Research: Prompt as Design Material

Date: 2026-03-10 Search queries used:

• “prompt as design material iterative design artifacts practice”
• “prompt engineering as design craft HCI research”
• “prompts as artifacts iterative workflow designers AI tools practice CHI”
• “digital design material concept Löwgren Stolterman reflective design”
• “Schon reflective practitioner conversation with materials design AI prompting”
• “prompt engineering tacit knowledge skill creativity design practice CHI 2024 2025”
• “prompt craft skill designers strategists writing practice iteration AI collaboration 2025”

Executive Summary

“Prompt as design material” is a framing that treats text prompts not as one-time commands but as iterated, owned, versioned artifacts with craft properties — closer to a sketch or wireframe than to a search query. The framing draws from Donald Schön’s “reflective conversation with materials” to argue that effective prompting requires the same non-linear, reflective back-and-forth that characterises all serious design practice. Academic HCI research (CHI, TEI 2025) is converging on the term “prompt craft” to replace “prompt engineering,” signalling a disciplinary shift from technical optimisation toward tacit skill, materiality, and judgement. For designers and strategists, the shift changes what the deliverable is: a well-crafted prompt library or instruction set is itself an intellectual and professional output, not merely a means to an end.

Key Sources

From Prompt Engineering to Prompt Craft

• URL: https://arxiv.org/abs/2411.13422
• Type: Pictorial research paper (ACM TEI 2025)
• Authors: Joseph Lindley & Roger Whitham (Lancaster University / Design Research Works)
• Key points:
  • Proposes “prompt craft” as a productive reframing of “prompt engineering”
  • Explores materiality and uncertainty as qualities of diffusion-based generative AI
  • Proposes craft-like navigation of the latent space within generative AI models
  • Research programme is practice-based: Shadowplay and Cardshark projects
  • Frames prompting as involving “tangible and embodied interactions”
• Tenet alignment: Strong alignment with Symbiotic Intelligence — craft frames the human as skilled navigator, not button-pusher. Aligns with Human Intent First — craft implies deliberate authorship of intent.
• Quote: “The pictorial proposes the notion of prompt craft as a productive reframing of prompt engineering.”

Design Generative AI for Practitioners: Exploring Interaction Approaches Aligned with Creative Practice

• URL: https://arxiv.org/html/2603.03074v1
• Type: Workshop paper, CHI 2026
• Authors: Xiaohan Peng, Wendy E. Mackay, Janin Koch (INRIA / Université Paris-Saclay)
• Key points:
  • Design is a “reflective conversation with materials” (Schön, 2017) — the paper explicitly applies this to GenAI
  • Identifies three interaction approaches that distribute control across intent, input, and process: DesignPrompt (multimodal scaffolds), FusAIn (manipulable visual prompts), DesignTrace (trackable iterative workflows)
  • Names “cumulative generation drift” as a core risk: repeated AI iterations frequently diverge from original design intent
  • Calls for fluid shifting of AI initiative: proactive (productive friction, divergence) vs. reactive (execute clear intent)
  • Alignment is a “dynamic negotiation,” not a fixed state
  • Professional designers resist conversation-driven interfaces — “we are visual thinkers, not verbal thinkers”
• Tenet alignment: Strong alignment with Human Intent First (intent decomposition, tracking intent over time), Context as Infrastructure (DesignTrace captures and persists design rationale), Symbiotic Intelligence (AI initiative shifts to match human need)
• Quote: “Alignment between designers and GenAI is not a static objective achieved through more controllable prompts and outputs, but a multi-stage, multi-mode coordination that unfolds dynamically across the creative process.”

The Craft of the Instruction

• URL: https://uxdesign.cc/the-craft-of-the-instruction-f0da34c445cb
• Type: Practice essay / UX Collective (Feb 2026)
• Author: Amber Bouabdallah (Product Designer)
• Key points:
  • Personal essay describing how iterating writing instructions for Claude forced articulation of tacit craft knowledge (Polanyi’s “we can know more than we can tell”)
  • Drew on Christopher Alexander’s Pattern Language as analogy: instructions = patterns, each specific enough to be actionable but open enough to generate variation
  • The instruction set is the IP; the craft of writing it is competitive advantage
  • Enterprise instruction-writing surfaces hidden organisational knowledge — tacit processes made explicit through prompting
  • Key move: from “content generation” to instructions as a reusable, owned, refined object
  • AI holds scaffolding; human does the building
  • “The instruction is not for Claude. It’s for me — a mirror held up to accumulated craft.”
• Tenet alignment: Strong alignment with Human Intent First (articulation of intent is the primary act). Aligns with Context as Infrastructure (instruction sets as persistent context infrastructure). Tension with Symbiotic Intelligence — the essay is collaborative but the AI is explicitly subordinate.
• Quote: “Writing instructions isn’t just configuration. It is externalization — converting what has lived in practice into something that can be shared, reviewed, and refined.”

