SHAPR: Operationalising Human-AI Collaborative Research Through Structured Knowledge Generation

SHAPR: Op erationalising Human–AI Collab orativ e Researc h Through Structured Kno wledge Generation Ka Ching Chan 1* Sc ho ol of Business, Univ ersity of Southern Queensland, Springeld Cen tral, QLD 4300, Queensland, Australia. Con tributing authors: kc.c han@unisq.edu.au ; Abstract SHAPR (Solo Human-Cen tred and AI-Assisted Practice) is a framework for researc h soft ware developmen t that integrates h uman-cen tred decision-making with AI-assisted capabilities. While prior w ork in tro duced SHAPR as a con- ceptual framew ork, this pap er fo cuses on its operationalisation as a structured, traceable, and knowledge-generating approach to AI-assisted research practice. W e present a set of interconnected mo dels describing how researc h activities are organised through iterativ e cycles (Explore–Build–Use–Ev aluate–Learn), how artefacts ev olv e through dev elopmen t and use, and ho w empirical evidence is systematically transformed into conceptual knowledge. Cen tral to this process is the notion of Structured Kno wledge Units (SKUs), which pro vide mo du- lar and reusable represen tations of insigh ts derived from practice, supp orting kno wledge accumulation and reuse across cycles. The framework further in tro- duces evidence and traceability as a cross-cutting mechanism linking human decisions, AI-assisted dev elopment, artefact evolution, and resulting knowledge, enabling transparency , reproducibility , and systematic renement. SHAPR is also p ositioned as an AI-executable research framework, as its structured pro- cesses and documentation practices can b e in terpreted and enacted by generativ e AI systems to guide research worko ws. At the same time, SHAPR supp orts a con tinuum of AI inv olv ement, allowing researchers to balance control, learning, and automation across dieren t researc h con texts. Bey ond individual worko ws, SHAPR is conceptualised as an integrated research system that combines LLM w orkspaces, dev elopment en vironments, cloud storage, and v ersion con trol to supp ort scalable, traceable, and knowledge-cen tred research practices. Overall, SHAPR provides a practical and theoretically grounded foundation for conduct- ing rigorous, transparen t, and repro ducible researc h in AI-assisted environmen ts, 1 con tributing to the developmen t of scalable and metho dologically sound research practices. Keyw ords: SHAPR, A ction design research, Human–AI collaboration, Softw are developmen t 1 In tro duction The rapid emergence of generative articial intelligence has signicantly transformed ho w researchers design, prototype, and ev aluate digital artefacts. Large language mo dels and AI-assisted developmen t to ols are increasingly capable of generating co de, prop osing architectural designs, and supporting debugging and exp erimen tation. These dev elopments reect a broader shift tow ard AI-augmented research en viron- men ts, where h uman and mac hine capabilities are increasingly in tegrated Shneiderman ( 2022 ). These capabilities enable researchers to dev elop complex research softw are artefacts more rapidly than ev er b efore. Ho w ever, they also in tro duce new methodolog- ical challenges related to the structure, traceability , and repro ducibility of AI-assisted researc h worko ws. In man y contemporary research environmen ts, substan tial p ortions of reason- ing, exp erimen tation, and design exploration o ccur within conv ersational interactions b et w een researchers and generative AI systems. While these interaction environmen ts can accelerate developmen t, they also make it dicult to do cumen t ho w design deci- sions are made and how research knowledge emerges from iterative exp erimen tation. Without a structured approach, AI-assisted developmen t risks b ecoming opaque, dicult to replicate, and challenging to ev aluate within a rigorous research pro cess. Design Science Research (DSR) and Action Design Research (ADR) provide well- established foundations for artefact-based kno wledge generation, emphasising iterativ e dev elopment and ev aluation to pro duce design kno wledge Hevner et al. ( 2004 ); P ef- fers et al. ( 2007 ); Sein et al. ( 2011 ). How ever, these metho dologies were dev elop ed in contexts where softw are developmen t was typically conducted by teams using con- v entional worko ws. They provide limited op erational guidance for researchers who increasingly conduct artefact developmen t individually while in teracting extensively with generativ e AI systems. This creates a metho dological gap: researc hers require structured approac hes for managing AI-assisted, often solo, research soft ware developmen t in a manner that remains traceable, repro ducible, and metho dologically rigorous. Addressing this gap is particularly important as AI to ols become more capable of generating complex artefacts. While generative AI can augment dev elopment activities, research remains fundamen tally a h uman epistemic pro cess in which the researcher retains resp onsibilit y for in terpretation, ev aluation, and knowledge claims. The SHAPR framework (Solo, Human-cen tred, AI-assisted PRactice) was in tro- duced in prior w ork as a conceptual framew ork for structuring AI-assisted research soft ware developmen t Chan ( 2026 ). SHAPR explicitly recognises the interaction 2 w orkspace b et ween human researc hers and generative AI systems as a central com- p onen t of contemporary research practice. Within this framework, AI systems act as collab orativ e tools that supp ort exploration, prototyping, and reasoning, while the h uman researc her maintains epistemic authority ov er design decisions and knowledge generation. SHAPR should not b e understoo d as a new research metho dology . Rather, it functions as an op erationalisation lay er that translates the principles of A ction Design Research into a worko w suitable for mo dern AI-assisted environmen ts. ADR remains the underlying research metho d, while SHAPR provides the practical struc- ture through whic h AI-assisted developmen t activities can b e organised, do cumented, and transformed into research knowledge. A cen tral premise of SHAPR is that iterative dev elopmen t activities can serve as a systematic source of research knowledge when properly structured and do cumen ted. In this framew ork, dev elopment cycles are treated not merely as engineering iterations but as knowledge-generating pro cesses in whic h exploration, artefact ev olution, and ev aluation contribute to the accum ulation of design insights. While prior work in tro duced SHAPR as a conceptual framework, this pap er extends the conceptual foundation in tro duced in prior work Chan ( 2026 ) by fo cus- ing on its op erationalisation, knowledge structures, and practical implementation in AI-assisted research en vironments. Sp ecically , the pap er presents a structured op er- ational worko w that connects human–AI in teraction, dev elopmen t cycles, researc h artefacts, and rep ository-based documentation, enabling exploratory AI-assisted dev elopment to b e transformed into traceable research evidence. The pap er makes four key contributions to AI-assisted research practice: • It op erationalises A ction Design Research for AI-assisted research environmen ts b y extending the traditional Build–Interv ene–Ev aluate cycle into an iterativ e dev elopment reasoning cycle: Explore–Build–Use–Ev aluate–Learn. • It introduces a traceable op erational worko w that links h uman–AI interac- tion workspaces, dev elopment cycles, artefact ev olution, and repository-based do cumen tation, enabling AI-assisted activities to b e systematically recorded as researc h evidence. • It prop oses SHAPR Knowledge Units (SKUs) as a mechanism for transforming dev elopment insigh ts into reusable and progressiv ely generalisable knowledge, supp orting structured knowledge accumulation across cycles. • It conceptualises SHAPR as b oth an AI-executable and system-oriented researc h framew ork, in whic h structured pro cesses can be interpreted by generative AI sys- tems and integrated with developmen t environmen ts, cloud storage, and version con trol to supp ort scalable and traceable research worko ws. T ogether, these con tributions p osition SHAPR as a framework for conducting rigorous, transparent, and scalable AI-assisted research softw are developmen t while preserving human epistemic authority . Imp ortan tly , SHAPR accommo dates v ary- ing lev els of AI in volv emen t, allowing researchers to balance con trol, learning, and automation across dierent research contexts. 3 The remainder of the pap er is organised as follows. Section 2 introduces the core concepts of the SHAPR framework. Section 3 describ es the relationship b et w een researc h practice, artefact ev olution, and kno wledge generation. Section 4 presen ts the SHAPR knowledge transformation cycle. Section 5 in tro duces the operational work- o w that structures AI-assisted developmen t. Section 6 explains SHAPR Kno wledge Units and the process of knowledge accumulation. Section 7 presents the inte- grated SHAPR op erational mo del. Section 8 discusses human–AI collaboration in SHAPR practice. Section 9 pro vides practical guidance for implementation. Section 10 discusses implications for AI-assisted research practice. Section 11 concludes the pap er. 2 SHAPR F ramew ork Ov erview The SHAPR framework (Solo, Human-centred, AI-assisted PRactice), in tro duced in prior work Chan ( 2026 ), pro vides a structured approach for organising researc h soft- w are dev elopment in en vironments where researc hers increasingly collab orate with generativ e AI systems. Rather than introducing a new research metho dology , SHAPR functions as an op erationalisation lay er that translates the principles of A ction Design Researc h (ADR) into a worko w suitable for contemporary AI-assisted developmen t en vironments. In Design Science Researc h (DSR) and ADR, knowledge is generated through the iterativ e dev elopment and ev aluation of artefacts Hevner et al. ( 2004 ); Sein et al. ( 2011 ). These approac hes emphasise the interaction b et ween problem context, arte- fact construction, and ev aluation as a mechanism for producing design knowledge. Ho wev er, their practical application often assumes collab orative dev elopmen t settings and con ven tional softw are engineering worko ws. In contrast, many mo dern research softw are pro jects are conducted by individual researc hers w orking closely with AI-assisted dev elopmen t to ols. Generative AI systems can supp ort code generation, debugging, design exploration, and reasoning ab out implemen tation alternatives. As a result, a substantial p ortion of design exploration no w o ccurs within con v ersational interaction environmen ts rather than traditional do cumen tation pro cesses. SHAPR explicitly incorp orates this human–AI in teraction w orkspace as a core comp onen t of the researc h process. Within the framew ork, AI systems act as cognitiv e collab orators that supp ort exploration and developmen t, while the human researcher retains resp onsibilit y for interpretation, ev aluation, and kno wledge generation. T o structure this pro cess, SHAPR distinguishes b et w een tw o complementary en vironments: • In teraction w orkspace – where researchers explore ideas and collab orate with AI systems during design and developmen t. • Rep ository w orkspace – where artefacts, documentation, and researc h evi- dence are formally recorded. 