Designers direct the system.
AI compresses the groundwork.
AI is being applied as an acceleration and systems-intelligence layer to support the foundational setup of the Tourism Australia Design System. Human-led, AI-assisted — designers direct, curate, and own all decisions; AI compresses the time-intensive work of discovery, extraction, and structuring.
The Experiment
AI is being used to audit multiple Tourism Australia domains and sub-properties — surfacing recurring UI patterns, shared foundations, divergent behaviours, and inconsistencies the design system will need to reconcile. From there, it assists in extracting and organising the underlying design primitives and identifying the components doing the majority of the UI heavy lifting across properties.
- Cross-domain auditing — systematic review of multiple TA domains and sub-properties to surface recurring UI patterns, shared foundations, divergent behaviours, and inconsistencies the system will need to reconcile.
- Foundation extraction & organisation — typography (scale, families, weights, line-height, responsive behaviour), colour systems (brand, semantic, surface, state, accessibility-paired), spacing, layout grids and breakpoints, elevation and shadows, radius, responsive and adaptive patterns, motion and interaction primitives, iconography, localisation (multi-script, RTL readiness, content expansion), and accessibility baselines.
- Component intelligence — identification of high-frequency components doing the majority of UI heavy lifting across domains, including variant patterns, states, and contextual usage. Informs prioritisation of what gets systematised first.
- Token & variable architecture — assistance setting up a scalable Figma variable and token architecture: primitive → semantic → component-level layering, with mode/theme structures anticipating future expansion (dark mode, sub-brand theming, campaign overlays).
- Taxonomy & documentation scaffolding — support for naming conventions, taxonomy structures, and documentation frameworks so the system speaks a consistent language across Figma, code, and written guidance.
- Where the time savings go — designer effort is redirected toward governance, accessibility, UX quality, system thinking, pattern coherence, and long-term maintainability.
AI is not autonomously designing the system. Every output is reviewed, curated, and refined under designer direction. Final decisions on UX quality, governance, accessibility, visual consistency, naming, and system architecture remain human-led and human-owned. AI is treated as a capable assistant for pattern recognition, extraction, and structuring — not as a decision-maker.
Risk Mitigation
Every AI-surfaced output passes through validation, review, and governance before it can influence the live system. Controls are designed to catch hallucination, over-fitting, bias, and accessibility regressions early — and to keep accountability with the design team end-to-end.
- Manual validation of extracted foundations, tokens, and component inventories against source domains.
- Design + Dev review for token accuracy, naming, and implementation feasibility before adoption.
- Governance checkpoints before any pattern enters the live system.
- Version-controlled variable architecture, naming conventions, and documentation; changes require an ADR.
- Accessibility verification — contrast, focus, semantic structure, WCAG conformance — on every foundation and component before sign-off.
- Source traceability — every AI-surfaced pattern links back to the domains and instances it was derived from; uncited patterns flagged for rejection.
- Bias & over-fitting check — designers verify AI hasn't over-weighted recent redesigns, the most visually prominent property, or surface-level repetition mistaken for system intent.
- Human approval gate — no foundation, token, or component goes live without designer sign-off.
Fall Back Plan
If AI assistance becomes unavailable, restricted, or deprioritised, the work reverts to a traditional manual design-system workflow. The outcome remains fully achievable — the trade-off is in time and operational overhead, not capability or quality.
- Manual cross-domain audits
- Manual token and foundation extraction
- Manual component inventory creation
- Manual Figma variable and library setup
- Manual documentation authoring and governance tracking
Migration Milestones & Component Plan
How design fits into the TA AEM → Sitecore migration. The highlighted column is what design owns.
| Milestone | Window | Highlights | Design |
|---|---|---|---|
| M1 · Discovery | Jun 2026 | Sign-off 30 Jun; Sprint 1 from 29 Jun. | Foundations by 18 Jun |
| M2 · Foundation | Jul–Aug 2026 | 90 components sign off in 5 batches. | Leads every batch |
| M3 · Build & Migrate | Sep–Nov 2026 | Corporate live 19 Oct · Business Events 30 Nov. | QA & sign-off in-sprint |
| M4 · Go Live | Jan–Mar 2027 | Signature + Aussie Specialist 1 Feb · australia.com 15 Mar. | UAT triage |
| M5 · Hypercare | Mar–Apr 2027 | ~30 days post go-live; Adobe ends Apr. | Governance handover |
Component Build Plan
Design builds all 90 components, handed over in 5 batches. The two plans below are the same work at different speeds: the realistic pace finishes end of August; the faster pace finishes the first week of August. Design has already finished 15.
| Batch | Total done | Realistic | Faster |
|---|---|---|---|
| 1 built | 10 | 3 Jul | 3 Jul |
| 2 | 25 | ~14 Jul | 10 Jul |
| 3 | 45 | ~25 Jul | 17 Jul |
| 4 | 65 | ~5 Aug | 24 Jul |
| 5 | 90 | ~28 Aug | 4 Aug |
Same work, two speeds — finishing on the faster plan depends only on each component's brief being signed off in time, not on design speed. The exact components per batch stay flexible; total is ~90.
System Progress & Health
A live reading of how far the system is built and how sound it is. The meter tracks overall build progress; the figures beneath it — variables, components, styles — are pulled from the live Figma library and refresh on each daily sync. Foundations are substantially in place; the element and component layers are where the remaining work sits. The dashboard syncs daily, drawing from the connected Figma library, the source-of-truth workbook, the Confluence page, and the project plan.
A component counts as complete only when all of its variants, every theme, and its documentation are finished. Partial builds report their true partial percentage — never rounded up. For example, Buttons reads as done only once every variant, every theme, and the written documentation are in place; until then it contributes the real fraction it has reached, not a rounded-up figure.
AI doesn't replace the design team.
It buys back their best hours.
The foundational phase of a design system is where the most consequential decisions get made — governance, accessibility, system thinking, long-term maintainability. It's also where the heaviest, most mechanical work lives: audits, extraction, inventory, scaffolding.
AI takes the mechanical layer. Designers keep the consequential one. If the assistance is pulled, the work continues — slower, heavier, but on the same trajectory.