Prompt-Craft Cards: A Toolkit for Developing Design Judgment Through Reverse Prompt Engineering

• URL: https://papers.academic-conferences.org/index.php/icair/article/download/4362/3944/15742
• Type: Conference paper (ICAIR 2025)
• Author: Kardelen Aysel (İzmir University of Economics)
• Key points:
  • Frames prompt practice as a vehicle for developing design judgement, not just AI literacy
  • “Reverse Prompt Engineering” (RPE): given an AI output, reconstruct the intent that generated it — practises articulation in reverse
  • Three scaffolded card decks aligned with Dreyfus model: novice → competent designer
  • Tacit-to-explicit knowledge transfer is the core pedagogical mechanism
  • “Ambiguity of relationship” (Gaver et al.) between intent and output is a feature, not a bug — it forces reflection
  • “The pedagogical goal of PCC is not to teach the craft of the prompt, but to use the prompt as a medium to teach the craft of design judgment.”
• Tenet alignment: Strongly aligns with Symbiotic Intelligence (prompt practice expanding human understanding). Pluralism — RPE surfaces the statistical/averaged nature of AI outputs, confronting learners with “the AI’s collective representation” vs. their individual worldview.
• Quote: “The central question we ask is not ‘how to write a better prompt?’ but rather ‘why did I make this specific design choice, and how can I articulate it?’”

Reconceptualizing Prompt Engineering as Reflective Professional Practice

• URL: https://www.mdpi.com/2079-9292/15/5/930
• Type: Academic paper (MDPI Electronics, 2025)
• Key points:
  • Proposes moving prompt engineering from technical competency to reflective professional practice
  • Systematic conceptual analysis drawing on Schön, reflective practice frameworks
  • Connects prompting to professional identity formation
• Tenet alignment: Aligns with Symbiotic Intelligence (expanding capability through reflection, not automation)

Major Positions

Position 1: Prompt as Engineering (Dominant Industry View)

• Proponents: Most LLM platform documentation, enterprise tooling, “prompt engineering” courses
• Core claim: Prompts are technical inputs that should be optimised for reliable outputs. Success = consistent, predictable results.
• Key arguments: Structured formats (chain-of-thought, few-shot) improve performance; version control treats prompts like code; teams need prompt management infrastructure
• Relation to site tenets: Weak alignment. Treats prompts as configuration rather than as expressions of intent. Human becomes a system tuner, not an author. Risks automation bias.

Position 2: Prompt as Craft (Emerging Academic / Practice Position)

• Proponents: Lindley & Whitham; Bouabdallah; Aysel; Paz Perez (UX Collective)
• Core claim: Prompts are craft objects. Their quality depends on tacit knowledge, iterative refinement, and a “reflective conversation with materials.” Effective prompting is a skill closer to writing or designing than to programming.
• Key arguments:
  • Prompts encode tacit knowledge made explicit (Polanyi)
  • Good prompts resist the “competent but hollow” AI default
  • Prompt iteration is a form of self-knowledge — it surfaces the prompter’s own judgement
  • Craft requires domain expertise, not just technical syntax
• Relation to site tenets: Strong alignment with Human Intent First (craft = deliberate authorship of intent). Strong alignment with Symbiotic Intelligence (skilled human directing capable AI, each contributing what the other lacks).

Position 3: Prompt as Inadequate Interface for Design (Critical HCI View)

• Proponents: Park et al. (NordiCHI 2024); Peng, Mackay & Koch (CHI 2026)
• Core claim: Text prompts are fundamentally misaligned with visual, design-oriented thinking. Forcing designers into verbal articulation creates a cognitive mismatch and loses tacit visual knowledge.
• Key arguments:
  • Professional designers self-describe as “visual thinkers, not verbal thinkers”
  • Single-shot prompt paradigm presupposes near-complete intent — incompatible with exploratory design
  • Cumulative drift erases design direction over iterations
  • Better: multimodal scaffolds, direct visual manipulation, trackable workflows
• Relation to site tenets: Aligns with Pluralism — surfaces that “prompt as medium” is not universal; different practitioners need different interaction models. Partially conflicts with treating text prompts as the primary design material — the position is that they are insufficient alone.

Key Debates

Debate 1: Craft vs. Engineering as the Right Frame

• Sides: “Prompt engineering” camp (optimise, systematise, automate) vs. “Prompt craft” camp (iterate, reflect, own)
• Core disagreement: Whether prompting is more like configuring a system or more like practising a skilled trade
• Current state: Ongoing. The engineering frame dominates product development; the craft frame is gaining ground in design education and practice-based HCI research. The Lindley/Whitham pictorial explicitly marks this as a disciplinary intervention.

Debate 2: Tacit Knowledge — Surface or Protect?

