pith. sign in

arxiv: 2605.22349 · v1 · pith:JIEDOL2Znew · submitted 2026-05-21 · 💻 cs.SE

At What Cost? Software Developers' Well-Being in the Age of GenAI

Mariam Guizani , Maduka Subasinghage , Sherlock A. Licorish , Sofia Ouhbi This is my paper

Pith reviewed 2026-05-22 04:15 UTC · model grok-4.3

classification 💻 cs.SE
keywords generative AIsoftware developerswell-beingcognitive loadburnoutresearch agendaprofessional norms
0
0 comments X

The pith

GenAI tools amplify cognitive load and oversight labor for software developers, raising stress and burnout risks.

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

This position paper examines how generative AI is changing software development beyond productivity gains. It points out that these tools add mental strain through new tasks like checking AI suggestions, push for faster output, and shift job expectations, which can lead to exhaustion and poor work-life balance. It also discusses how this affects who enters the field and creates uneven access. Readers should care because focusing only on speed might harm the people doing the work and the quality of software over time.

Core claim

GenAI tools can amplify cognitive load, introduce new forms of oversight labor, and escalate expectations around output and pace, contributing to stress, burnout, and diminished work-life balance, while also transforming professional norms, altering career entry points, demanding continuous adaptation, and deepening inequalities in access and support.

What carries the argument

A theoretical framework to investigate GenAI in software development that centers human experience, social context, and sustainable productivity instead of only performance metrics.

If this is right

  • Investigations into GenAI must include effects on developer mental health and well-being.
  • New oversight tasks from AI outputs require study for their contribution to workload.
  • Research should address how GenAI changes career paths and adaptation needs for developers.
  • Efforts to reduce inequalities in GenAI access and support are needed to prevent widening gaps.

Where Pith is reading between the lines

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

  • If the mechanisms hold, development teams may need policies to limit pace expectations and provide training on effective AI use.
  • Similar well-being impacts could appear in other knowledge work fields adopting GenAI, suggesting broader studies.
  • Companies adopting GenAI might see higher turnover if burnout increases without mitigation strategies.
  • Future work could test the framework by measuring well-being metrics in GenAI-using vs non-using developer groups.

Load-bearing premise

That the negative effects on well-being from amplified cognitive load and new oversight labor are common and large enough across the industry to justify shifting the whole research focus away from performance metrics.

What would settle it

Empirical data from surveys or experiments showing that software developers using GenAI report similar or lower levels of stress and burnout compared to those not using it, or no change from pre-GenAI times.

Figures

Figures reproduced from arXiv: 2605.22349 by Maduka Subasinghage, Mariam Guizani, Sherlock A. Licorish, Sofia Ouhbi.

Figure 1
Figure 1. Figure 1: Proposed Theoretical Framework the context of GenAI [PITH_FULL_IMAGE:figures/full_fig_p004_1.png] view at source ↗
read the original abstract

Generative Artificial Intelligence (GenAI) is rapidly reshaping software development, with growing emphasis on accelerating productivity and optimizing performance. However, excessive focus on such dimensions risks overlooking the critical implications for developer well-being. GenAI tools can amplify cognitive load, introduce new forms of oversight labor, and escalate expectations around output and pace, contributing to stress, burnout, and diminished work-life balance. The GenAI movement is also transforming professional norms, altering career entry points, demanding continuous adaptation, and deepening inequalities in access and support. This position paper calls for a reorientation of the GenAI research agenda in software development and proposes a theoretical framework to move beyond narrow performance metrics toward investigations that also center on human experience, social context, and sustainable productivity.

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. This position paper argues that the emphasis on productivity gains from GenAI in software development overlooks critical impacts on developer well-being. It describes mechanisms by which GenAI can amplify cognitive load, introduce oversight labor, escalate output expectations and pace, thereby contributing to stress, burnout, and reduced work-life balance. The paper further claims that GenAI is transforming professional norms, altering career entry points, requiring continuous adaptation, and exacerbating inequalities in access and support. It calls for reorienting the GenAI research agenda in software engineering toward investigations that center human experience, social context, and sustainable productivity, and proposes a theoretical framework to support this shift.