4 A cen tral principle of SHAPR is that interaction with AI systems b ecomes researc h evidence only when translated in to artefact c hanges and do cumen ted within the repos- itory . This ensures that exploratory reasoning and AI-assisted exp erimen tation remain traceable and repro ducible. A t the core of the framework is a knowledge transformation cycle consisting of the stages Explor e, Build, U se, Evaluate, and L e arn . This cycle structures iterativ e artefact dev elopment and pro vides a mechanism for transforming developmen t activities into researc h knowledge. T o support systematic knowledge generation, SHAPR introduces the concept of the SHAPR K now le dge Unit (SKU) , a structured representation of an insight explaining ho w a design decision inuences artefact b ehaviour. Over successiv e cycles, SKUs accum ulate and may b e synthesised into patterns and design principles, contributing to broader metho dological knowledge. T ogether, these elements form a coherent framew ork that links in teraction w orkspaces, developmen t cycles, artefact ev olution, and kno wledge do cumentation. Through this in tegration, SHAPR enables researchers to conduct rigorous, traceable, and kno wledge-generating softw are developmen t in AI-assisted environmen ts. The k ey concepts of the SHAPR framew ork are summarised b elow: SHAPR F ramework for structuring solo, AI-assisted researc h softw are developmen t. Inter action W orksp ac e En vironment for human–AI collab oration during exploration and developmen t. R ep ository W orksp ac e V ersion-controlled en vironment for storing arte- facts and research evidence. SHAPR Cycle Iterativ e cycle: Explore → Build → Use → Ev al- uate → Learn. SHAPR K now le dge Unit (SKU) Structured insigh t derived from dev elopmen t cycles. K now le dge A c cumulation Pro cess through which SKUs evolv e into patterns and design principles. R ese ar ch A rtefact Soft ware system dev elop ed as part of the researc h pro cess. 3 The Practice–Artefact–Kno wledge Relationship in SHAPR A cen tral premise of the SHAPR framework is that researc h soft ware developmen t can function as a systematic pro cess for generating research kno wledge when the rela- tionship b et w een research practice, artefact developmen t, and knowledge extraction is made explicit. Figure 2 illustrates this relationship as a triadic in teraction b etw een practice, artefacts, and knowledge. Rather than treating artefact construction as a purely technical activit y , SHAPR conceptualises it as an epistemic process through whic h insights emerge from iterative developmen t and use of research softw are. 5 Solo Research Practice Human-Centred Decision Making (Decision Authority) AI-Assisted Development (Cognitive Augmentation) Practice → Artefact → Knowledge (Structured Knowledge Generation) Fig. 1 SHAPR Conceptual F oundation. This gure illustrates the conceptual foundation of SHAPR, where solo research practice in tegrates human-cen tred decision-making and AI-assisted developmen t. These comp onen ts conv erge to pro duce practice-based artefacts that contribute to structured knowl- edge generation, highlighting SHAPR’s emphasis on human accountabilit y supp orted by AI-assisted developmen t. Research Practice (Design & Exploration) Research Artefact (System / Prototype) Research Knowledge (Insights, SKUs, Principles) Build / Design Evaluate / Observe Learn / Inform Iterative Knowledge Generation Cycles through SHAPR Fig. 2 SHAPR Iterativ e Cycle (ADR-Aligned). This gure presents the iterative cycle underly- ing SHAPR, aligned with the build–interv ene–ev aluate logic of Action Design Research. The cycle captures the contin uous interaction b etw een human decision-making and AI-assisted developmen t, supported by iterative renemen t, and highlights how SHAPR op erationalises cyclical research prac- tice in AI-assisted environmen ts. This p erspective aligns with the principles of design-oriented researc h, where arte- facts embo dy theoretical and practical insights while simultaneously providing a medium through whic h new knowledge can b e disco vered. How ev er, SHAPR extends this p erspective by explicitly m odelling how knowledge emerges from the interaction b et w een human-cen tred practice and evolving artefacts in AI-assisted developmen t con texts. 6 In SHAPR, the relationship b etw een these three elemen ts forms a knowledge- pro ducing cycle. Research practice pro duces artefacts through design and exp erimen- tation; artefacts rev eal insights through their b eha viour, p erformance, and use; and these insights inform subsequen t researc h practice. Making this relationship explicit pro vides a conceptual foundation for structuring AI-assisted research as a traceable and kno wledge-generating pro cess. 3.1 Practice as Researc h A ctivity Within SHAPR, practice refers to the activities undertaken by the researcher during the developmen t and exploration of research softw are. These activities include prob- lem framing, design exploration, coding, testing, experimentation, interpretation of outcomes, and interaction with generative AI systems. Practice is therefore not merely implementation work but constitutes the primary site of researc h activity . Through iterative dev elopmen t, researchers explore design alternativ es, examine system b eha viours, and ev aluate p oten tial solutions. Decisions made during these activities—such as selecting algorithms, structuring data ows, or rening system architectures—generate insights that extend b ey ond the immediate artefact. In AI-assisted en vironments, generativ e mo dels expand the design space b y accel- erating co ding, prototyping, and experimentation. How ev er, SHAPR positions these to ols as assistiv e collaborators rather than autonomous agents. The h uman researcher retains resp onsibilit y for framing research questions, in terpreting results, and v alidat- ing knowledge claims. Research practice thus remains fundamentally human-cen tred, with AI augmenting exploratory capability . 3.2 Artefacts as Kno wledge-Carrying Structures The second element of the SHAPR relationship is the researc h artefact, t ypically represen ted by soft ware systems, computational models, analytical to ols, or exp er- imen tal platforms developed during the research process. These artefacts em b ody design decisions and theoretical assumptions. In SHAPR, artefacts are not static outcomes but evolving structures that change across iterative developmen t cycles. As artefacts are extended, tested, and applied, they reveal behaviours and prop erties that may conrm, challenge, or rene the researc her’s understanding of the problem domain. Because artefacts enco de design choices, algorithms, and system architectures, they function as kno wledge-carrying structures. Observing how artefacts behav e under dieren t conditions provides the empirical basis for generating researc h insights. 3.3 Kno wledge Extraction from Artefact Developmen t The third element of the SHAPR relationship is knowledge, which represents insights deriv ed from the interaction b etw een research practice and artefact evolution. Unlike artefacts, this kno wledge is intended to b e transferable b ey ond a single system or implemen tation. 7 SHAPR emphasises the imp ortance of explicitly capturing insigh ts generated during dev elopmen t. T o support this process, the framework introduces Structured Kno wledge Units (SKUs)—documented insigh ts that explain ho w specic design deci- sions inuence artefact behaviour. Each SKU represen ts a traceable unit of kno wledge link ed to particular developmen t activities or ev aluation outcomes. As SKUs accumulate across developmen t cycles, recurring patterns ma y emerge. These patterns can b e synthesised into higher-level design principles or metho dological insigh ts, contributing to the accumulation of research knowledge. 3.4 The Practice–Artefact–Knowledge Cycle The interaction b et ween practice, artefacts, and knowledge forms a contin uous learn- ing lo op. Research practice generates artefacts through design and exp erimentation; artefacts exp ose insights through ev aluation and use; and these insights inform subsequen t developmen t activities. This triadic relationship transforms researc h softw are dev elopmen t in to a struc- tured knowledge generation pro cess. Rather than treating artefacts as nal outputs, SHAPR positions them as evolving instrumen ts for exploration, experimentation, and learning. The next section builds up on this conceptual relationship by in tro ducing the SHAPR knowledge transformation cycle, which op erationalises these interactions through an iterative developmen t pro cess aligned with Action Design Research. 