• Sides: Bouabdallah / Aysel argue that forcing tacit knowledge into explicit prompt form is valuable — it surfaces and develops expertise. Others (implicit in the Park et al. critique) argue the forced verbalization of tacit visual/design knowledge distorts it.
• Core disagreement: Is articulating tacit knowledge as prompt instructions a feature (reflection, self-knowledge) or a lossy compression (strips nuance, favours verbal thinkers)?
• Current state: Open. Likely both — depends on whether articulation is used to improve the human’s own understanding or merely to satisfy the AI.

Debate 3: Prompt as Atomic vs. Prompt as Process

• Sides: Atomic view: a prompt is a message you send; iterative view: a prompt library/instruction set is a persistent artifact you maintain
• Core disagreement: What is the unit of prompt design work?
• Current state: DesignTrace research (Peng et al.) shows that the process dimension — the full history of generative iterations, branches, and rationale — matters as much as any individual prompt. The atomic view is insufficient for serious creative or strategic work.

Historical Timeline

Potential Article Angles

Based on this research, an article could:

1. “What makes a prompt a design artifact?” — Concept piece distinguishing the engineering view from the craft view; defines the qualities of a prompt-as-artifact (iterative, owned, versioned, tacit knowledge made explicit, carrying intent). Aligns with Human Intent First and Context as Infrastructure.

2. “The instruction set as professional output” — Applied topic piece for designers and strategists. Argues that a well-curated prompt library / instruction set is a deliverable in its own right — a form of IP encoding professional judgement. Connects to Symbiotic Intelligence (skill of the human is what differentiates output quality).

3. “When prompts lie: the drift problem in iterative AI collaboration” — Concept or argument piece on cumulative generation drift and intent erosion. What architectural and behavioural practices prevent drift? Aligns with Human Intent First and Context as Infrastructure.

4. “Verbal thinkers, visual thinkers, and the prompt problem” — Argument: text-only prompting is an epistemological constraint that favours certain kinds of thinkers and certain kinds of knowledge. The move toward multimodal prompting is a pluralism issue. Aligns with Pluralism of Perspectives.

When writing, follow obsidian/project/writing-style.md:

• Front-load the key claim: what prompt-as-design-material means and why it changes practice
• Named-anchor forward references for “tacit knowledge,” “cumulative drift,” “prompt craft”
• Connect to the Symbiotic Intelligence tenet via the craft frame (human skill + AI capability = better than either alone)
• Avoid the “prompt engineering is not X, it is Y” construct

Gaps in Research

• No longitudinal studies on how design teams’ prompt libraries evolve and whether they decay (intent drift in the library itself)
• Almost no research on prompt-as-material in text-heavy work (strategists, researchers, writers) as distinct from visual designers; most empirical work uses image generation
• The “reflective conversation with materials” framing is borrowed from Schön but not rigorously applied — how does Schön’s back-talk mechanism map to AI’s stochastic responses?
• Collective / team prompting as design material is underexplored: prompt libraries are often individual; what makes them work as shared team infrastructure?
• Relationship between prompt craft and automation bias (Tenet 3 gap): skilled prompt craft requires the human to maintain authorial judgment — but does iterative AI collaboration erode this over time even when the practitioner is skilled?
• Non-Western / non-English-language practitioners: all identified sources are from North American or European institutions; how does the verbal/textual nature of prompting disadvantage practitioners working in other languages or epistemological traditions?

Citations

1. Lindley, J., & Whitham, R. (2024). From Prompt Engineering to Prompt Craft. arXiv:2411.13422. [TEI 2025]. https://arxiv.org/abs/2411.13422

2. Peng, X., Mackay, W. E., & Koch, J. (2026). Design Generative AI for Practitioners: Exploring Interaction Approaches Aligned with Creative Practice. Workshop paper, CHI 2026. https://arxiv.org/html/2603.03074v1

3. Bouabdallah, A. (2026, February 19). The Craft of the Instruction. UX Collective. https://uxdesign.cc/the-craft-of-the-instruction-f0da34c445cb

4. Aysel, K. (2025). Prompt-Craft Cards: A Toolkit for Developing Design Judgment Through Reverse Prompt Engineering. ICAIR 2025. https://papers.academic-conferences.org/index.php/icair/article/download/4362/3944/15742

5. Schön, D. A. (1983/2017). The Reflective Practitioner: How Professionals Think in Action. Routledge.

6. Alexander, C., Ishikawa, S., & Silverstein, M. (1977). A Pattern Language. Oxford University Press.

7. Polanyi, M. (1966). The Tacit Dimension. Doubleday.

8. Park, H., Eirich, J., Luckow, A., & Sedlmair, M. (2024). “We are visual thinkers, not verbal thinkers!”: a thematic analysis of how professional designers use generative AI image generation tools. NordiCHI ‘24. https://doi.org/10.1145/3679318.3685370

9. Peng, X., Koch, J., & Mackay, W. E. (2024). DesignPrompt: using multimodal interaction for design exploration with generative AI. ACM DIS ‘24. https://doi.org/10.1145/3643834.3661588