Concept·Workflow Infrastructure·Agency·Emerging·CON-024

Design Skills, Contracts and Evals Development

Value hypothesis

Builds organisational AI capability through reusable files that reduce variance in AI output quality across team members and compress the time practitioners spend learning to prompt effectively. Inc.

Learning · Efficiency

The design team builds, uses and maintains organisational AI infrastructure to help codify execution and review quality of repeatable activities. These instruction files, system prompts, work contracts, and design evals help standardise how team members use AI tools. "Prompt ops" encourages shared practice and facilitates onboarding. Quality management goes from reactive to preventive: teams integrate compliance, heuristics, and functional baselines into the generation context before work begins, rather than correcting AI output afterwards.

Risks in application

Pseudoproductivity

Instruction files may appear to standardise practice while encoding shallow or misaligned guidance; teams may believe they have quality governance despite templates being too generic to materially improve output quality. "Eval theater" aligns quality on fulfilling automated tests rather than demonstrable value.

Deskilling

Instruction files created without documented reasoning become organisational dependencies no one can explain or adapt when AI tools change; the rationale for specific constraints is lost before anyone thinks to record it.

Expertise that differentiates

Design System Logic

Structuring instruction files with the same architectural discipline as design system components: composable, reusable, maintainable, and consistent across different team members' workflows.

Business Framing

Defining which organisational constraints, quality standards, and brand requirements need to be encoded, and prioritising which workflows benefit most from standardisation.

AI Fluency that assures

Performance Discernment

Humans must the supply organisational insight necessary to define constraints, standards, and requirements, and to prioritize which workflows benefit most from standardisation.

Outputs must be monitored to verify that desired quality improvements are delivered, not just that the templates are well-structured.

Related

Possible Indicators

Knowledge asset creation

Growth of template libraries; rate of revision and maintenance.

Practice diffusion

Rate at which AI-enabled practices spread across the team; reuse of prompt and skill templates.

Sources

Wen (2025). Don't Trust the Design Process. Anthropic.

Newman (n.d.). The Infrastructure No One Talks About: How to Build an AI-Powered Design Workflow That Actually Holds. Medium.

Chawla (n.d.). How Top Companies Are Using AI in Their Design Workflows. UX Collective.

Baldwin (2026). Prototyping for the unknown. Into Design Systems AI Conference 2026.

Rousseau (2026). WhatsApp Web: Reclaiming UI Excellence through Vibe Coding. Into Design Systems AI Conference 2026.

Perez-Cruz (2026). Product Primitives: The New Material of Design System. Into Design Systems AI Conference 2026.

Yan and Giordimaina (2026). The Path to an AI-Enabled Design System. Into Design Systems AI Conference 2026.

Kavcic (2026). Agentic Design Systems. Into Design Systems AI Conference 2026.

Stockman (2026). I'm not an engineer but I ship code: How designers can ship production code and work like an engineer. Into Design Systems AI Conference 2026.

Gardner (2026). Context > Probability: Design systems as AI infrastructure. Into Design Systems AI Conference 2026.

Six (2026). Building real design systems with agents. Into Design Systems AI Conference 2026.

Morales Achiardi (2026). Encoding governance on agentic design systems. Into Design Systems AI Conference 2026.

Wolosin (2026). Machine-Readable Design Systems for MCP and LLMs. Into Design Systems AI Conference 2026.

Fehre (2026). From falling for markdown to solving real problems with scripts. Into Design Systems AI Conference 2026.

Fung (2026). Ship It! Vibe Coding Your First Figma Plugin. Into Design Systems AI Conference 2026.

Sandu et al. (2026). Designers Who Ship: Building a Real Plugin in 48 Hours with AI. Into Design Systems AI Conference 2026.

Frost et al. (2026). AI Without the Chaos: Context-Based Design Systems to the Rescue. Into Design Systems AI Conference 2026.

Yildirim et al. (2023). Creating Design Resources to Scaffold the Ideation of AI Concepts. ACM DIS '23.

Author unknown (2026). State of the Designer 2026. Figma.