Significance. If the described mechanisms prove prevalent and net-negative relative to pre-GenAI baselines, the position could usefully broaden the research agenda beyond narrow performance metrics. The call for a theoretical framework that incorporates human and social dimensions is a constructive suggestion for the field, though its impact hinges on whether the framework is developed with sufficient specificity and testability in future work.

major comments (2)
  1. [Abstract] Abstract and main argument: The claims that GenAI amplifies cognitive load, introduces new oversight labor, and escalates expectations are presented as established effects driving stress and burnout, yet the text provides no citations to empirical studies, developer surveys, time-use data, or longitudinal comparisons with pre-GenAI conditions. This leaves the magnitude and direction of the asserted harms unquantified and makes the call for full research-agenda reorientation rest on descriptive assertion rather than evidenced prevalence.
  2. [Abstract] Proposed theoretical framework: The manuscript states that a theoretical framework is proposed to move beyond performance metrics, but offers no outline of its core constructs, relationships, or how it would operationalize well-being outcomes. Without this detail, it is difficult to assess whether the framework can address the load-bearing assumptions about mechanism prevalence and net impact.
minor comments (1)
  1. [Abstract] The abstract and position statement would benefit from explicit references to related work in software engineering or HCI on developer well-being (e.g., prior studies on cognitive load or burnout) to situate the normative claims.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the careful reading and constructive feedback on our position paper. We address each major comment below, clarifying the scope of a position paper while committing to revisions that strengthen grounding and specificity without altering the core argument.

read point-by-point responses

  1. Referee: [Abstract] Abstract and main argument: The claims that GenAI amplifies cognitive load, introduces new oversight labor, and escalates expectations are presented as established effects driving stress and burnout, yet the text provides no citations to empirical studies, developer surveys, time-use data, or longitudinal comparisons with pre-GenAI conditions. This leaves the magnitude and direction of the asserted harms unquantified and makes the call for full research-agenda reorientation rest on descriptive assertion rather than evidenced prevalence.

    Authors: We agree that the abstract presents the mechanisms in a manner that could be read as asserting established effects. As this is a position paper, the claims are intended as reasoned extrapolations from industry reports, early empirical work on AI coding assistants, and documented patterns in software engineering labor. To address the concern directly, we will revise the abstract and introduction to (a) add citations to relevant studies on cognitive load and developer experience with GenAI tools, (b) explicitly frame the mechanisms as hypothesized pathways supported by emerging evidence rather than definitively quantified harms, and (c) emphasize that the paper's primary contribution is a call for the empirical work needed to test prevalence and net impact. These changes will make the evidential basis clearer while preserving the position that the research agenda should expand beyond performance metrics. revision: partial

  2. Referee: [Abstract] Proposed theoretical framework: The manuscript states that a theoretical framework is proposed to move beyond performance metrics, but offers no outline of its core constructs, relationships, or how it would operationalize well-being outcomes. Without this detail, it is difficult to assess whether the framework can address the load-bearing assumptions about mechanism prevalence and net impact.

    Authors: The referee correctly notes that the current manuscript mentions the framework at a high level without providing an outline of constructs or operationalization. In the revised version we will add a dedicated subsection that (1) defines the core constructs (cognitive load amplification, oversight labor, expectation escalation, norm shifts, and access inequalities), (2) describes the proposed relationships among these constructs and their links to well-being and sustainable productivity, and (3) sketches initial operationalization approaches, such as mixed-methods protocols combining validated well-being scales with time-use diaries and longitudinal cohort studies. This expansion will allow readers to evaluate the framework's potential to support the agenda reorientation we advocate. revision: yes

Circularity Check

0 steps flagged

Position paper makes normative claims without derivations or self-referential reductions

full rationale

This is a position paper advancing a call to reorient the GenAI research agenda in software engineering. It asserts mechanisms such as amplified cognitive load and new oversight labor based on logical description and normative framing, without equations, fitted parameters, uniqueness theorems, or any derivation chain. No self-citations appear in the provided text, and the central claims do not reduce by construction to their own inputs. The argument is self-contained as a perspective piece and does not rely on circular steps.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The central argument rests on domain assumptions about GenAI's negative effects on developers; no free parameters, invented entities, or mathematical axioms are introduced.