4 The SHAPR Kno wledge T ransformation Cycle T o op erationalise the relationship b et ween practice, artefacts, and kno wledge, the SHAPR framew ork in tro duces an iterative developmen t process referred to as the SHAPR Knowledge T ransformation Cycle. This cycle structures research softw are dev elopment in to a sequence of activities that systematically transform exploratory practice in to do cumen ted research knowledge. Compared to traditional ADR formulations, the SHAPR cycle makes explicit the role of exploration and learning in AI-assisted en vironments, where iterativ e in teraction with generative AI systems pla ys a central role in shaping artefact dev elopment. The cycle consists of ve stages: Explore → Build → Use → Ev aluate → Learn, as sho wn in Figure 3 . Eac h stage represen ts a distinct form of research activity , collectiv ely enabling artefact dev elopment to function as a mechanism for generating insigh ts. 4.1 Explore The Explore stage inv olves inv estigating p oten tial design directions, iden tifying lim- itations in the current artefact, and considering alternativ e approaches. Exploration ma y inv olv e conceptual reasoning, literature reection, system analysis, or interaction with generativ e AI systems to generate p otential design options. 8 💡 Explore ⚙ Build (ADR: Build) ▶ Use (ADR: Intervene) 📊 Evaluate (ADR: Evaluate) 📝 Learn Knowledge T ransformation Cycle Fig. 3 SHAPR Conceptual F oundation. This gure illustrates the conceptual foundation of SHAPR, where solo research practice in tegrates human-cen tred decision-making and AI-assisted developmen t. These tw o comp onen ts conv erge to pro duce practice-based artefacts that contribute to structured knowledge generation. The mo del highlights SHAPR’s emphasis on main taining human accountabilit y while leveraging AI to supp ort developmen t and exploration. This stage is particularly imp ortant in AI-assisted research environmen ts, where con versational interaction with large language mo dels can rapidly expand the design space. How ever, SHAPR emphasises that exploration remains human-guided, with the researc her determining which alternatives are relev ant and worth pursuing. 4.2 Build In the Build stage, the selected design direction is implemented through mo dications to the researc h artefact. This stage may inv olv e writing co de, restructuring system comp onen ts, implemen ting algorithms, or integrating new to ols. Artefact dev elopment in SHAPR is delib erately incremen tal. Rather than attempt- ing to construct complete systems in a single step, the framew ork encourages small, traceable mo dications that can b e clearly ev aluated in subsequent stages. 4.3 Use The Use stage corresponds to the application of the artefact within its in tended con text. This may inv olv e running exp erimen ts, testing system features, analysing datasets, or simulating user worko ws. Using the artefact allows the researcher to observ e how the system b ehav es under realistic conditions. These observ ations often rev eal unexp ected b eha viours, limitations, or opp ortunities for impro vemen t that were not apparent during imple- men tation. 4.4 Ev aluate During the Ev aluate stage, the researcher assesses the outcomes pro duced during the use phase. Ev aluation ma y in volv e analysing system performance, examining usabilit y , iden tifying errors, or comparing alternative design solutions. This stage plays a critical role in transforming developmen t activit y into research evidence. Through systematic ev aluation, the researcher determines whether the implemen ted design ac hieves the intended ob jectiv es and what new insights can b e deriv ed from the observed outcomes. 9 4.5 Learn The nal stage of the cycle is Learn, where insigh ts deriv ed from ev aluation are explicitly captured and do cumen ted. These insights form the basis of SHAPR Knowl- edge Units (SKUs), representing structured observ ations ab out how particular design decisions inuence artefact b eha viour. Do cumen ting learning ensures that kno wledge produced during dev elopment is not lost. Instead, insights accumulate across cycles and may later b e synthesised into design patterns or metho dological principles. 5 The SHAPR Op erational W orko w While the SHAPR knowledge transformation cycle structures the epistemic pro cess of artefact developmen t, researchers also require a practical w orkow that integrates exploration, implementation, and evidence management. The SHAPR framework therefore introduces an operational worko w that connects the researcher’s inter- action environmen t, dev elopment cycles, evolving artefacts, and repository-based do cumen tation. This worko w ensures that developmen t activities p erformed during research soft- w are construction are systematically transformed in to traceable research evidence. Rather than treating softw are developmen t as an informal or ad ho c activit y , SHAPR pro vides a structured mec hanism through which exploratory reasoning, artefact evolu- tion, and knowledge do cumen tation are explicitly linked. This emphasis on traceabilit y aligns with broader concerns in computing research regarding repro ducibilit y and transparency Sto dden et al. ( 2010 ). The SHAPR op erational worko w consists of four primary comp onen ts: Chat W orksp ac e → SHAPR Cycle → A rtefact → R ep ository . Figure 4 illustrates this end- to-end pro cess, making explicit how human–AI interaction in the workspace leads to artefact evolution and the accumulation of structured knowledge within a rep ository . T ogether, these comp onen ts establish a traceable and repro ducible worko w that supp orts iterative developmen t while preserving research rigour. 5.1 Chat W orkspace: Exploration and Reasoning The rst comp onen t of the w orkow is the interaction workspace, t ypically represen ted b y conv ersational environmen ts such as generativ e AI interfaces or researcher notes. In this environmen t, the researcher explores ideas, ev aluates design alternatives, and renes implemen tation strategies. In AI-assisted researc h con texts, this workspace often inv olv es in teraction with large language mo dels that supp ort reasoning, co de generation, debugging, and con- ceptual exploration. These interactions can signicantly accelerate the exploration of design alternativ es and reduce the time required to prototype new ideas. Ho wev er, SHAPR emphasises that reasoning performed in this w orkspace do es not automatically constitute researc h evidence. Conv ersations and exploratory discussions primarily function as cognitive scaolding. They supp ort the researc her’s thinking 10 Chat Workspace SHAPR Cycle (Explore → Build → Use → Evaluate → Learn) Artefact (Evolving System) Repository (Code, Cycle Records, SKUs) Fig. 4 SHAPR W orko w from W orkspace to Rep ository . This gure illustrates the op erational ow of SHAPR from the chat workspace through the SHAPR cycle to artefact developmen t and reposi- tory storage. The SHAPR cycle (Explore–Build–Use–Ev aluate–Learn) drives artefact ev olution, while outputs are systematically captured in a rep ository containing co de, cycle records, and structured knowledge units (SKUs), supp orting traceability and knowledge accum ulation. pro cess but only b ecome part of the formal research record when translated into artefact c hanges and do cumented within the rep ository . 5.2 SHAPR Cycle: Iterativ e Dev elopment The second comp onen t of the worko w is the SHAPR developmen t cycle, whic h op erationalises the knowledge transformation pro cess describ ed earlier. Eac h cycle follows the sequence: Explor e → Build → U se → Evaluate → L e arn . During a cycle, the researc her inv estigates a design idea, implemen ts changes to the artefact, ev aluates system behaviour, and documents the resulting insigh ts. In this w ay , exploratory reasoning is progressively transformed into structured developmen t activit y . The SHAPR cycle acts as the primary unit of research progress, ensuring that dev elopment pro ceeds through small, traceable iterations. Each completed cycle con tributes simultaneously to artefact evolution and knowledge generation. 5.3 Artefact: Evolving Researc h System The third comp onen t of the w orko w is the researc h artefact, typically represented b y a soft ware system under developmen t. Examples include analytical platforms, sim ulation environmen ts, exp erimen tal to ols, or prototype applications. 11 The artefact evolv es incrementally as successive SHAPR cycles introduce new fea- tures, rene system architecture, or mo dify algorithms. Observing artefact b ehaviour during testing and use provides the empirical basis for generating research insights. In SHAPR, the artefact therefore plays a dual role: • as a technical system b eing developed, and • as an exp erimen tal platform for generating research knowledge. This dual role is cen tral to SHAPR, as it links soft ware developmen t directly to kno wledge pro duction. 5.4 Rep ository: Evidence and Kno wledge Storage The nal comp onen t of the w orkow is the researc h rep ository , where dev elopment activities and insigh ts are formally recorded. The rep ository typically contains source co de, cycle do cumen tation, prompts, and knowledge units derived from SHAPR cycles. The rep ository serves as the primary lo cation for research evidence, ensuring that artefact ev olution and associated learning remain transparen t and repro ducible. A key principle of SHAPR is that developmen t activities b ecome part of the research record only when they are do cumen ted in the rep ository . By main taining structured records, the repository enables researc hers to trace ho w sp ecic design decisions inuence artefact behaviour and contribute to knowledge generation. 5.5 T raceabilit y and Evidence The integration of the four worko w comp onen ts enables SHAPR to maintain trace- abilit y across the entire dev elopmen t pro cess. Exploratory reasoning in the workspace informs developmen t cycles; cycles modify the artefact; artefact b eha viour pro duces observ ations; and these observ ations are recorded and transformed in to structured kno wledge within the rep ository . Through this w orko w, researc h softw are dev elopmen t b ecomes a traceable kno wl- edge pro duction pro cess rather than an opaque technical activit y . The rep ository therefore functions not only as a storage lo cation for artefacts but also as a historical record of the research pro cess. While Figure 4 illustrates ho w artefacts and records are generated and stored, it do es not fully explain how kno wledge is derived from these artefacts. This transformation is elab orated in Figure 5 . 