Design Systems Are Data Systems: The Token Economy

Every senior designer has felt the sting of a rebrand cascading through a product — twenty shades of blue, three button radii, and a typography stack that someone’s intern approved in 2019. Design tokens were supposed to fix that. And they do, mostly. But as AI tooling embeds itself deeper into design workflows, tokens have quietly picked up a second job: they are now a cost variable.

Speckyboy’s Eric Karkovack put it plainly — AI tool costs are climbing, and every prompt, every generation, every iteration burns through tokens in both the literal and metaphorical sense. Undisciplined design systems amplify that burn rate dramatically.

Tokens as a Data Architecture Decision

If you work at the intersection of data and campaign execution — which is where most growth teams in Southeast Asia actually operate — design tokens should feel familiar. They are, structurally, a lookup table: a named variable that maps an abstract concept (color.brand.primary) to a concrete value (#E8431A). Change the value once, propagate everywhere.

The data parallel matters because it reframes how leadership should resource design system governance. This is not a cosmetic exercise. A well-structured token library reduces the surface area of ambiguity that AI tools need to resolve on every generation pass. Fewer ambiguous inputs mean fewer regeneration loops, which means lower compute costs — whether you are paying per-token on a generative API or burning seats on Figma’s AI tier.

For teams managing multi-language interfaces across Thai, Bahasa Indonesia, and Vietnamese — a routine reality in Southeast Asian brand operations — tokens also enforce the discipline that prevents localisation from fragmenting your visual system. A button label that expands 40% in Thai still renders correctly if padding and font-size are tokenised rather than hardcoded.

What Undisciplined Token Use Actually Costs

Karkovack’s argument is essentially about waste: teams that treat AI tools as infinite-capacity interns end up paying for the disorganisation they failed to fix upstream. The same logic applies to design systems at scale.

Consider a regional e-commerce brand running campaigns across Shopee, Lazada, and a standalone app. Each surface has its own component spec. Without a shared token layer, every design handoff requires a manual translation — and every AI-assisted asset generation requires re-specifying constraints that should already be codified. That friction compounds. A Sequoia-backed Indonesian startup that rebuilt its design system around a strict token taxonomy in 2024 reported a 30% reduction in design-to-development cycle time, largely because engineers stopped asking clarifying questions about values that were now machine-readable.

---

Expressive Systems Don’t Contradict Structured Ones

There is a tempting false binary here: systematic versus expressive. The Los Cabbalos identity work for New York venue Public Records, covered by It’s Nice That, pushes back on that assumption effectively. The studio built a modular system — block and lino-esque print components, figurative illustration — that is simultaneously rigorous and kinetic. The structure enables the expressiveness rather than suppressing it.

This is the mature position for brand design teams. A token system does not mean visual sameness. It means the rules of variation are explicit and intentional rather than emergent and accidental. For Southeast Asian brands navigating Ramadan campaigns, Chinese New Year palettes, and platform-specific dark mode requirements simultaneously, this distinction is not philosophical — it is operational.

The implementation pathway is less glamorous than the theory: audit your current component library for hardcoded values, establish a token naming convention that your development team can also own (W3C Design Token Community Group format is the current standard), and gate AI-assisted design generation behind a style dictionary that injects your token values as constraints rather than suggestions.

The Human Judgment That Still Anchors the System

The UX Collective’s recent issue touched on something structurally interesting: the small delay that makes a brainstorm actually productive. Speed without pause collapses the distance between stimulus and output, which is precisely where creative quality lives. AI tools are extraordinarily good at eliminating that pause if you let them.

Design tokens are one mechanism for reintroducing principled constraint into AI-assisted workflows. But the governance layer — who decides when a token value changes, what the approval path looks like, how brand decisions propagate to engineering systems — still requires human judgment. The data analogy holds here too: a clean data model is worthless without data stewardship. The same is true of a token library without a named owner.

For marketing directors sitting one level above the design and engineering team, this is the conversation worth having now, before AI tooling costs become a line item that nobody can explain.

Key Takeaways

• Treat design tokens as a governed data layer with named ownership — not a one-time setup task — to prevent visual debt from compounding across channels.
• Gate AI-assisted design generation behind a style dictionary that injects token values as hard constraints, reducing regeneration loops and controlling compute costs.
• For Southeast Asian multi-language deployments, tokenise spacing and typography scale separately to accommodate script expansion without breaking visual hierarchy.

The deeper question is whether most marketing organisations are structured to treat design infrastructure with the same governance rigour they apply to data infrastructure. The teams that answer yes are going to find AI tooling genuinely multiplicative. The teams that don’t will spend 2027 wondering why their AI spend keeps climbing while output quality stays flat.

---

At grzzly, we work with regional brand and growth teams to connect design system architecture to campaign performance — mapping where visual inconsistency is creating friction in your conversion data, and helping teams build the governance structures that make AI-assisted production actually scalable across Southeast Asian markets. Let’s talk