axioms (1)
  • domain assumption GenAI tools amplify cognitive load, introduce oversight labor, and escalate output expectations in software development
    Invoked in the abstract as the basis for claiming contributions to stress and burnout without supporting measurements.

pith-pipeline@v0.9.0 · 5666 in / 1294 out tokens · 42890 ms · 2026-05-22T04:15:48.713942+00:00 · methodology

discussion (0)

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

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

What do these tags mean?
matches
The paper's claim is directly supported by a theorem in the formal canon.
supports
The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends
The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses
The paper appears to rely on the theorem as machinery.
contradicts
The paper's claim conflicts with a theorem or certificate in the canon.
unclear
Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

26 extracted references · 26 canonical work pages

  1. [1]

    Saleema Amershi, Andrew Begel, Christian Bird, Robert DeLine, Harald Gall, Ece Kamar, Nachiappan Nagappan, Besmira Nushi, and Thomas Zimmermann. 2019. Software engineering for machine learning: A case study. In2019 IEEE/ACM 41st At What Cost? Software Developers’ Well-Being in the Age of GenAI FSE Companion ’26, July 05–09, 2026, Montreal, QC, Canada Inte...

    work page 2019

  2. [2]

    Joel Becker, Nate Rush, Elizabeth Barnes, and David Rein. 2025. Measuring the impact of early-2025 AI on experienced open-source developer productivity. arXiv preprint arXiv:2507.09089(2025)

    work page arXiv 2025

  3. [3]

    Licorish, Christoph Treude, and Markus Wagner

    Scott Blyth, Sherlock A. Licorish, Christoph Treude, and Markus Wagner. 2025. Static Analysis as a Feedback Loop: Enhancing LLM-Generated Code Beyond Correctness. In2025 IEEE International Conference on Source Code Analysis & Manipulation (SCAM). 100–109. doi:10.1109/SCAM67354.2025.00017

    work page doi:10.1109/scam67354.2025.00017 2025

  4. [4]

    Tiziana C Callari and Lucia Puppione. 2025. Can generative artificial intelligence productivity tools support workplace learning? A qualitative study on employee perceptions in a multinational corporation.Journal of Workplace Learning37, 3 (2025), 266–283

    work page 2025

  5. [5]

    Noelle Chesley. 2014. Information and communication technology use, work intensification and employee strain and distress.Work, employment and society 28, 4 (2014), 589–610

    work page 2014

  6. [6]

    TOI Tech Desk. 2026. Nvidia CEO Jensen Huang to engineers: I want you to stop coding and start...Times of India Tech Desk (2026). doi:technology/tech-news/nvidia-ceo-jensen-huang-to-engineers-i- want-you-to-stop-coding-and-start-/amp_articleshow/126543905.cms

    work page arXiv 2026

  7. [7]

    Zixuan Feng, Sadia Afroz, and Anita Sarma. 2025. From Gains to Strains: Modeling Developer Burnout with GenAI Adoption.arXiv preprint arXiv:2510.07435(2025)

    work page arXiv 2025

  8. [8]

    2025.Quality of Life Framework for Canada

    Government of Canada. 2025.Quality of Life Framework for Canada. https://www.canada.ca/en/government/system/government-wide-reporting- spending-operations/government-spending/quality-life-framework- canada.html

    work page 2025

  9. [9]

    John Hopkins and Sean Gallagher. 2024. Generative AI at work: Empowering employee mental wellbeing. (2024)

    work page 2024

  10. [10]

    Nanna Inie, Jeanette Falk, and Steve Tanimoto. 2023. Designing participatory ai: Creative professionals’ worries and expectations about generative ai. InExtended Abstracts of the 2023 CHI Conference on Human Factors in Computing Systems. 1–8

    work page 2023

  11. [11]

    Will I be replaced?