6 SHAPR Kno wledge Units and Knowledge A ccum ulation While the SHAPR op erational worko w structures the developmen t process, a k ey c hallenge in research soft ware developmen t is ensuring that insights generated dur- ing artefact construction are systematically captured and transformed in to research 12 Research Artefact (System / Prototype) Observations & Results Insights Structured Knowledge Units (SKUs) Design Principles Conceptual Knowledge (Insights, SKUs, Design Principles) Empirical Evidence (Data, Observation) Fig. 5 Artefact-to-Knowledge T ransformation in SHAPR. This gure illustrates ho w SHAPR trans- forms empirical evidence into conceptual knowledge. The upp er lay er represents empirical evidence generated through researc h artefacts, while the low er lay er represents conceptual kno wledge, including insights, structured knowledge units (SKUs), and design principles. The branching structure high- lights that knowledge generation is non-linear and can o ccur at multiple levels of abstraction. kno wledge. Without delib erate documentation and reection, v aluable learning obtained during developmen t cycles may remain implicit and b e lost ov er time. T o address this c hallenge, SHAPR introduces the concept of SHAPR K now le dge Units (SKU s) and a structured mo del of kno wledge accumulation. This mec hanism ensures that insigh ts derived from artefact dev elopment are progressiv ely transformed in to reusable and generalisable design knowledge. 6.1 SHAPR Knowledge Units (SKUs) A SHAPR Knowledge Unit (SKU) is a structured representation of an insigh t derived from a developmen t cycle that explains ho w a sp ecic design decision inuences arte- fact behaviour. SKUs serv e as the fundamen tal unit of kno wledge captured within SHAPR. Eac h SKU typically contains four core elements: Comp onen t Description Context The design situation or problem addressed. Design De cision The sp ecic implementation or mo dication applied. Observation The outcome or b eha viour observed during ev aluation. Insight The in terpretation explaining why the outcome o ccurred. By explicitly linking design decisions to observ ed outcomes, SKUs capture the causal relationship b et ween developmen t actions and artefact b ehaviour. This structured representation enables insights generated during developmen t to b e systematically recorded, analysed, and reused. SKUs are t ypically extracted during the L e arn stage of the SHAPR cycle, where the researc her reects on ev aluation outcomes and identies insights that extend b ey ond 13 the immediate dev elopmen t task. Depending on the researc h con text, SKUs ma y align with or complement established forms of knowledge representation: Metho d Kno wledge Unit Exp eriment Result Case Study Finding Design Scienc e Design Principle SHAPR Structured Kno wledge Unit (SKU) 6.2 F rom Observ ations to Design Kno wledge While individual SKUs represen t discrete insights, knowledge accumulation in SHAPR emerges through the progressive abstraction of these insights across cycles. SHAPR conceptualises knowledge developmen t as a multi-lev el pro cess: Stage Kno wledge Pro duced Observation Outcome of a developmen t cycle. Insight In terpretation of the observed outcome. SKU Structured and contextualised insight. Pattern Recurring relationships across multiple SKUs. Design Principle Generalised kno wledge applicable across systems. SHAPR R enement A daptation or extension of the framework. This hierarch y reects the gradual transformation of practice-based observ ations in to increasingly generalisable knowledge. As SKUs accum ulate, recurring patterns ma y emerge, revealing consistent relationships b et ween design decisions and system b eha viour. These patterns can then b e synthesised into design principles that inform future dev elopment. 6.3 Kno wledge Accum ulation A cross Cycles Kno wledge accumulation in SHAPR occurs through the systematic documentation and syn thesis of SKUs across successive developmen t cycles. Each cycle contributes new observ ations and insights, which are recorded and stored within the rep ository . As the num b er of cycles increases, the rep ository ev olves into a structured kno wledge base that captures b oth artefact evolution and associated learning. This accum ulation enables researchers to identify relationships that may not b e visible within individual cycles. Imp ortan tly , knowledge accum ulation may extend across multiple artefacts within a research programme. Insigh ts derived from one artefact can inform subsequen t dev elopments, allowing knowledge to propagate across pro jects and contexts. Through this pro cess, artefact dev elopment b ecomes a cum ulativ e researc h activity in whic h each cycle contributes incrementally to a broader b o dy of knowledge. 14 6.4 F rom Kno wledge Accum ulation to F ramework Renemen t A t a higher lev el of abstraction, kno wledge accum ulated through SKUs and design principles can contribute to the renement of the SHAPR framework itself. Pat- terns observed across artefacts and developmen t cycles ma y rev eal methodological insigh ts ab out how AI-assisted researc h softw are developmen t can b e structured more eectiv ely . These insigh ts ma y lead to improv emen ts in w orko w design, documentation prac- tices, or metho dological guidelines. In this w ay , SHAPR evolv es through empirical evidence generated from practice, rather than solely through theoretical dev elopment. The kno wledge accumulation pro cess therefore op erates across multiple levels: • dev elopment cycles generate SKUs, • SKUs rev eal patterns, • patterns pro duce design principles, and • design principles inform framework renement. 6.5 T raceabilit y and Non-Linear Kno wledge Developmen t Kno wledge accumulation in SHAPR is not strictly linear. A single set of observ ations ma y generate multiple SKUs at dierent lev els of abstraction, which ma y later b e rened, combined, or generalised in to broader principles. This reects the dynamic nature of practice-based research, where knowledge evolv es alongside artefacts. The in tegration of evidence and traceabilit y strengthens this pro cess. By main- taining explicit links b etw een actions, artefacts, and resulting kno wledge, SHAPR enables researchers to trace ho w sp ecic insights and SKUs were derived. This sup- p orts v alidation, repro ducibilit y , and reective learning, while facilitating the transfer of kno wledge across pro jects and contexts. Ov erall, SHAPR positions knowledge accum ulation as a structured, traceable, and iterativ e process tigh tly integrated with artefact dev elopment. Through SKUs and rep ository-based organisation, practice-based evidence is systematically transformed in to reusable and progressively rened knowledge. 7 In tegrated SHAPR Op erational Mo del Building on the conceptual foundation (Section 3 ), op erational cycles (Section 4 ), worko w and traceabilit y (Section 5 ), and structured kno wledge accumulation (Section 6 ), SHAPR can b e understo od as an in tegrated op erational mo del that co ordinates h uman decision-making, AI-assisted dev elopment, artefact evolution, and kno wledge generation within a unied framework. Figure 6 presents this in tegrated view as a swimlane mo del, where each lane rep- resen ts a distinct but in terconnected asp ect of the SHAPR process: h uman-cen tred decision-making, AI-assisted developmen t, researc h artefact ev olution, and conceptual kno wledge generation. The h uman-centred decision-making lay er captures activities such as problem framing, in terpretation of results, and iterativ e renement. These activities emphasise 15 Integrated SHAPR Operational Model Human-Centred Decision Making Problem Framing & Practice Observation, Interpretation & Reflection Learning, Decision & Refinement AI-Assisted Development Artefact Generation Iteration / Simulation AI Suggestions Research Artefact Evolution Prototype V ersioning Deployment / Use Conceptual Knowledge Generation Insights SKUs Design Principles Fig. 6 Integrated SHAPR Operational Model. This gure presents a swimlane view of SHAPR, where the four lanes represent human-cen tred decision-making, AI-assisted developmen t, researc h artefact evolution, and conceptual knowledge generation. Artefacts evolve through iterative cycles, generating insights, structured knowledge units (SKUs), and design principles. Evidence and traceabil- ity op erate as a cross-cutting lay er, linking these asp ects and supporting transparency , repro ducibilit y , and systematic renement. h uman accountabilit y , critical judgement, and con textual understanding. While AI systems con tribute to developmen t and analysis, SHAPR main tains that responsibility for decision-making and ev aluation remains with the researcher. The AI-assisted developmen t lay er supp orts the execution of tasks across the SHAPR cycle, including code generation, design suggestions, debugging, and do c- umen tation. Rather than replacing human reasoning, AI acts as a collab orator that augmen ts exploration and accelerates dev elopment while op erating within the structure dened by the framework. The research artefact ev olution la yer represen ts the incremental developmen t of the system under study . Artefacts evolv e through successiv e SHAPR cycles, with eac h iteration in tro ducing new features, mo dications, or renemen ts. As artefacts are developed and used, they generate empirical observ ations that form the basis for kno wledge extraction. The conceptual knowledge generation lay er captures the transformation of empir- ical evidence into structured kno wledge, including insigh ts, Structured Knowledge Units (SKUs), patterns, and design principles. This la yer