Draft:Agentic Design


Agentic Design (also known as Agentic Design Systems, ADS, or Agentic Design Methodology) is a design system methodology developed by Diego Packer Martins for building design systems in which artificial intelligence agents are first-class consumers alongside human designers and engineers.[1] The methodology extends Atomic Design with machine-queryable semantics, executable constraints, and a protocol-first architecture, replacing the component library as the central artifact with a versioned, machine-readable schema called the manifest.

The terms Agentic Design and Agentic Design Systems were first coined by Diego Packer Martins, whose work on AI-native design infrastructure was developed through large-scale design system practice across multiple enterprise product organizations. The methodology is documented in the book ADS: Agentic Design Systems — From Blanding to Brand: Building Design Systems for the Age of Agents, published at agenticdesign.diegomartins.com.

ADS addresses a structural limitation of earlier design system methodologies: that design systems were built to be read by humans but cannot be effectively reasoned about by AI agents. Without a semantic protocol layer, AI-generated interfaces revert to statistically common patterns drawn from training data, producing aesthetic convergence across products — a phenomenon the methodology terms blanding.

History and origin

The Agentic Design methodology emerged from Diego Packer Martins's practice building large-scale, cross-product design systems at enterprise technology organizations. The methodology developed in response to a gap observed as AI coding agents and design tools became active participants in product development workflows: existing design system frameworks, including the widely adopted Atomic Design methodology introduced in 2013, had been conceived for a world where human designers and engineers were the sole consumers of a design system.

As AI agents became capable of generating UI from prompts, parsing component libraries, and writing implementation code, a structural limitation became apparent. AI agents could read a design system — locating components, extracting token values, copying documented patterns — but could not reason about it. They could not determine whether a given component was valid in a particular context, whether a color carried semantic appropriateness for a given use case, or whether a generated output satisfied the system's intent-level constraints. The result was AI-generated output that passed visual inspection while violating design intent.

Martins formalized the methodology and coined the terms Agentic Design and Agentic Design Systems to describe a design system architecture built from the ground up for machine queryability. The approach positions the manifest — a versioned, machine-readable schema encoding the full semantic knowledge of the design system — as the primary artifact, with the component library recast as a downstream expression of the manifest.

Methodology

ADS defines design systems as protocols rather than references. Where a traditional design system is a library that human practitioners consult, an Agentic Design system is a queryable protocol that agents and humans alike can interrogate for deterministic, semantically correct answers. The methodology is organized around five levels of abstraction, seven non-negotiable governance rules, and one core artifact: the manifest.

The methodology is described as:

  • AI-native — architected from the ground up for machine reasoning
  • AI-powered — enabling AI agents to generate on-brand, on-system output by querying the manifest directly
  • AI-ready — providing a structured path for existing design systems to adopt the protocol-first architecture incrementally

Distinction from Atomic Design

ADS explicitly positions itself as an extension of Atomic Design rather than a replacement. Both methodologies use a five-level hierarchy. The key distinction is that Atomic Design's levels — atoms, molecules, organisms, templates, and pages — define compositional structure. ADS's corresponding levels — Signals, Intents, Contracts, Protocols, and Experiences — define semantic meaning and constraint, encoding what each element means, where it is valid, and what rules govern its use.

A further distinction is the position of the component library. In Atomic Design, the component library is the primary product. In ADS, the component library is a downstream artifact generated from the manifest; the manifest is the product.

The five levels

ADS Level Atomic Design Equivalent Definition
Signal Atom The smallest meaningful design decision: a value with a semantic role, a constraint, and valid and invalid contexts. A Signal without a constraint is a variable, not a Signal.
Intent Molecule A named design goal — what a pattern is for, not what it looks like. The same Intent can resolve to different components on different surfaces or themes.
Contract Organism A machine-readable specification defining what a component can do, cannot do, what it requires, and what it promises. Contract violations fail the build.
Protocol Template A machine-readable configuration defining how Intents resolve for a specific surface, platform, and theme. Protocols are executable resolution maps, not layout skeletons.
Experience Page A validated interface output — generated by a human or AI agent — verified against the relevant Protocol and Contract set before it ships.

