pith. sign in

arxiv: 2605.07717 · v2 · pith:6B66HKCWnew · submitted 2026-05-08 · 💻 cs.SE · cs.AI

The AI-Native Large-Scale Agile Software Development Manifesto

Ricardo Britto , Fredrik Palmgren , Nishrith Saini , Marcus Ohlin This is my paper

Pith reviewed 2026-05-25 06:24 UTC · model grok-4.3

classification 💻 cs.SE cs.AI
keywords large-scale agileAI-native developmentsoftware development manifestoagile principlesAI agents in software engineeringadaptive software processesintent-driven developmentverification-first assurance
0
0 comments X

The pith

A manifesto redefines large-scale agile development by making AI a first-class participant instead of a peripheral tool.

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

The paper introduces the AI-Native Large-Scale Agile Software Development Manifesto to address why true agility at scale remains difficult under current human-centric frameworks. It grounds the manifesto in six principles that move away from coordination meetings, artifact synchronization, and sequential handoffs. The shift treats AI systems as active participants that enable parallel processes, intent-driven coordination, living knowledge bases, verification-first assurance, orchestrated agent workforces, and reusable blueprints. A sympathetic reader would see this as turning development into an intelligent, adaptive, continuously learning system. The claim matters because existing large-scale frameworks still rely on manual, document-heavy practices that limit real-time adaptation even as AI tools advance rapidly in software engineering.

Core claim

The AI-Native Large-Scale Agile Software Development Manifesto consists of six principles—parallel processes, intent-driven teams, living knowledge, verification-first assurance, orchestrated agent workforces, and reusable blueprints—that together replace meeting-driven, document-heavy, sequential processes with an intelligent, adaptive, continuously learning system in which AI acts as a first-class participant.

What carries the argument

The six principles of the manifesto, which redefine organization when AI becomes a first-class participant.

If this is right

  • Coordination shifts from scheduled meetings to intent-driven interactions mediated by AI.
  • Knowledge artifacts become continuously updated living systems rather than static documents.
  • Verification occurs first and continuously instead of at the end of phases.
  • Agent workforces are orchestrated as reusable, scalable components.
  • Development processes gain the capacity for real-time adaptation through parallel AI-supported flows.

Where Pith is reading between the lines

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

  • Team roles would likely evolve from role-based handoffs to oversight of AI-orchestrated workflows.
  • Existing large-scale frameworks such as SAFe or LeSS would require substantial redesign to incorporate the six principles rather than incremental AI tool adoption.
  • Training data for the AI components would need to capture not only code but also intent and verification history to sustain the living-knowledge principle.
  • Organizations adopting the manifesto might first test it on greenfield projects before retrofitting legacy scaled-agile setups.

Load-bearing premise

AI systems can reliably handle coordination, knowledge maintenance, verification, and orchestration at scale without creating new coordination or verification problems.

What would settle it

A controlled deployment in which AI agents attempt to maintain living knowledge and orchestrate work across multiple teams for six months while measuring the rate of introduced coordination failures or verification gaps against a human-only baseline.

Figures

Figures reproduced from arXiv: 2605.07717 by Fredrik Palmgren, Marcus Ohlin, Nishrith Saini, Ricardo Britto.

Figure 1
Figure 1. Figure 1: AI-native R&D overview: human intent and machine velocity in cross-functional end-to [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: The relationship between humans, AI personas, agents, skills, MCP connectors, and the [PITH_FULL_IMAGE:figures/full_fig_p006_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: AI-native R&D blueprint: a cross-functional end-to-end team supported by a local agent [PITH_FULL_IMAGE:figures/full_fig_p007_3.png] view at source ↗
read the original abstract

Despite the widespread adoption of agile methods, achieving true agility at scale remains elusive. Large-scale agile frameworks remain largely human-centric and manual, relying on coordination meetings, artifact synchronization, and role-based handoffs that inhibit real-time adaptation. Meanwhile, rapid advances in AI, particularly large language models, have begun transforming software engineering, yet their potential for organizational-level agility remains underexplored. We present the AI-Native Large-Scale Agile Software Development Manifesto: a set of values and principles that redefine how large-scale software development is organized when AI becomes a first-class participant rather than a peripheral tool. The manifesto is grounded in six principles, parallel processes, intent-driven teams, living knowledge, verification-first assurance, orchestrated agent workforces, and reusable blueprints, that together shift development from a meeting-driven, document-heavy, sequential process to an intelligent, adaptive, continuously learning system.