reects the pro cesses describ ed in Section 6 , where kno wledge accumulates through iterative abstraction and syn thesis. Imp ortan tly , these four lay ers are not indep enden t but op erate as an inter- connected system. Human decisions guide AI-assisted developmen t; AI-supp orted activities contribute to artefact evolution; artefact b ehaviour pro duces observ ations; 16 and these observ ations are transformed in to structured knowledge. This kno wledge, in turn, informs subsequent decisions and developmen t cycles. Evidence and traceability operate as a cross-cutting mec hanism linking all lay- ers. By maintaining explicit connections b etw een decisions, artefacts, and knowledge, SHAPR enables researchers to trace how insights are deriv ed and how artefacts ev olve ov er time. This traceability supp orts repro ducibilit y , v alidation, and reective learning. The integrated mo del also accommo dates v arying levels of AI in volv emen t. Researc hers may apply SHAPR in a predominantly human-driv en manner, using AI selectiv ely for supp ort, or adopt more automated, agen t-assisted w orkows. In both cases, the framework provides a consistent structure that ensures traceabilit y and metho dological rigour. Ov erall, the integrated SHAPR op erational mo del p ositions research softw are dev elopment as a coordinated system of h uman judgement, AI capabilit y , artefact evo- lution, and structured knowledge generation. By making these relationships explicit, SHAPR provides a foundation for conducting rigorous, transparent, and scalable researc h in AI-assisted environmen ts. 8 Human–AI Collab oration in SHAPR Practice Ha ving established the in tegrated SHAPR op erational mo del, it is imp ortan t to clarify how human researchers and AI systems interact within this framework. The rapid adv ancement of generative articial in telligence has signican tly transformed the w ay softw are systems can be designed and implemented. Large language mo d- els and related AI systems are increasingly capable of generating co de, prop osing arc hitectural solutions, analysing data, and assisting with technical problem solving. These capabilities raise imp ortan t questions regarding the role of human researchers in soft ware-driv en research environmen ts. The SHAPR framework addresses this challenge by p ositioning articial in telli- gence not as a replacemen t for the researc her but as a collaborative tool that augmen ts h uman researc h practice. This persp ectiv e aligns with established principles of human– AI collab oration that emphasise human-cen tred design and decision-making Amershi et al. ( 2019 ); Shneiderman ( 2022 ). While AI systems can assist with exploration, dev elopment, and exp erimen tation, the interpretation of results and the v alidation of researc h knowledge remain fundamentally human resp onsibilities. In this sense, SHAPR supp orts an AI-augmented research practice in which generativ e AI expands the researc her’s capacity to explore design alternativ es and accelerate developmen t cycles, while preserving the central role of human judgement in kno wledge pro duction. 8.1 AI as an Exploratory Partner A key contribution of generative AI in research softw are developmen t is its abil- it y to supp ort rapid exploration of design ideas. Through con versational interaction, researc hers can ev aluate alternativ e architectures, generate protot yp e implemen ta- tions, and inv estigate p otential solutions to technical problems. 17 Within the SHAPR worko w, this capability primarily supp orts the Explor e and Build stages of the dev elopment cycle. AI systems can assist with generating co de frag- men ts, suggesting debugging strategies, and prop osing alternative design approac hes. This accelerates exp erimen tation and enables researchers to explore a broader design space than would typically b e feasible. Ho wev er, SHAPR emphasises that exploration remains h uman-guided. The researc her determines which alternativ es are relev ant, ev aluates the plausibility of AI- generated suggestions, and decides ho w these suggestions should inuence artefact dev elopment. 8.2 Human Epistemic A uthorit y Although AI systems can assist with tec hnical implemen tation and reasoning, they do not independently generate v alidated researc h kno wledge. Within SHAPR, the human researc her retains epistemic authorit y , meaning that resp onsibility for interpreting observ ations, v alidating insights, and establishing knowledge claims remains with the researc her. This distinction is essential for maintaining the scien tic integrit y of AI-assisted researc h. AI systems ma y pro duce plausible suggestions or explanations, but deter- mining whether an observ ation constitutes meaningful kno wledge requires contextual understanding, metho dological reasoning, and critical ev aluation. SHAPR therefore distinguishes betw een artefact generation and knowledge v ali- dation. While AI con tributes to artefact construction, the extraction and v alidation of researc h knowledge are grounded in human interpretation and reection. 8.3 Expanding the Design Space Generativ e AI systems can assist with code generation, debugging, documentation, and design exploration. When integrated within structured worko ws, these sys- tems can signicantly accelerate developmen t and expand the design space explored b y researchers, as observed in recen t studies of large language mo del-assisted programming V aithilingam et al. ( 2022 ). Within the SHAPR cycle, this expanded design space enhances the exploratory capabilities of the Explor e stage and accelerates implementation in the Build stage. By reducing the eort required to protot yp e new ideas, AI systems enable researc hers to complete more developmen t cycles within a given timeframe. This increased iteration capacit y strengthens SHAPR’s knowledge accumula- tion process, as more cycles generate a larger set of observ ations and insights that con tribute to the formation of SKUs and design principles. 8.4 Structured Human–AI Researc h Practice The integration of AI in to research w orkows introduces the risk that developmen t activities ma y become opaque or dicult to reproduce, particularly when substan- tial p ortions of code or reasoning are generated through conv ersational in teraction. SHAPR addresses this challenge by structuring the interaction b et ween human reasoning, AI assistance, artefact developmen t, and rep ository-based do cumentation. 18 By requiring that developmen t activities and insights b e captured through do cu- men ted SHAPR cycles and stored within the rep ository , the framework ensures that AI-assisted dev elopmen t remains transparen t, traceable, and reproducible. This struc- tured worko w aligns AI-supp orted activities with metho dological requirements and preserv es the integrit y of the research pro cess. 8.5 SHAPR as an AI-Guidable F ramew ork An imp ortant c haracteristic of SHAPR is that its op erational structure can b e inter- preted and follo w ed b y generativ e AI systems. Because SHAPR denes explicit cycles, do cumen tation practices, and kno wledge extraction procedures, the framew ork can be em b edded within AI prompts, worko ws, or knowledge bases. In practice, researc hers can pro vide SHAPR do cumen tation—including conceptual descriptions, op erational guidelines, and templates—to AI systems in order to obtain structured assistance during research softw are developmen t. The AI system can then supp ort the execution of SHAPR cycles, assist with do cumen tation, and help identify p oten tial kno wledge units. This capabilit y p ositions SHAPR as an AI-guidable and AI-executable research framew ork. It enables AI systems to supp ort not only technical implementation but also adherence to structured research pro cesses, while preserving human-cen tred decision-making and epistemic authority . 9 Practical Implemen tation of SHAPR While the preceding sections describe the conceptual and op erational structure of SHAPR, its eectiveness depends on the abilit y of researchers to apply the frame- w ork consisten tly in real developmen t environmen ts. This section outlines practical mec hanisms that enable SHAPR to b e implemented in AI-assisted research softw are pro jects. The implemen tation of SHAPR is grounded in three key elements: • structured dev elopment cycles, • rep ository-based evidence management, and • standardised templates for do cumen ting insights. T ogether, these elemen ts transform exploratory softw are dev elopmen t in to a trace- able and reproducible researc h pro cess. Detailed templates and examples are pro vided in App endix A. 9.1 Cycle-Based Developmen t Records The fundamental op erational unit of SHAPR is the dev elopment cycle, whic h captures a single iteration of exploration, implemen tation, ev aluation, and learning. T o support systematic do cumen tation, SHAPR introduces a structured cycle record template. Eac h cycle record do cuments the reasoning, implementation, ev aluation, and learn- ing asso ciated with a developmen t iteration. A t ypical record includes the follo wing sections: 19 Cycle Section Purp ose Explor e Describ e the design problem or opp ortunity inv estigated. Build Do cumen t artefact mo dications and implementation details. U se Explain ho w the artefact was tested or applied. Evaluate Analyse outcomes, b eha viour, or p erformance. L e arn Summarise insigh ts derived from the cycle. SKU Extr action Record reusable insights generated from the cycle. Next Dir e ction Iden tify directions for subsequent cycles. These records ensure that developmen t activities remain traceable and repro- ducible, allowing researchers to understand how artefacts evolv e and how insights are generated across iterations. A minimal template for do cumen ting SHAPR cycles is provided in Appendix A (Section 11 ) and can b e adapted to dierent developmen t contexts. 