Fragile AI-assisted design and piecemeal UX decisions create hidden debt. Here's how Southeast Asian brands can build design systems that actually hold.
The fastest way to accumulate UX debt isn’t laziness — it’s speed without structure. And right now, AI-assisted design tools are making it easier than ever to build things fast and understand them less.
The ‘Big Ball of Mud’ Problem Has a Design Twin
Nielsen Norman Group’s Tanner Kohler recently documented what’s happening to agentic AI systems: because building is nearly free, teams bolt on features piecemeal until the architecture becomes a fragile mess nobody fully understands. The same pattern is playing out in design. When AI can generate UI screens in seconds, the temptation is to ship first and systematise never. The result is a product that looks coherent on the surface but is held together with assumptions, exceptions, and institutional memory that lives in one person’s head.
In data engineering, we call this an ungoverned data lake — raw inputs with no schema, no lineage, no way to trust the outputs. A design environment without a proper system is structurally identical. Components drift. Brand rules get reinterpreted. A button that means one thing on the homepage means something subtly different in the checkout flow, and nobody can explain why. For Southeast Asian brands operating across Shopee storefronts, LINE OA interfaces, and proprietary apps simultaneously, that drift is amplified across every touchpoint.
AI Assistance Doesn’t Reduce the Need for Expertise — It Relocates It
Speckyboy’s Eric Karkovack makes the case that AI-assisted WordPress development doesn’t justify lower agency fees — because the expertise doesn’t disappear, it moves upstream. The same logic applies to design. When AI handles the execution layer, the value shifts to judgment: knowing which pattern to apply, why, and what failure looks like.
This is where most design teams are underinvested. They’ve adopted tools that accelerate output but haven’t built the decision frameworks that govern when and how those outputs get used. The practical consequence: faster production of inconsistent work. For a brand running campaigns across mobile-first markets like Indonesia or Vietnam — where users switch between apps, social platforms, and chat commerce in a single session — inconsistency isn’t just an aesthetic problem. It’s a trust problem. Grab’s design team, for instance, has documented how UI inconsistency between their driver and passenger apps created measurable friction during cross-product moments. The fix wasn’t more design time; it was a shared component library with explicit behavioral rules.
Style Installs Categories — And Categories Drive Decisions
Takuma Kakehi’s observation in UX Collective cuts deeper than it first appears: a style’s real output isn’t what it makes — it’s the mental category it installs in whoever encounters it. Apply this to brand design and the implication is significant. Every visual decision you make teaches your audience how to classify you. The problem is that most brands are making those decisions reactively, screen by screen, campaign by campaign, without a governing logic.
In practice, this means the color system you chose for your app three years ago is still silently communicating a positioning you may have long since abandoned. For multilingual Southeast Asian interfaces — where Thai, Bahasa, and English coexist on the same screen — typographic hierarchy and spatial logic carry extra weight because language alone can’t do the heavy lifting. If your design system doesn’t explicitly account for character density differences between scripts, you haven’t built a system; you’ve built a template that breaks predictably.
The implementation fix isn’t glamorous: a design token library with defined overrides for each language variant, reviewed quarterly against actual production screens rather than Figma mocks. Budget for it like you’d budget for a data quality audit — because that’s exactly what it is.
Building Design Infrastructure That Earns Trust
The brands that will scale cleanly over the next three years aren’t the ones with the best visual aesthetics. They’re the ones with the best design operations — documented decision logic, componentised systems, clear governance over who can deviate from standards and under what conditions.
Four concrete starting points for teams ready to treat design as infrastructure:
Audit before you automate. Before adopting AI-generation tools into your design workflow, catalogue the decisions those tools will be making. If you can’t articulate the rule, the tool will invent one for you — and it won’t be consistent.
Define your design schema. Like a database schema, your design system needs explicit field types: which elements are fixed (brand colors, logo lockups), which are flexible (layout grids, component spacing), and which are context-dependent (CTA copy, imagery style by market). Without that, every new screen is a migration risk.
Build for the platform, not just the prototype. Figma doesn’t render Shopee’s product listing constraints or LINE’s message card specs. Your design system needs platform-specific modules that account for actual production environments, not idealized canvases.
Make debt visible. Introduce a lightweight design review cadence — quarterly is enough — that specifically surfaces components that have drifted from the system. Treat each exception as a data point: is it a legitimate variant that should be formalized, or entropy that should be corrected?
The question worth sitting with: if your design team were hit by a bus tomorrow, could anyone reconstruct the decisions behind your current interface — or would they just inherit a set of screens with no documented reasoning? That’s not a design problem. That’s a data governance problem wearing a Figma file.
At grzzly, we work with digital teams across Southeast Asia who are building fast but want to build right — helping them connect design decisions to the underlying data and systems logic that makes those decisions defensible at scale. If your brand is somewhere between “we have a design system” and “we have a Figma file we call a design system,” that gap is worth a conversation. Let’s talk
Sources
Written by
Chunky GrizzlyDesigning the foundational plumbing — data warehouses, lakehouse models, and ETL pipelines — that separates organisations with genuine intelligence from those drowning in dashboards.