo and others , title =. 2021 , eprint =

  11. [11]

    Information Technology and People , volume =

    Eshraghian, Farjam and Hafezieh, Najmeh and Farivar, Farveh , title =. Information Technology and People , volume =. 2024 , doi =

  12. [12]

    , title =

    Fan, Angela and Gokkaya, Beliz and Harman, Mark and Lyubarskiy, Mitya and Sengupta, Shubho and Yoo, Shin and Zhang, Jie M. , title =. 2023 IEEE/ACM International Conference on Software Engineering: Future of Software Engineering , pages =. 2023 , doi =

  13. [13]

    2023 , eprint =

    Jimenez-Gomez, Carlos and Yang, John and Wettig, Alexander and others , title =. 2023 , eprint =

  14. [14]

    2017 , howpublished =

    Karpathy, Andrej , title =. 2017 , howpublished =

  15. [15]

    Journal of Systems and Software , volume =

    Li, Zengyang and Avgeriou, Paris and Liang, Peng , title =. Journal of Systems and Software , volume =. 2015 , doi =

  16. [16]

    Science , volume =

    Li, Yujia and Choi, David and Chung, Junyoung and Kushman, Nate and Schrittwieser, Julian and Leblond, Remi and Eccles, Tom and Keeling, James and Gimeno, Felix and Dal Lago, Agustin and others , title =. Science , volume =. 2022 , doi =

  17. [17]

    and Guba, Egon G

    Lincoln, Yvonna S. and Guba, Egon G. , title =

  18. [18]

    and Ray, Baishakhi and Roychoudhury, Abhik , title =

    Lyu, Michael R. and Ray, Baishakhi and Roychoudhury, Abhik , title =. ACM Transactions on Software Engineering and Methodology , volume =. 2025 , doi =

  19. [19]

    and Norris, Jill M

    Nowell, Lorelli S. and Norris, Jill M. and White, Deborah E. and Moules, Nancy J. , title =. International Journal of Qualitative Methods , volume =. 2017 , doi =

  20. [20]

    2023 , doi =

    Can Architecture Knowledge Guide Software Development With Generative AI? , journal =. 2023 , doi =

  21. [21]

    , title =

    Parnas, David L. , title =. Communications of the ACM , volume =. 1972 , doi =

  22. [22]

    Proceedings of the 43rd IEEE Symposium on Security and Privacy , pages =

    Pearce, Hammond and Ahmad, Baleegh and Tan, Benjamin and Dolan-Gavitt, Brendan and Karri, Ramesh , title =. Proceedings of the 43rd IEEE Symposium on Security and Privacy , pages =. 2022 , publisher =

  23. [23]

    2023 , eprint =

    Peng, Sida and Kalliamvakou, Eirini and Cihon, Peter and Demirer, Mert , title =. 2023 , eprint =

  24. [24]

    and Tarkoma, Sasu and Klemettinen, Mika and Riekki, Jukka and Doermann, David , title =

    Sauvola, J. and Tarkoma, Sasu and Klemettinen, Mika and Riekki, Jukka and Doermann, David , title =. Automated Software Engineering , volume =. 2024 , doi =

  25. [25]

    and Fieldsend, Megumi , title =

    Smith, Jonathan A. and Fieldsend, Megumi , title =. APA Handbook of Research Methods in Psychology , publisher =. 2021 , doi =

  26. [26]

    and Osborn, Mike , title =

    Smith, Jonathan A. and Osborn, Mike , title =. Qualitative Psychology: A Practical Guide to Research Methods , publisher =. 2004 , doi =

  27. [27]

    How Secure Is AI-Generated Code: A Large-Scale Comparison of Large Language Models , journal =

    Tihanyi, Norbert and Bisztray, Tam\'. How Secure Is AI-Generated Code: A Large-Scale Comparison of Large Language Models , journal =. 2024 , doi =

  28. [28]

    Journal of Documentation , volume =

    VanScoy, Amy and Evenstad, Solveig Beyza Narli , title =. Journal of Documentation , volume =. 2015 , doi =

  29. [29]

    and Kaiser, Lukasz and Polosukhin, Illia , title =

    Vaswani, Ashish and Shazeer, Noam and Parmar, Niki and Uszkoreit, Jakob and Jones, Llion and Gomez, Aidan N. and Kaiser, Lukasz and Polosukhin, Illia , title =. Advances in Neural Information Processing Systems , volume =. 2017 , doi =

  30. [30]

    Xia, Chunqiu Steven and Deng, Yinlin and Dunn, S. M. and others , title =. Proceedings of the ACM on Software Engineering , volume =. 2025 , doi =

  31. [31]

    Science China Information Sciences , volume =

    Zhang, Quanjun and Fang, Chunrong and Xie, Yang and Zhang, Yaxin and Yu, Shengcheng and Sun, Weisong and Yang, Yun and Chen, Zhenyu , title =. Science China Information Sciences , volume =. 2026 , doi =

  32. [32]

    Journal of Systems and Software , volume =

    Zhou, Xiyu and Liang, Peng and Zhang, Beiqi and others , title =. Journal of Systems and Software , volume =. 2025 , doi =

  33. [33]

    2026 , howpublished =

  34. [34]

    2025 , howpublished =