9.2 Rep ository-Based Evidence Management A second practical comp onent of SHAPR is the use of a research rep ository to store artefacts, dev elopmen t records, and knowledge units. The rep ository serv es as the cen tral lo cation for preserving research evidence and maintaining traceability . T ypical rep ository con tents include: • source co de and system comp onen ts, • cycle do cumen tation, • extracted SKUs, • exp erimen tal results and ev aluation data. Organising these materials within a structured repository ensures that artefact ev olution and asso ciated knowledge remain transparent and repro ducible. A t ypical rep ository structure may include: project_repository/ docs/ cycles/ insights/ src/ system_code/ data/ experiments/ analysis/ design/ prds/ framework_notes/ This separation of implementation artefacts and research do cumen tation enables clear distinction b et w een system developmen t and knowledge generation. 20 9.3 Capturing Knowledge Through SKUs As discussed in Section 6 , the extraction of SKUs is central to transforming develop- men t exp erience into reusable knowledge. In practice, SKUs are recorded as concise, structured en tries within the rep ository . Eac h SKU captures a reusable insight deriv ed from a developmen t cycle and t ypically includes: SKU Component Description Context Situation or problem addressed. Design De cision Implemen tation approach used. Observation Behaviour observed during ev aluation. Insight Explanation of why the outcome o ccurred. This structured representation enables researchers to identify patterns across cycles and artefacts, supp orting the dev elopmen t of design principles and metho d- ological insigh ts ov er time. 9.4 In tegrating AI Assistance in to the W orko w Generativ e AI systems can supp ort SHAPR worko ws by assisting with tasks such as co de generation, debugging, do cumen tation, and design exploration. When in tegrated within the structured SHAPR process, AI can signican tly accelerate developmen t cycles and expand the range of design alternatives explored. Ho wev er, AI-generated outputs do not automatically constitute research knowl- edge. All artefact changes, observ ations, and insights must b e ev aluated by the researc her and do cumen ted through cycle records and SKU extraction. This ensures that AI-assisted developmen t remains aligned with SHAPR’s emphasis on traceabilit y and metho dological rigour. 9.5 Supp orting AI-Guided SHAPR W orko ws An additional adv antage of SHAPR is that its structured cycles, templates, and do c- umen tation practices can b e em b edded within AI prompts or kno wledge bases. By pro viding SHAPR do cumentation to AI systems, researc hers can obtain structured assistance in applying the framework during developmen t. In practice, AI systems can assist with: • generating cycle records from developmen t discussions, • suggesting p oten tial SKUs based on ev aluation outcomes, • iden tifying patterns across multiple developmen t cycles. T o supp ort this pro cess, SHAPR pro vides reusable templates (see Appendix A), including: • cycle templates, • prompt templates, 21 • reection templates, • pro duct or pro ject design templates. This capability enables SHAPR to function as b oth a human-guided metho dology and an AI-guidable framework, supp orting consisten t and scalable research w orkows across dieren t levels of AI inv olvemen t. 10 Implications for AI-Assisted Researc h Practice The emergence of generativ e AI technologies has signicantly altered the landscap e of research softw are dev elopment. T o ols capable of generating code, assisting with debugging, and prop osing design alternativ es can dramatically accelerate develop- men t. Ho wev er, these capabilities also introduce metho dological challenges related to transparency , repro ducibility , and the role of human judgement in research. The SHAPR framew ork addresses these challenges by providing a structured and traceable approac h for in tegrating AI assistance in to researc h practice while main tain- ing the rigour required for knowledge generation. The implications of this approach extend b eyond individual to ols to the design of AI-assisted research worko ws, systems, and educational practices. 10.1 Structuring AI-Assisted Researc h W orko ws A key implication of SHAPR is that AI-assisted dev elopment can b e organised into a repro ducible and traceable research w orkow. Without structured pro cesses, dev elop- men t activities supp orted by generative AI may b ecome dicult to interpret, verify , or replicate, particularly when substan tial p ortions of reasoning and implementation are generated through conv ersational interaction. By structuring dev elopment into explicit cycles and requiring do cumentation of artefacts, observ ations, and insights, SHAPR ensures that AI-assisted activities remain transparent and repro ducible. Rep ository-based records further support this pro cess by maintaining p ersisten t links betw een actions, artefacts, and resulting kno wledge. This enables researc hers to b enet from AI capabilities while preserving metho dological integrit y . 10.2 Supp orting Solo Research Soft w are Developmen t Another imp ortan t implication concerns the growing prev alence of solo research soft- w are developmen t, particularly in contexts where researchers increasingly rely on AI to ols to supp ort implementation. In such en vironments, researchers may lack the col- lab orativ e structures and resources traditionally asso ciated with large-scale softw are engineering pro jects. SHAPR provides a framework that enables individual researchers to conduct sys- tematic, traceable dev elopment while main taining researc h rigour. By organising w ork in to manageable cycles and emphasising structured do cumentation, the framew ork 22 transforms iterative developmen t into a disciplined pro cess of knowledge genera- tion. This low ers the barrier for conducting complex research softw are pro jects while main taining academic standards. 10.3 Enhancing Repro ducibilit y in Soft w are-Based Researc h Repro ducibilit y remains a signican t concern in research inv olving complex softw are systems. T raditional research outputs often fo cus on nal artefacts or exp erimen tal results without fully do cumen ting the pro cesses that pro duced them. SHAPR improv es repro ducibilit y by capturing developmen t decisions, exp erimen- tal outcomes, and learning across iterative cycles. Through structured do cumentation and rep ository-based organisation, the evolution of artefacts and the deriv ation of insigh ts are made explicit. This traceabilit y supp orts not only replication but also critical ev aluation of t he design and reasoning underlying research artefacts. 10.4 SHAPR as an AI-Executable Research F ramew ork A distinctive feature of SHAPR is that its operational structure can b e interpreted not only b y human researchers but also b y generative AI systems. The framework denes explicit cycles, structured do cumen tation practices, and knowledge extraction pro cedures, whic h can b e embedded within AI prompts, worko ws, or knowledge bases. As a result, SHAPR is b oth human-readable and AI-interpretable, enabling its use as an executable guide for AI-assisted research. This capability supp orts a new form of AI-guided research metho dology , in which AI systems assist not only with implemen tation tasks but also with adherence to struc- tured research processes. By enco ding SHAPR principles in to prompts or agent-based systems, AI can guide exploration, supp ort artefact developmen t, structure ev aluation, and facilitate kno wledge extraction, while preserving h uman-cen tred decision-making. Emerging systems such as OpenClaw illustrate ho w SHAPR can b e operationalised as an executable research worko w. In such environmen ts, SHAPR cycles and asso- ciated do cumen tation structures can b e embedded into AI-driven systems, enabling partial automation of research activities while maintaining traceability and metho d- ological discipline. This p ositions SHAPR as a bridge b et w een human-led researc h design and AI-assisted execution. 10.5 T o w ard Integrated AI-Assisted Researc h Systems SHAPR is designed to b e tool-indep enden t, allowing researchers to adopt dierent congurations of to ols dep ending on their ob jectiv es, exp ertise, and desired level of automation. AI inv olv ement ma y range from light weigh t, h uman-driven w orkows to highly automated, agen t-based environmen ts. These congurations, as sho wn in T able 1 , should b e understo o d as illustrative examples rather than prescriptive require- men ts, drawing on contemporary AI-assisted developmen t to ols and environmen ts GitHub ( 2023 ); Go ogle ( 2024a , 2 ); Op enAI ( 2023 ). These congurations demonstrate that SHAPR do es not dep end on sp ecic to ols but rather on the structure of in teraction betw een h uman decision-making, AI-assisted dev elopment, artefact evolution, and knowledge documentation. Regardless of the level 23 T able 1 Illustrative Lev els of AI Inv olvemen t in SHAPR W orkows Lev el Conguration Characteristics Low-AI Inv olv ement ChatGPT + IDE (e.g., PyCharm) + Git + Cloud Storage (e.g., OneDrive) Human-driven worko w with AI used for code assistance, debugging, and concep- tual support. F ull control and strong learn- ing orientation. Moderate-AI Inv olv ement ChatGPT / Gemini + Note- bo okLM + GitHub + Docu- mentation T emplates AI supp orts reasoning, summarisation, and do cumen tation. Researcher remains central but benets from structured AI- assisted worko ws. High-AI Inv olv ement AI-assisted dev elopment envi- ronments (e.g., GitHub Copi- lot, Cursor) + automated do c- umentation to ols AI actively assists implementation and developmen t cycles, accelerating iteration while main taining human o versigh t and v alidation. Agent-Based W orkows Autonomous or semi- autonomous agents (e.g., OpenClaw-style systems) managing developmen t cycles AI systems co ordinate parts of the SHAPR cycle, including implementation and docu- mentation, with human sup ervision guid- ing direction and v alidation. Integrated AI Research Sys- tems End-to-end en vironmen ts combining LLMs, rep ositories, IDEs, and knowledge systems (e.g., Notebo okLM + Gemini + cloud-based worko ws) Highly in tegrated systems supporting con- tinuous cycles of dev elopment, do cumen- tation, and knowledge extraction, aligned with SHAPR structure. of automation, the framework maintains human epistemic authority while enabling exible in tegration of emerging AI technologies. Bey ond individual worko ws, SHAPR suggests a system-lev el view of research practice in whic h m ultiple comp onen ts are in tegrated in to a cohesiv e en vironment. In this view, LLM w orkspaces supp ort reasoning, prompting, and documentation; in tegrated developmen t environmen ts (IDEs) support artefact construction and exe- cution; cloud storage provides p ersisten t access to artefacts and records; and version con trol systems such as Git ensure traceability of changes. T ogether, these comp onen ts form a cloud-connected and version-con trolled researc h infrastructure that supp orts contin uous iteration, structured do cumen tation, and knowledge accum ulation. Within this environmen t, SHAPR op erates as the co or- dinating framework that links h uman decisions, AI-assisted developmen t, artefact ev olution, and conceptual knowledge generation. This system persp ectiv e reinforces the importance of evidence and traceabilit y as core elements of AI-assisted research. By main taining explicit links b etw een deci- sions, artefacts, and kno wledge, SHAPR enables transparent and repro ducible research practices, addressing long-standing c hallenges in computational and softw are-based researc h Sto dden et al. ( 2010 ). 