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

1 major / 0 minor

Summary. The paper presents the AI-Native Large-Scale Agile Software Development Manifesto, which consists of six principles—parallel processes, intent-driven teams, living knowledge, verification-first assurance, orchestrated agent workforces, and reusable blueprints—intended to redefine large-scale software development by making AI a first-class participant, shifting from meeting-driven, document-heavy processes to an intelligent, adaptive, continuously learning system.

Significance. If the proposed principles prove effective in practice, the manifesto could serve as a foundational document for integrating AI into agile methodologies at scale, potentially influencing software engineering practices and research directions. As a conceptual contribution without empirical data or formal validation, its significance lies in stimulating debate and guiding future implementations rather than providing immediately actionable or tested guidance.

major comments (1)
  1. [Abstract] Abstract: The central assumption that AI systems can reliably act as first-class participants in coordination, knowledge maintenance, verification, and orchestration without introducing new coordination or verification problems is presented as a premise but lacks any supporting analysis, references to related work on multi-agent systems challenges, or discussion of potential pitfalls; this assumption is load-bearing for the manifesto's claims of achieving true agility.

Simulated Author's Rebuttal

1 responses · 0 unresolved

We thank the referee for the constructive comment on the abstract. We address the point below and propose targeted revisions to strengthen the presentation of the manifesto's foundational assumptions.

read point-by-point responses

  1. Referee: [Abstract] Abstract: The central assumption that AI systems can reliably act as first-class participants in coordination, knowledge maintenance, verification, and orchestration without introducing new coordination or verification problems is presented as a premise but lacks any supporting analysis, references to related work on multi-agent systems challenges, or discussion of potential pitfalls; this assumption is load-bearing for the manifesto's claims of achieving true agility.

    Authors: We agree that the assumption regarding AI systems acting as reliable first-class participants is central to the manifesto and is presented without explicit supporting analysis or caveats in the abstract. As a conceptual manifesto rather than an empirical study, the paper's purpose is to articulate a forward-looking vision grounded in observed trends in AI and agile practices. However, we recognize that referencing challenges from multi-agent systems literature (e.g., coordination failures, verification overheads, and emergent behaviors) and briefly noting potential pitfalls would provide necessary balance and context. We will revise the abstract to acknowledge the assumption explicitly and add a dedicated subsection in the full manuscript discussing related work on multi-agent challenges along with potential limitations of the proposed principles. revision: partial

Circularity Check

0 steps flagged

No significant circularity

full rationale

The paper is a conceptual manifesto that defines six principles for AI-native large-scale agile development without any formal derivations, equations, quantitative predictions, or self-referential definitions. The central claim consists solely in proposing the principles themselves as a shift from human-centric processes; no load-bearing step reduces by construction to fitted parameters, prior self-citations, or renamed inputs. The document contains no derivation chain that could be inspected for circularity.