The manifest

The manifest is the central artifact of an Agentic Design system. It is a single, versioned, machine-readable schema encoding the complete semantic knowledge of the design system: all Signals with their roles and constraints, all Intents with their resolution logic, all Contracts with their rules, and all Protocols with their surface-specific resolution maps.

In ADS, the manifest is described as "the product" — the primary output of design systems practice — with the component library, Figma libraries, and platform implementations all recast as downstream artifacts derived from it.

Manifest server

A conforming ADS exposes the manifest through a manifest server: an API infrastructure that makes the full semantic knowledge of the design system queryable in real time. The manifest server specifies five endpoints: a full manifest endpoint, a query endpoint, a validate endpoint, a diff endpoint, and a surface-resolved manifest endpoint.

The validate endpoint is integrated into CI/CD pipelines, making Contract compliance a build-time enforcement rather than a post-hoc review.

The context cascade

The ADS methodology describes a property of the manifest-driven workflow termed the context cascade: the characteristic by which each stage of the design and development process inherits the full semantic context of all decisions made in prior stages, rather than receiving a lossy translation through human handoff.

Blanding

Blanding is a term introduced by the Agentic Design methodology to describe the aesthetic convergence of digital products toward a shared visual language that belongs to no specific brand. The term is distinct from inconsistency in that blanding describes the convergence of technically correct, professionally executed products that are visually indistinguishable from one another despite representing different organizations.

AI blanding

The methodology identifies a more severe form termed AI blanding. When AI agents generate user interfaces without a semantic design system to reason against, they draw on training data weighted toward the most frequently documented patterns in the history of digital product design. The output is competent and accessible but statistically indistinguishable across products built by different organizations using the same agents and prompts.

AI blanding is distinguished from conventional blanding by its speed: where conventional blanding accumulates over years, AI blanding operates at the speed of inference. The ADS manifest is proposed as the structural solution, providing AI agents with a brand-specific semantic protocol to reason against.

Governance

The ADS methodology specifies a four-loop governance model:

  • Continuous loop — fires on every code commit, running manifest validation automatically via CI integration
  • Weekly loop — signal triage by a dedicated manifest owner reviewing override rates and validation failure patterns
  • Monthly loop — shipping versioned manifest updates with auto-generated migration guides for breaking changes
  • Quarterly loop — a full semantic audit comparing the manifest against actual product output

A dedicated role, the manifest owner, is specified as a non-optional governance requirement, responsible for the semantic correctness and currency of the manifest.

See also

References

Category:Design systems Category:Artificial intelligence Category:Software design Category:Design methodology Category:Product design

Content Disclaimer

Informasi ini disarikan dari Wikipedia dan disajikan kembali untuk tujuan edukasi. Konten tersedia di bawah lisensi CC BY-SA 3.0. Kami tidak bertanggung jawab atas ketidakakuratan data yang bersumber dari kontribusi publik tersebut.

  1. The information displayed on this website is sourced in part or in whole from Wikipedia and has been adapted for the purpose of restating it. We strive to provide accurate and relevant information, however:
  2. There is no guarantee of absolute accuracy. Wikipedia is an open, collaborative project that can be edited by anyone, so information is subject to change.
  3. It is not intended to constitute professional advice. The content displayed is for informational and educational purposes only. For important decisions (e.g., medical, legal, or financial), please consult a professional.
  4. Content copyright. Wikipedia is licensed under the Creative Commons Attribution-ShareAlike License (CC BY-SA). This means that content may be reused with appropriate attribution and shared under a similar license.
  5. Responsible use. Any risk arising from the use of information from this website is entirely the responsibility of the user.