10.6 Flexible Levels of AI Inv olvemen t An important characteristic of SHAPR is its ability to supp ort v arying lev els of AI in volv emen t within the researc h pro cess. While the framework enables AI-assisted and p oten tially AI-executable worko ws, it do es not prescrib e a xed level of automation. 24 Instead, researc hers can adapt the degree of AI inv olvemen t based on their ob jectiv es, preferences, and level of exp ertise. A t one end of the spectrum, SHAPR can b e applied in a predominantly manual mo de, where researchers retain close control o ver design decisions, artefact dev el- opmen t, and knowledge extraction, using AI primarily for limited support suc h as co de suggestions or do cumen tation assistance. This mode is particularly v aluable for learning, experimentation, and main taining deep engagemen t with the research pro cess. A t the other end, SHAPR can b e implemented in more automated or agent-based en vironments, where AI systems assist with m ultiple stages of the cycle, including generation, ev aluation support, and kno wledge structuring. In suc h cases, SHAPR pro- vides the structure that ensures these activities remain traceable and metho dologically aligned. This exibilit y highligh ts that SHAPR is not tied to a specic lev el of AI capability or autonomy . Rather, it functions as a human-cen tred framework that accommo dates a contin uum of AI inv olvemen t, enabling researchers to balance control, learning, eciency , and automation according to their needs. 10.7 Implications for Higher Education and HDR Researc h While SHAPR was initially developed to supp ort solo HDR research practice Chan ( 2026 ), its operationalisation in this pap er reveals broader implications for higher education, researc h sup ervision, and assessment in AI-assisted environmen ts. T able 2 Levels of AI Inv olvemen t in SHAPR for Higher Education and HDR Contexts Lev el Studen t Role Sup ervisor Role Assessmen t F o cus Low-AI Inv olv ement Manual developmen t with selective AI sup- port Close guidance, skill developmen t emphasis Understanding, rea- soning, and process transparency Moderate-AI Inv olv ement AI-assisted coding, documentation, and exploration Guided sup ervision, emphasis on decision- making Justication of design decisions and interpre- tation High-AI Inv olv ement Extensive AI- supported developmen t workows Supervisory focus on v alidation and critical ev aluation Ev aluation of outputs, reproducibility , and insight generation Agent-Based / Automated AI executes parts of developmen t cycles Supervisor ensures methodological rigour and integrit y Assessment of research design, traceability , and knowledge contri- bution Integrated AI Research Sys- tems F ully integrated AI- supported research environmen ts Supervisor acts as mentor for researc h direction and ethics Contribution to knowl- edge, originality , and methodological clarity The concept of exible levels of AI inv olv ement has direct implications for higher education and HDR research. Dierent congurations of AI-assisted worko ws corresp ond to dierent stages of learning, supervision approaches, and assessment strategies. Low er lev els of AI in volv emen t emphasise skill dev elopment, conceptual 25 understanding, and direct engagemen t with implemen tation. In con trast, higher lev els shift the fo cus tow ard critical ev aluation, research design, and knowledge generation. In HDR contexts, SHAPR pro vides a structured approach for managing solo researc h practice in AI-assisted environmen ts. By organising developmen t in to trace- able cycles and capturing insights as structured knowledge units (SKUs), the framew ork enables systematic do cumen tation of researc h progress. This supports more transparen t sup ervision, where supervisors can ev aluate not only outcomes, suc h as the nal artefacts, but also the reasoning, exp erimen tation, and learning pro cesses underlying artefact developmen t. The framew ork also has important implications for assessment. In AI-assisted en vi- ronmen ts, ev aluating only nal artefacts b ecomes insucient, as signican t p ortions of developmen t ma y b e supp orted b y generative AI systems. SHAPR enables pro cess- orien ted assessmen t by capturing developmen t cycles, decision-making, and knowledge extraction. This shifts the fo cus from artefact correctness to traceability , justication, and epistemic contribution. A critical implication of AI-assisted researc h environmen ts concerns the refram- ing of academic in tegrity . T raditional approaches often fo cus on whether studen ts ha ve used external assistance, including generativ e AI systems, in the pro duction of their work. In AI-augmented contexts, how ever, suc h binary distinctions b ecome increasingly dicult to dene and enforce. SHAPR supp orts a shift from tool-based ev aluation to w ard process-based accoun t- abilit y . Rather than asking whether AI w as used, the emphasis mov es to whether the researcher or student can explain, justify , and trace the developmen t pro cess and resulting artefacts. This includes demonstrating how design decisions were made, ho w AI-assisted outputs were ev aluated, and ho w insights were derived and v alidated. By structuring dev elopment through documented cycles and capturing insigh ts as structured kno wledge units (SKUs), SHAPR enables transparen t tracing of the researc h pro cess. Academic in tegrity is th us grounded in epistemic resp onsibilit y , where the legitimacy of work is determined by the researcher’s ability to account for their reasoning, decisions, and knowledge contributions, regardless of the level of AI assistance in volv ed. More broadly , SHAPR supp orts a reconguration of the roles of students and sup ervisors. As AI systems increasingly assist with implementation and exploration, studen ts are required to take greater resp onsibilit y for interpretation, v alidation, and kno wledge construction. Sup ervisors, in turn, shift from direct instruction tow ard guiding research design, ensuring metho dological rigour, and supp orting critical reection. These implications suggest that SHAPR can serve not only as a research frame- w ork but also as a p edagogical structure for AI-assisted learning en vironmen ts. F uture researc h can explore how SHAPR-informed worko ws supp ort student learning, sup ervision practices, and assessment design across dierent levels of education. 10.8 T o w ard Practice-Centred Researc h Metho dologies More broadly , SHAPR supports a reconguration of the roles of studen ts and supervi- sors. As AI systems increasingly assist with implementation and exploration, studen ts 26 tak e greater resp onsibility for interpretation, v alidation, and knowledge construction. Sup ervisors, in turn, shift from direct instruction tow ard guiding research design, ensuring metho dological rigour, and supp orting critical reection. These implications suggest that SHAPR can function not only as a researc h frame- w ork but also as a p edagogical structure for AI-assisted learning en vironmen ts. F uture researc h can explore how SHAPR-informed worko ws supp ort student learning, sup ervision practices, and assessment design across dierent levels of education. 11 Conclusion The gro wing capabilities of generative articial intelligence are rapidly transform- ing the practice of researc h softw are developmen t. While these technologies enable faster exp erimen tation and implementation, they also introduce challenges related to metho dological structure, traceability , and the preserv ation of h uman judgement in kno wledge generation. This pap er presented SHAPR as a structured framework for op erationalising A ction Design Research in the con text of solo, AI-assisted research softw are develop- men t. The framew ork conceptualises research soft ware developmen t as a kno wledge- generating pro cess b y explicitly linking research practice, artefact evolution, and kno wledge extraction. The pap er mak es four key contributions. First, it op erationalises ADR for AI- assisted en vironments through an iterative developmen t reasoning cycle consisting of Explore, Build, Use, Ev aluate, and Learn. Second, it introduces a traceable op er- ational w orkow that connects human–AI interaction, developmen t cycles, artefact ev olution, and repository-based do cumen tation. Third, it prop oses SHAPR Knowl- edge Units (SKUs) as a mechanism for transforming developmen t insights into reusable and progressiv ely generalisable kno wledge. F ourth, it p ositions SHAPR as an AI-executable and system-orien ted framew ork that integrates human-cen tred decision- making, AI-assisted developmen t, and supp orting infrastructure to enable scalable and repro ducible research worko ws. A key feature of SHAPR is its emphasis on h uman epistemic authorit y within AI-assisted environmen ts. While generative AI systems can signicantly expand the design space and accelerate developmen t, the interpretation, v alidation, and general- isation of knowledge remain the resp onsibility of the h uman researc her. In this wa y , SHAPR supports AI-augmented researc h practice while preserving methodological rigour and accountabilit y . Imp ortan tly , SHAPR is designed to b e to ol-independent in an environmen t where AI technologies are evolving rapidly . New to ols, platforms, and agent-based systems con tinue to emerge, oering dieren t capabilities for reasoning, developmen t, and kno wledge management. SHAPR accommo dates this dynamic landscap e by enabling researc hers to adopt dieren t congurations of to ols and v arying levels of AI inv olve- men t, ranging from h uman-driven w orkows to highly automated, AI-executable systems. This exibilit y allows the framework to supp ort diverse researc h and edu- cational ob jectives while remaining robust to changes in the underlying tec hnology ecosystem. 