pith. sign in

arxiv: 2606.04543 · v1 · pith:G5GMQ55Snew · submitted 2026-06-03 · 💻 cs.CY

Agentic AI and Pedagogical Best Practice: The Tension Between Automation and Learning

Steve Woollaston , Brendan Flanagan , Isanka Wijerathne , Hiroaki Ogata This is my paper

Pith reviewed 2026-06-28 04:18 UTC · model grok-4.3

classification 💻 cs.CY
keywords agentic AIAI in educationlearner agencypedagogical principlesscaffoldingmetacognitionformative assessmentintentional friction
0
0 comments X

The pith

Agentic AI risks undermining learner agency and cognitive effort unless designs prioritize intentional friction, dynamic scaffolding, human oversight, and selective use.

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

The paper reviews how proactive AI agents that initiate and pursue goals could automate educational tasks that build skills through effort. It examines six established pedagogical principles—prior knowledge activation, collaborative learning, problem-based learning, formative assessment, scaffolding, and metacognition—and shows where agentic AI might bypass the cognitive work those principles require. The central tension is that convenience from automation can reduce learner independence and deep processing. The authors propose concrete design choices to keep AI in a supporting role rather than replacing the learner's effort. A sympathetic reader would care because these systems are already appearing in classrooms and could reshape how students develop agency if left unchecked.

Core claim

Agentic AI shifts from passive tools to goal-directed agents that can initiate interactions; when applied to education this creates a tension with learning because it can reduce the agency and cognitive effort learners normally expend. Examining the six principles reveals specific points of risk, such as AI handling problem-solving steps that should remain with students or providing scaffolds that never fade. The paper therefore advances design recommendations that embed intentional friction, adjust scaffolding dynamically, keep humans in the loop, and limit AI to appropriate moments so that automation supports rather than supplants human learning.

What carries the argument

The application of the six pedagogical principles (prior knowledge activation, collaborative learning, problem-based learning, formative assessment, scaffolding, metacognition) as a lens for identifying where agentic AI automation displaces learner effort.

If this is right

  • AI agents should insert deliberate friction rather than complete tasks automatically.
  • Scaffolding must remain dynamic and gradually withdraw to preserve learner effort.
  • Human oversight must stay active rather than fully delegated to the agent.
  • AI use should be selective so that only suitable tasks are automated.

Where Pith is reading between the lines

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

  • Real-world classroom trials could measure whether the recommended designs produce measurable differences in long-term skill retention compared with unchecked agentic systems.
  • The same friction-and-oversight logic could apply to workplace training environments where automation risks reducing employee problem-solving capacity.
  • Future agent designs might embed explicit prompts that require learners to articulate their own reasoning before the AI intervenes.

Load-bearing premise

That these six principles identify the main ways agentic AI affects learning outcomes and that the listed design recommendations will maintain balance without needing direct empirical tests of their impact.

What would settle it

A study that measures learner agency, metacognitive skill, and knowledge retention in matched groups using agentic AI with versus without the recommended friction and oversight features.

read the original abstract

Artificial intelligence in education is evolving from passive chatbots to proactive AI agents capable of initiation and goal-directed interactions. While offering opportunities for personalised learning, this shift risks undermining learner agency and cognitive effort. This paper reviews six pedagogical principles-prior knowledge activation, collaborative learning, problem-based learning, formative assessment, scaffolding, and metacognition-through the lens of agentic AI. We discuss the tension between automation and learning, proposing design recommendations that prioritise intentional friction, dynamic scaffolding, human-in-the-loop oversight, and considered AI utilisation to ensure AI supports rather than supplants human learning.

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 reviews six pedagogical principles (prior knowledge activation, collaborative learning, problem-based learning, formative assessment, scaffolding, and metacognition) through the lens of agentic AI in education. It identifies risks that proactive AI agents may undermine learner agency and cognitive effort, and proposes four design recommendations—intentional friction, dynamic scaffolding, human-in-the-loop oversight, and considered AI utilisation—to ensure AI supports rather than supplants human learning.