Axiom & Free-Parameter Ledger

0 free parameters · 2 axioms · 0 invented entities

The proposal rests on domain assumptions about AI capability and organizational behavior rather than new free parameters or invented physical entities. No quantitative fitting occurs.

axioms (2)
  • domain assumption AI systems (particularly LLMs) can function as reliable first-class participants in coordination, knowledge, and verification tasks at organizational scale.
    Invoked in the abstract when stating that AI becomes a first-class participant rather than a peripheral tool.
  • domain assumption Current large-scale agile frameworks are limited by human-centric manual processes that inhibit real-time adaptation.
    Stated as the motivation for the manifesto in the opening sentences.

pith-pipeline@v0.9.0 · 5679 in / 1261 out tokens · 22977 ms · 2026-05-25T06:24:04.632088+00:00 · methodology

Review history (2 revisions) →

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

20 extracted references · 20 canonical work pages

  1. [1]

    2021 , issn =

    Adapting Behavior Driven Development (BDD) for large-scale software systems , journal =. 2021 , issn =. doi:https://doi.org/10.1016/j.jss.2021.110944 , author =

    work page doi:10.1016/j.jss.2021.110944 2021

  2. [2]

    Empirical Software Engineering , volume =

    Dingsøyr, Torgeir and Moe, Nils Brede and Fægri, Tor Erlend and Seim, Eva Amdahl , title =. Empirical Software Engineering , volume =

    work page

  3. [3]

    Journal of Systems and Software , volume =

    Rolland, Colette and Prakash, Naveen and Ben Achour, Camille , title =. Journal of Systems and Software , volume =

    work page

  4. [4]

    and Lines, Mark , title =

    Ambler, Scott W. and Lines, Mark , title =

    work page

  5. [5]

    Larman, Craig and Vodde, Bas , title =

    work page

  6. [6]

    2020 , howpublished =

    work page 2020

  7. [7]

    , title =

    Guo, Philip J. , title =. Communications of the ACM , volume =

    work page

  8. [8]

    IEEE Software , volume =

    Zhang, Jie and Wang, Xin , title =. IEEE Software , volume =

    work page

  9. [9]

    arXiv preprint , year =

    Dingsøyr, Torgeir and Moe, Nils Brede and Seim, Eva Amdahl and Fægri, Tor Erlend , title =. arXiv preprint , year =

    work page

  10. [10]

    Behavior-Driven Development , howpublished =

    work page

  11. [11]

    2025 , howpublished =

    Khomenko, Kateryna , title =. 2025 , howpublished =

    work page 2025

  12. [12]

    Agent Skills , howpublished =

    work page

  13. [13]

    Dingsøyr, N

    T. Dingsøyr, N. B. Moe, T. E. Fægri, and E. A. Seim, ``Exploring software development at the very large scale: a revelatory case study and research agenda for agile method adaptation,'' Empirical Software Engineering, vol. 23, no. 1, pp. 490--520, 2018

    work page 2018

  14. [14]

    Rolland, N

    C. Rolland, N. Prakash, and C. Ben Achour, ``Challenges of large-scale agile: a systems perspective,'' Journal of Systems and Software, vol. 125, pp. 1--14, 2016

    work page 2016

  15. [15]

    , `` SAFe 5.0 framework overview,'' https://scaledagileframework.com, 2020

    Scaled Agile Inc. , `` SAFe 5.0 framework overview,'' https://scaledagileframework.com, 2020

    work page 2020

  16. [16]

    Larman and B

    C. Larman and B. Vodde, Large-Scale Scrum: More with LeSS. 1em plus 0.5em minus 0.4em Addison-Wesley, 2016

    work page 2016

  17. [17]

    S. W. Ambler and M. Lines, Disciplined Agile Delivery: A Practitioner's Guide to Agile Software Delivery in the Enterprise. 1em plus 0.5em minus 0.4em IBM Press, 2012

    work page 2012

  18. [18]

    P. J. Guo, ``The state of AI -assisted programming: emerging opportunities and challenges,'' Communications of the ACM, vol. 66, no. 11, pp. 68--77, 2023

    work page 2023

  19. [19]

    Irshad, R

    M. Irshad, R. Britto, and K. Petersen, ``Adapting behavior driven development (bdd) for large-scale software systems,'' Journal of Systems and Software, vol. 177, p. 110944, 2021

    work page 2021

  20. [20]

    AgentSkills , ``Agent skills,'' https://agentskills.io/

    work page