27 By structuring dev elopment cycles, enabling systematic knowledge capture, and supp orting traceability across the research pro cess, SHAPR provides a practical foundation for transforming iterative softw are developmen t into a rigorous and repro- ducible research activity . The framework is particularly relev an t for researchers who increasingly rely on AI to ols to supp ort complex, softw are-driv en inv estigations. More broadly , SHAPR contributes to the developmen t of AI-assisted research metho dologies that are structured, transparen t, and scalable. By accommo dating v ary- ing lev els of AI in v olvemen t and enabling in tegration with dev elopment en vironments, rep ositories, and AI systems, SHAPR pro vides a exible foundation for future researc h practice. F uture w ork will extend SHAPR across multiple research softw are artefacts, domains, and educational contexts to ev aluate its eectiveness and generalisabilit y . In particular, further researc h can explore its application in HDR supervision, AI- assisted learning environmen ts, and assessmen t design, as well as its integration with emerging AI tools and agent-based systems. As AI technologies con tinue to ev olv e, SHAPR provides a stable conceptual and op erational foundation for in vestigating ho w h uman-centred and AI-assisted researc h practices can co-ev olv e, enabling a broad and ev olving research agenda. App endix A. Applying SHAPR in LLM-Supp orted Researc h W orko ws This app endix provides a practical guide for applying SHAPR in LLM-supported researc h en vironments. It demonstrates ho w researchers can use SHAPR as an AI- readable and executable framew ork b y em b edding its structures, cycles, and templates in to interactions with large language mo dels (LLMs). A.1 SHAPR as an AI-Readable F ramew ork SHAPR is designed to b e b oth human-readable and AI-interpretable. Its structured cycles (Explore–Build–Use–Ev aluate–Learn), explicit artefacts, and do cumentation practices enable it to b e directly incorp orated in to LLM worko ws. By providing SHAPR do cumen tation to an AI system, researchers can guide the AI to follo w consisten t research pro cesses aligned with SHAPR principles. This enables a worko w in which researchers: • Pro vide SHAPR framework do cumen ts to an AI system, • Instruct the AI to follow SHAPR cycles, • Dev elop artefacts collab orativ ely with AI assistance, • Do cumen t cycles and extract structured knowledge, • A ccumulate reusable knowledge for future research. A.2 Typical SHAPR–LLM W orko w A t ypical worko w using SHAPR in an LLM workspace is as follows: 28 1. Upload SHAPR framew ork do cuments (e.g., Papers 1–3, templates, and notes). 2. Initialise the AI with instructions to follow SHAPR principles. 3. Begin the Explore phase by dening the problem and research ob jectiv es. 4. Progress through Build, Use, Ev aluate, and Learn stages with AI assistance. 5. Do cument each cycle, including artefacts, observ ations, and reections. 6. Extract Structured Know ledge Units (SKUs) and up date the rep ository . A.3 Example Initialisation Prompt The follo wing prompt can b e used to initialise an LLM for SHAPR-based research: You are assisting in a research project using the SHAPR framework (Solo Human-Centred and AI-Assisted Practice). Follow these principles: - Maintain human-centred decision-making - Support AI-assisted development - Structure work using SHAPR cycles: Explore → Build → Use → Evaluate → Learn - Help document each cycle clearly - Assist in extracting Structured Knowledge Units (SKUs) - Ensure traceability between decisions, artefacts, and knowledge Your role is to support the researcher while preserving their decision authority and ensuring alignment with SHAPR. A.4 Cycle Execution Prompt T emplate F or each SHAPR cycle, t he following prompt structure can b e used: We are in the [STAGE] phase of the SHAPR cycle. Context: [Describe current problem, artefact, or goal] Tasks: - Guide the next steps for this stage - Suggest artefact development or refinement - Identify what should be observed or evaluated - Help structure outputs and documentation Output: - Clear next steps - Suggested artefact changes - Key observations to capture - Potential SKUs to extract 29 A.5 SKU Extraction Prompt T emplate Based on the following observation s and results: [Insert observations] Please: - Identify key insights - Convert them into Structured Knowledge Units (SKUs) - Indicate possible reuse contexts - Suggest whether any design principles can be derived A.6 Rep ository Structure (Example) A SHAPR rep ository may b e structured as follows: project_root/ ��� artefacts/ � ��� prototype_v1/ � ��� prototype_v2/ � ��� cycles/ � ��� cycle_01.md � ��� cycle_02.md � ��� knowledge/ � ��� skus.json � ��� design_principles.md � ��� prompts/ � ��� initialisation.txt � ��� cycle_template.txt � ��� README.md A.7 T o w ard AI-Executable Researc h Systems The structured nature of SHAPR enables its integration into AI-driv en environ- men ts and research to ols. Systems such as OpenClaw illustrate ho w SHAPR cycles, prompts, and do cumen tation practices can be em b edded in to AI-assisted w orkows, enabling partial automation of research activities while maintaining traceability and metho dological rigour. This suggests a future in whic h SHAPR functions not only as a conceptual frame- w ork but also as a practical foundation for AI-executable researc h systems, supporting scalable, repro ducible, and knowledge-generating research practices. 30 A.8 Integrated SHAPR Dev elopmen t En vironmen t SHAPR worko ws can be further simplied and strengthened b y integrating LLM w orkspaces with dev elopment environmen ts, cloud storage, and version con trol sys- tems. In suc h a setup, the LLM (e.g., ChatGPT) is used for guidance, prompting, and do cumen tation, while an integrated developmen t environmen t (IDE) such as Visual Studio Co de or PyCharm is used for artefact developmen t and execution. Cloud storage (e.g., OneDriv e, Go ogle Drive, or cloud-based rep ositories) enables p ersisten t storage of artefacts, cycle records, and knowledge outputs, ensuring accessibilit y and contin uity across sessions. V ersion control systems such as Git pro- vide systematic trac king of c hanges to code, do cumen ts, and knowledge artefacts, supp orting traceability and repro ducibilit y . In this integrated environmen t, SHAPR cycles can b e executed seamlessly: • The LLM guides the researcher through SHAPR stages and do cumen tation. • The IDE supp orts artefact developmen t, testing, and renement. • Cloud storage maintains structured records of cycles, artefacts, and knowledge. • Git captures version histories, enabling traceability of decisions and changes. This in tegration reduces friction in applying SHAPR and reinforces its emphasis on structured, traceable, and iterative research practice. It also supp orts the dev elopment of scalable and collaborative research worko ws, where artefacts and kno wledge can b e shared, reused, and extended across pro jects and teams. In this sense, SHAPR can b e viewed as a cloud-connected and version-con trolled researc h worko w that integrates h uman judgemen t, AI assistance, and persistent kno wledge accumulation. References Amershi, S., W eld, D., V orvorean u, M., et al. (2019). Guidelines for human–AI inter- action. In Pr o c e e dings of the 2019 CHI Confer enc e on Human F actors in Computing Systems (pp. 1–13). https://doi.org/10.1145/3290605.3300233 Chan, K. (2026). SHAPR: A Solo Human-Centred and AI-Assisted Practice F rame- w ork for Research Softw are Developmen t. arXiv pr eprint , Cursor. (2024). Cursor AI. https://cursor.sh GitHub. (2023). GitHub Copilot. https://gith ub.com/features/copilot Go ogle. (2024a). Gemini. https://deepmind.google/technologies/gemini Go ogle. (2024b). Noteb o okLM. https://notebo oklm.google Hevner, A. R., Marc h, S. T., P ark, J., & Ram, S. (2004). Design science in information systems researc h. MIS Quarterly , 28(1), 75–105. 31 Op enAI. (2023). ChatGPT. https://c hat.op enai.com P eers, K., T uunanen, T., Rothen b erger, M. A., & Chatterjee, S. (2007). A design science research metho dology for information systems research. Journal of management information systems , 24(3), 45–77. Sein, M. K., Henfridsson, O., Purao, S., Rossi, M., & Lindgren, R. (2011). Action design researc h. MIS Quarterly , 35(1), 37–56. Shneiderman, B. (2022). Human-c enter e d AI . Oxford Universit y Press. h ttps://doi.org/10.1093/oso/9780192845290.001.0001 Sto dden, V. C., Donoho, D., ... & F omel, S. (2010). Reproducible researc h: A ddressing the need for data and code sharing in computational science. Computing in Scienc e & Engine ering , September/Octob er 2010, 8–12. V aithilingam, P ., Zhang, T., & Glassman, E. L. (2022, April 29–Ma y 5). Exp ectation vs. exp erience: Ev aluating the usability of co de generation to ols pow ered by large language mo dels. Pr o c e e dings of the CHI Confer- enc e on Human F actors in Computing Systems, New Orle ans, LA, USA. h ttps://doi.org/10.1145/3491101.3519665 32