Significance. If the proposed design recommendations can be shown to mitigate the identified risks to learner agency, the paper would offer useful conceptual guidance for the design of agentic AI tools in education. As a literature-based discussion without new data or derivations, its primary value would lie in synthesizing tensions between automation and established pedagogical mechanisms.

major comments (2)
  1. [Design recommendations] The manuscript asserts that the four design recommendations (intentional friction, dynamic scaffolding, human-in-the-loop oversight, and considered AI utilisation) will counteract the risks to learner agency and cognitive effort identified in the six principles, but supplies no formal mapping, counter-example analysis, or qualitative case study demonstrating this effectiveness (see the section proposing design recommendations).
  2. [Review of pedagogical principles] The claim that the six listed pedagogical principles capture the primary mechanisms through which agentic AI affects learning outcomes is presented without justification or evidence that these principles are exhaustive or the most load-bearing for agency risks (see the review of pedagogical principles).
minor comments (1)
  1. [Abstract] The abstract is dense; splitting the final sentence would improve readability.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive feedback on our conceptual review. We address the two major comments point by point below, noting the manuscript's scope as a literature synthesis without new empirical data.

read point-by-point responses

  1. Referee: [Design recommendations] The manuscript asserts that the four design recommendations (intentional friction, dynamic scaffolding, human-in-the-loop oversight, and considered AI utilisation) will counteract the risks to learner agency and cognitive effort identified in the six principles, but supplies no formal mapping, counter-example analysis, or qualitative case study demonstrating this effectiveness (see the section proposing design recommendations).

    Authors: As a conceptual review without new data or derivations, the recommendations are derived through logical analysis of the tensions identified for each principle rather than through empirical testing. We do not claim to demonstrate effectiveness via mapping or case studies, which would require a separate empirical study. We will revise to add an explicit limitations subsection clarifying the theoretical nature of the proposals and outlining the need for future validation work. revision: partial

  2. Referee: [Review of pedagogical principles] The claim that the six listed pedagogical principles capture the primary mechanisms through which agentic AI affects learning outcomes is presented without justification or evidence that these principles are exhaustive or the most load-bearing for agency risks (see the review of pedagogical principles).

    Authors: The six principles were selected as representative examples from established pedagogical frameworks (e.g., constructivism and self-regulated learning) where agency and cognitive effort are central. They are not asserted to be exhaustive. We will add a brief justification paragraph in the introduction explaining the selection criteria based on their relevance to automation risks and prominence in the education literature. revision: yes

Circularity Check

0 steps flagged

No circularity: conceptual review draws on external pedagogical literature

full rationale

The manuscript is a discussion paper that reviews six established pedagogical principles (prior knowledge activation, collaborative learning, problem-based learning, formative assessment, scaffolding, metacognition) drawn from standard literature and proposes four design recommendations (intentional friction, dynamic scaffolding, human-in-the-loop oversight, considered AI utilisation) as conceptual safeguards. No equations, fitted parameters, self-citations, or uniqueness theorems are present. The derivation chain consists of narrative mapping from the reviewed principles to the recommendations; these steps are not reduced by construction to the paper's own inputs or prior self-work. The central claims remain independent assertions open to external empirical testing rather than self-referential closures.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

As a review paper the central discussion rests on the selection and relevance of six standard pedagogical principles as load-bearing for the tension analysis; no free parameters, new axioms beyond domain assumptions, or invented entities are introduced.

axioms (1)
  • domain assumption The six pedagogical principles (prior knowledge activation, collaborative learning, problem-based learning, formative assessment, scaffolding, and metacognition) are the key mechanisms relevant to agentic AI effects on learning.
    The paper structures its review explicitly around these six principles as the lens for analysis.

pith-pipeline@v0.9.1-grok · 5628 in / 1127 out tokens · 28921 ms · 2026-06-28T04:18:03.373329+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

27 extracted references

  1. [1]

    Agentic AI: Autonomous intelligence for complex goals—a comprehensive survey.IEEE Access: Practical Innovations, Open Solutions, 13:18912–18936, 2025

    Deepak Bhaskar Acharya, Karthigeyan Kuppan, and B Divya. Agentic AI: Autonomous intelligence for complex goals—a comprehensive survey.IEEE Access: Practical Innovations, Open Solutions, 13:18912–18936, 2025

