Your brand is an operating system, not a PDF

← All insights

Most brand guidelines fail at scale because they’re built like operating systems from 1995. You design a system, lock it in a PDF, and hope people follow it. They don’t.

I spent 18 years watching enterprise brands degrade silently. A 50,000-person organisation across 100+ countries doesn’t fail because someone read a PDF wrong. It fails because the system was never meant to scale beyond a single region, a single campaign, a single medium. PDF-based guidelines are artefacts of a world where consistency meant static templates and brand managers meant gatekeepers.

That world is gone.

A design token is a decision capsule — a single piece of brand intent in a format machines can read, understand, and apply consistently across every medium. When you move to a token-based system, the brand team stops making individual decisions about what red should be on a poster versus a digital interface versus an AI-generated asset. One decision. The system propagates it everywhere.

I saw this principle at work through my Adobe Express work with AECOM — structured template governance across 50,000 people in 100+ countries brought template production time down by over 90%, improved consistency across every market, and gave designers more creative freedom. Designers redirected energy from policing hex codes to solving actual problems. Design tokens are the next evolution of that principle: encoding brand decisions into a system so deeply that consistency happens by default, across every platform, without human enforcement.

In October 2025, this moved from best practice to open standard. The Design Tokens Community Group announced the first stable version of the Design Tokens Specification (2025.10) — a production-ready, vendor-neutral format for sharing design decisions across tools and platforms — backed by Adobe, Google, Microsoft, Meta, Figma, Salesforce, Disney, and Shopify, among others. The fragmentation problem, where tokens lived in incompatible proprietary formats across different tools, now has an industry answer. New brands, products, or regions can be spun up in hours rather than months when this architecture is in place.

Tokens also bridge human intent and machine generation. Hand an AI system your tokens and it inherits your brand DNA. No 500-page guide required.

A composable brand system treats every component — colour, layout, typography — as independently deployable and recombineable. At scale, you don’t have uniform needs. A bid team in Singapore needs different templates than a recruiter in Toronto. A composable system accommodates that reality.

Enterprise design systems are evolving from passive repositories into active systems of interaction — providing the semantic intelligence that allows AI agents to build entire features while staying on-brand and on-system. The most mature systems stack permissions and constraints. A non-designer works inside pre-built templates. A designer modifies components within defined boundaries. Brand leadership controls the tokens at the top. No bottleneck. No gatekeeping. No PDFs.

Where previous years were about AI assisting designers, 2026 is about AI agents orchestrating the design-to-code pipeline — with modern systems using autonomous agents to detect design drift before it even reaches production.

The tools are catching up to the principle. Figma’s MCP server — which entered beta in 2025 — lets AI agents identify where to apply or introduce design tokens, suggest implementations that comply with existing standards, and audit differences between designs and code. The flywheel: AI strengthens your design system, which powers better AI code generation.

On the asset generation side, Adobe expanded access to Firefly Custom Models in public beta on 19 March 2026, letting teams train a model on their own images to capture a specific style, character, or photographic look — with trained models preserving stroke weight, colour palettes, and character features consistently across generations. For enterprise teams: Adobe Firefly Foundry enables organisations to deploy proprietary generative AI models trained on their own intellectual property across image, video, audio, vector, and 3D content, with early adopters including Walt Disney Imagineering.

There’s one more dimension that didn’t exist two years ago. Nearly one-third of digital marketing leaders now prioritise generative engine optimisation as the most critical performance hurdle for 2026 — because AI answer engines act as gatekeepers, presenting only the brands they deem worthy in synthesised responses. A brand that lives in a PDF is invisible to those systems. A brand built as structured, machine-readable data gets found.

The gap between organisations that understand this and those still managing PDFs is widening. Here’s where to start.

• Audit before you automate. Identify where your code and design have drifted — map every rogue component and hard-coded hex value. You can’t automate what you haven’t cleaned.
• Migrate to DTCG-standard token naming. Organise tokens into three layers: primitives (raw values), semantics (intent-based, like action.primary), and components (specific to an element). This structure is what makes tokens consumable by AI agents across any platform and essential for multi-brand scaling.
• Make your documentation machine-readable. Replace static PDFs with live, searchable documentation that pulls real code and design assets into one place. By making documentation machine-readable via MCP, your AI coding assistants stay compliant with the latest brand standards automatically.
• Train a custom model on your brand assets. The capability to train AI generation on your own creative IP is live and in public beta. For enterprise teams still generating assets generically, this gap will widen fast.
• Redesign the brand leadership role. Only 21% of enterprise leaders currently have a mature governance model for autonomous agents, even though these systems can initiate actions, interface with customers, and interact with core business processes. Brand leaders who understand system design — alongside visual identity — are the ones whose role compounds in value as AI execution scales.

This article was accurate as of late March 2026. Given the pace of change in AI tooling, specific platform capabilities referenced here may have evolved.