    Mohammad Amin Kuhail, Sujith Samuel Mathew, Ashraf Khalil, Jose Berengueres, and Syed Jawad Hussain Shah. 2024. “Will I be replaced?” Assessing ChatGPT’s effect on software development and programmer perceptions of AI tools.Science of Computer Programming235 (2024), 103111

    work page 2024

  12. [12]

    Sherlock A Licorish, Ansh Bajpai, Chetan Arora, Fanyu Wang, and Kla Tan- tithamthavorn. 2025. Comparing human and LLM generated code: The jury is still out!arXiv preprint arXiv:2501.16857(2025)

    work page arXiv 2025

  13. [13]

    Phelipe Silva Malheiros, Rayfran Rocha Lima, and Ana Carolina Oran. 2024. Impact of Generative AI Technologies on Software Development Professionals’ Perceptions of Job Security. InProceedings of the XXIII Brazilian Symposium on Software Quality. 169–178

    work page 2024

  14. [14]

    Melissa Mazmanian, Wanda J Orlikowski, and JoAnne Yates. 2013. The autonomy paradox: The implications of mobile email devices for knowledge professionals. Organization science24, 5 (2013), 1337–1357

    work page 2013

  15. [15]

    Amr Mohamed, Maram Assi, and Mariam Guizani. 2025. The impact of LLM- assistants on software developer productivity: A systematic literature review. arXiv preprint arXiv:2507.03156(2025)

    work page arXiv 2025

  16. [16]

    Charles A O’Reilly III and Michael L Tushman. 2013. Organizational ambidexter- ity: Past, present, and future.Academy of management Perspectives27, 4 (2013), 324–338

    work page 2013

  17. [17]

    Srishti Palani and Gonzalo Ramos. 2024. Evolving roles and workflows of creative practitioners in the age of generative AI. InProceedings of the 16th Conference on Creativity & Cognition. 170–184

    work page 2024

  18. [18]

    Apoorva Pandey, Mridul Maheshwari, and Nishtha Malik. 2025. A systematic literature review on employee well-being: Mapping multi-level antecedents, moderators, mediators and future research agenda.Acta Psychologica258 (2025), 105080

    work page 2025

  19. [19]

    Corina Pelau, Dan-Cristian Dabija, and Mihaela Stanescu. 2024. Can I trust my AI friend? The role of emotions, feelings of friendship and trust for consumers’ information-sharing behavior toward AI.Oeconomia Copernicana15, 2 (2024), 407–433

    work page 2024

  20. [20]

    2025.Technostress: Orgs Give Employees AI, then Demand More Work

    David Ramel. 2025.Technostress: Orgs Give Employees AI, then Demand More Work. https://virtualizationreview.com/articles/2025/06/03/technostress-orgs- give-employees-ai-then-demand-more-work.aspx Accessed: 2025-01-18

    work page 2025

  21. [21]

    Siddharth Sarkar, Vikas Menon, Susanta Padhy, and Preethy Kathiresan. 2024. Mental health and well-being at the workplace.Indian Journal of Psychiatry66, Suppl 2 (2024), S353–S364

    work page 2024

  22. [22]

    Ben Shneiderman. 2020. Human-centered artificial intelligence: Reliable, safe & trustworthy.International Journal of Human–Computer Interaction36, 6 (2020), 495–504

    work page 2020

  23. [23]

    Alex Singla, Alexander Sukharevsky, Bryce Hall, Lareina Yee, Michael Chui, and Tara Balakrishnan. 2025. The State of AI: How Organizations Are Rewiring to Capture Value. https://www.mckinsey.com/capabilities/quantumblack/our- insights/the-state-of-ai. McKinsey & Company. Accessed: 2025-03-22

    work page 2025

  24. [24]

    Macy Takaffoli, Sijia Li, and Ville Mäkelä. 2024. Generative AI in user experience design and research: how do UX practitioners, teams, and companies use GenAI in industry?. InProceedings of the 2024 ACM Designing Interactive Systems Conference. 1579–1593

    work page 2024

  25. [25]

    Monideepa Tarafdar, Qiang Tu, Bhanu S Ragu-Nathan, and TS Ragu-Nathan

    work page

  26. [26]

    The impact of technostress on role stress and productivity.Journal of management information systems24, 1 (2007), 301–328

    work page 2007