    2025

  2. [2]

    Te wh¯ ariki: An early childhood curriculum in a superdiverse New Zealand.New Zealand Journal of Educational Studies, 54(2):245–259, 2019

    Angel Chan. Te wh¯ ariki: An early childhood curriculum in a superdiverse New Zealand.New Zealand Journal of Educational Studies, 54(2):245–259, 2019

    2019

  3. [3]

    Cognitive offloading or cognitive overload? how AI alters the mental architecture of coping.Frontiers in Psychology, 16:1699320, 2025

    Ginto Chirayath, K Premamalini, and Jeena Joseph. Cognitive offloading or cognitive overload? how AI alters the mental architecture of coping.Frontiers in Psychology, 16:1699320, 2025

    2025

  4. [4]

    Scaffolding generative AI as a tutor: A quasi-experimental study of learning outcomes and motivational, cognitive and metacognitive processes.Education Sciences, 16(4):651, 2026

    Melanou Chrysanthi and Beege Maik. Scaffolding generative AI as a tutor: A quasi-experimental study of learning outcomes and motivational, cognitive and metacognitive processes.Education Sciences, 16(4):651, 2026

    2026

  5. [5]

    Proactive conversational AI: A comprehensive survey of advancements and opportunities.ACM Transactions on Information Systems, 43(3):1–45, 2025

    Yang Deng, Lizi Liao, Wenqiang Lei, Grace Hui Yang, Wai Lam, and Tat-Seng Chua. Proactive conversational AI: A comprehensive survey of advancements and opportunities.ACM Transactions on Information Systems, 43(3):1–45, 2025

    2025

  6. [6]

    J Dewey.How we think: A restatement of reflective thinking to the educative process. D.C. Heath and Company, 1933

    1933

  7. [7]

    The use of active learning strategies to foster effective teaching in higher education institutions.Zanco for the Humanities, 28(4):328–351, 2024

    Aryan Hussein Sulaiman Dzaiy and Saman Ahmed Abdullah. The use of active learning strategies to foster effective teaching in higher education institutions.Zanco for the Humanities, 28(4):328–351, 2024

    2024

  8. [8]

    Five high-impact teaching practices: A list of possibilities.Collected Essays on Learning and Teaching, 9:3, 2016

    L Dee Fink. Five high-impact teaching practices: A list of possibilities.Collected Essays on Learning and Teaching, 9:3, 2016

    2016

  9. [9]

    Springer Nature, 2025

    Antonio Gull´ ı.Agentic design patterns: a hands-on guide to building intelligent systems. Springer Nature, 2025

    2025

  10. [10]

    Erica R Hamilton, Joshua M Rosenberg, and Mete Akcaoglu. The substitution augmentation mod- ification redefinition (SAMR) model: A critical review and suggestions for its use.TechTrends: For Leaders in Education & Training, 60(5):433–441, 2016. 7 S. Woollaston et al. Agentic AI and Pedagogical Best Practice

    2016

  11. [11]

    Active learning in engineering education

    Marcela Hern´ endez-de Men´ endez, Antonio Vallejo Guevara, Juan Carlos Tud´ on Mart´ ınez, Diana Hern´ endez Alc´ antara, and Ruben Morales-Menendez. Active learning in engineering education. a review of fundamentals, best practices and experiences.International Journal on Interactive Design and Manufacturing (IJIDeM), 13(3):909–922, 2019

    2019

  12. [12]

    Great pedagogy: Nine claims from research

    Chris Husbands and Jo Pearce. Great pedagogy: Nine claims from research. Research report, National College for School Leadership, 2012

    2012

  13. [13]

    Routledge, 2014

    Mary James and Andrew Pollard.TLRP’s ten principles for effective pedagogy: rationale, develop- ment, evidence, argument and impact, pages 14–67. Routledge, 2014

    2014

  14. [14]

    Learning from productive failure.Learning: Research and Practice, 1(1):51–65, 2015

    Manu Kapur. Learning from productive failure.Learning: Research and Practice, 1(1):51–65, 2015

