A single headless reactive core (Solid signals, class composition) with two manifestations from the same state: a 2D DOM face (Solid) and a 3D three.js face that renders the same DOM face into an HTMLTexture on a panel. Toggle or select anywhere — these controls, the DOM list, or by clicking a row in 3D — and both faces update in lockstep.
empirical probe: how do the HTMLTexture + pick pipelines fare off WebGPU? Shared choice (all DOM-in-3D test routes), persisted; toggling reloads — a canvas can't swap context kinds in place.
This hands-on workshop examined what AI can and cannot do for the legacy systems public agencies have inherited — systems built decades ago, by people who have since moved on, with documentation that is thin or wrong. The session framed the territory between two unhelpful extremes: the hype that AI will translate your COBOL by Friday, and the dismissal that AI engineering is just slop.
1. The hard part is understanding, not translating
2. Translation alone reproduces the past
3. Think in three disciplines: design, architecture, development
4. AI raises the value of human skills
three face — the same DOM face as an HTMLTexture — click a row in 3D to drive the shared core. Uses the WICG html-in-canvas API (Chrome 150+ flag) or the bundled polyfill.