    2015

  15. [15]

    PhD thesis, University of Eastern Finland, 2020

    Mahmud Hasan Khan.Epistemic Network Analysis in Problem-Based Learning (PBL). PhD thesis, University of Eastern Finland, 2020

    2020

  16. [16]

    Scaffolding problem solving in technology-enhanced learning environments (TELEs): Bridging research and theory with practice.Computers & Education, 56(2):403–417, 2011

    Minchi C Kim and Michael J Hannafin. Scaffolding problem solving in technology-enhanced learning environments (TELEs): Bridging research and theory with practice.Computers & Education, 56(2):403–417, 2011

    2011

  17. [17]

    Student success skills: An evidence-based cognitive and social change theory for student achievement.Journal of Education, 192(2-3):89–99, 2012

    Matthew E Lemberger, Greg Brigman, Linda Webb, and Molly M Moore. Student success skills: An evidence-based cognitive and social change theory for student achievement.Journal of Education, 192(2-3):89–99, 2012

    2012

  18. [18]

    Learned helplessness: Theory and evidence.Journal of Experimental Psychology: General, 105(1):3, 1976

    Steven F Maier and Martin E Seligman. Learned helplessness: Theory and evidence.Journal of Experimental Psychology: General, 105(1):3, 1976

    1976

  19. [19]

    Active learning strategies: A mini review of evidence- based approaches.Acta Pedagogia Asiana, 4(1):43–54, 2025

    Maria Eugenia Martinez and Valeria Gomez. Active learning strategies: A mini review of evidence- based approaches.Acta Pedagogia Asiana, 4(1):43–54, 2025

    2025

  20. [20]

    where is agency moving to?

    Ana Mouta, Ana Mar´ ıa Pinto-Llorente, and Eva Mar´ ıa Torrecilla-S´ anchez. “where is agency moving to?”: Exploring the interplay between AI technologies in education and human agency.Digital Society: Ethics, Socio-Legal and Governance of Digital Technology, 4(2), 2025

    2025

  21. [21]

    Preservice student views of teacher judgement and practice in the age of artificial intelligence.Southern African Review of Education, 25(1):70–88, 2019

    N Petersen and J Batchelor. Preservice student views of teacher judgement and practice in the age of artificial intelligence.Southern African Review of Education, 25(1):70–88, 2019

    2019

  22. [22]

    Distributed scaffolding: Helping students learn science from design.Cognition and Instruction, 21(1):1–47, 2003

    S Puntambekar and J L Kolodner. Distributed scaffolding: Helping students learn science from design.Cognition and Instruction, 21(1):1–47, 2003

    2003

  23. [23]

    Principles of instruction: Research-based strategies that all teachers should know.American Educator, 36(1):12, 2012

    Barak Rosenshine. Principles of instruction: Research-based strategies that all teachers should know.American Educator, 36(1):12, 2012

    2012

  24. [24]

    Thinking-fast, slow, and artificial: How AI is reshaping human reasoning and the rise of cognitive surrender.Available at SSRN 6097646, 2026

    Steven D Shaw and Gideon Nave. Thinking-fast, slow, and artificial: How AI is reshaping human reasoning and the rise of cognitive surrender.Available at SSRN 6097646, 2026

    2026

  25. [25]

    Do not worry that generative AI may compromise human creativity or intelli- gence in the future: It already has.Journal of Intelligence, 12(7):69, 2024

    Robert J Sternberg. Do not worry that generative AI may compromise human creativity or intelli- gence in the future: It already has.Journal of Intelligence, 12(7):69, 2024

    2024

  26. [26]

    Scaffolding learning: Principles for effective teaching and the design of classroom resources

    Keith S Taber. Scaffolding learning: Principles for effective teaching and the design of classroom resources. InEffective teaching and learning: Perspectives, strategies and implementation, pages 1–43. Nova Science Publishers, 2018

    2018

  27. [27]

    Reflections on pedagogy: A journey of collaboration.Journal of Management Education, 35(1):154–167, 2011

    Christine Woods. Reflections on pedagogy: A journey of collaboration.Journal of Management Education, 35(1):154–167, 2011. 8

    2011