Place two images side by side. The first: a room on the first floor of a Venetian palazzo, around 1770. Three Murano chandeliers throw light off a wall of mirrors. A green baize table sits at the centre. Six figures in white baùta masks and black tabarro cloaks lean over a hand of cards. The second: a smartphone screen in 2026. The background is a digital green baize. The card animations slide with subtle drop shadows. A small animated chandelier glints in one corner. A red-velvet border frames the interface. Beneath the table, a row of gold buttons waits for the user's thumb.

The two images appear, on first inspection, to belong to entirely different worlds. On second inspection they share an unsettling amount of vocabulary. Green baize. Gold. Chandelier. Velvet. Cards. The same visual scenography, separated by two and a half centuries, reassembled in pixels. This essay is about how that migration happened — and, more interestingly, about how the way that migration happens has itself migrated through three successive generations of design tooling, ending in 2026 at a point where the citation is being made by AI systems that have never seen the original they are citing.

The grammar of the salon

The visual apparatus of the eighteenth-century European gaming salon was not arbitrary decoration. Each element did a specific job. The green felt rested the eye through long sessions and produced optimal contrast with white cards and gold coin. Mirrors and chandeliers built a chamber of light in which every gesture was visible and every facial expression legible — when not concealed by the mask. The gilding and ornament communicated the room's place in a specific social hierarchy: salons were expensive to build and maintain, and the expense was part of the message. The intimate room scale — a few square metres, six to eight players — produced a concentration of attention that larger halls could not.

Every component, in short, had a function. Green was ergonomic. Mirrors were social. Gold was classificatory. Scale was cognitive. The mask was egalitarian inside the room. The visual apparatus was not ornament: it was a finely tuned cultural device assembled over generations to produce a particular experience.

The Belle Époque casinos of Monte-Carlo and Baden-Baden, between 1860 and 1914, codified this vocabulary at architectural scale, with Charles Garnier's 1878-79 expansion of Monte-Carlo importing the entire ornamental syntax of the Paris Opera into a gambling environment. By the early twentieth century the grammar had stabilised into an international design standard that every casino floor anywhere in the world would, knowingly or otherwise, imitate.

When digital translation arrived, the language was already in place.

First migration — skeuomorphism, 1995-2007

The first generation of digital entertainment interfaces, from roughly the mid-1990s through the late 2000s, did the migration by hand. Engineers and designers worked together to translate the visible signs of the salon into bitmap and vector graphics. Green felt textures were drawn pixel by pixel. Card images were photographed and digitised. Chip stacks were modelled in three dimensions and rendered with subtle drop shadows. Chandeliers, where they appeared, were static decorative elements imported from stock libraries.

This was the era of explicit skeuomorphism, the design philosophy associated most prominently with early Apple interfaces but pervasive across virtually every consumer software category. The argument was that digital surfaces, being new and disorienting to users, should imitate familiar physical materials to make them legible. The argument carried weight. It also produced a great deal of design that looked like its analogue ancestor at the cost of taking advantage of what the new medium could actually do.

Programming this period was a manual exercise. Each visual asset was hand-crafted, the interface was assembled in code that referred to those specific assets by name, and the design system — to the extent that the term applied — was a folder of PNGs and a stylesheet. The aesthetic was inherited, but the inheritance was performed by individual labour, one detail at a time. A designer who wanted a green baize background drew one. A developer who wanted a chip stack imported the image their teammate had produced. The salon was citing itself through a series of explicit human acts.

Second migration — design systems, 2008-2022

The second generation, beginning around 2008 with the rise of mobile and accelerating through the 2010s with the maturation of design systems, abstracted the manual labour into reusable infrastructure. Material Design, Apple's Human Interface Guidelines, Bootstrap, Tailwind CSS, Figma component libraries — together these produced a working environment in which the visual vocabulary of an interface could be assembled from pre-existing components, with the aesthetic decisions encoded in design tokens rather than authored fresh for each project.

For digital entertainment platforms specifically, this period saw the emergence of what the industry called "live dealer" interfaces — video streams of real dealers in studios in Riga, Bucharest, and Manila, overlaid with HTML interfaces that let the user place bets on real cards being dealt. The visual language was deliberately archaic: jacket, white shirt, bow tie, mahogany table, exact reproduction of a Monte-Carlo gaming environment minus the building. The HTML overlay used the same design tokens that the rest of the platform used for purely digital tables — green, gold, red, the same typography, the same button styles.

Programming had also changed. By 2018-2020, a typical digital entertainment interface was built in React or Vue, styled with Tailwind, populated with components from a shared design system, and assembled by a developer who almost never made aesthetic decisions from scratch. The aesthetic was inherited from the tokens; the tokens were inherited from the Belle Époque; the Belle Époque was a codification of the Venetian salon. The chain of citation was now mediated by code rather than performed by hand.

It was, in retrospect, the high point of the citation-driven model of interface design. The aesthetic was stable, the toolchain was mature, and the decisions made by individual developers and designers were narrowly bounded by the design system's existing vocabulary. Everyone in the industry agreed on what a casino interface looked like, and the agreement was encoded in the tokens.

Third migration — vibe-coding and AI generation, 2023-2026

The third migration is happening now, and it is doing something different from the previous two.

The term "vibe-coding" entered the working vocabulary of software engineering in early 2025, when Andrej Karpathy used it to describe a way of working with AI coding assistants in which the developer describes the desired outcome in approximate natural language and lets the model produce the implementation. The phrase has expanded since then to cover a broader practice: working with AI assistants for design and front-end development by specifying aesthetic and behavioural intent without specifying the implementation details. A developer using a tool like Cursor, Claude Code, or GitHub Copilot to build a digital entertainment interface in 2026 is often working in a mode that would have been unrecognisable in 2018. They describe what they want — "make this look like a high-end European casino" — and the model produces components, colour palettes, animations, typography choices, layout decisions, and code, in seconds.

The interesting question, for anyone watching this happen, is where the aesthetic specification is coming from. The developer typing the prompt does not necessarily know what a high-end European casino looks like. The model does. It knows because its training data includes thousands of examples of digital entertainment interfaces produced during the previous two generations of design migration, plus photographs of physical casinos, plus design system documentation, plus art-historical material on Belle Époque architecture and Venetian salons. The model has learned the vocabulary not by visiting Monte-Carlo but by reading the work of designers who had themselves learned it from earlier generations who had themselves learned it from people who had perhaps, somewhere far back, actually been in a Venetian palazzo.

What is striking about the AI generation is that it does not produce surprising results. A prompt asking for "a casino interface" in 2026 reliably produces green baize, gold accents, red velvet, ornate typography, animated chandeliers, sometimes a stylised silhouette of a chip stack. The model is not innovating; it is averaging. The output is the centroid of two centuries of citation, smoothed by statistical aggregation and rendered in whatever the current front-end framework happens to be. The aesthetic that was once an explicit decision by a Garnier or an Apple designer or a Material Design committee is now, increasingly, a default output of a language model that has read everything those decisions produced.

The consequence — citation without source

The consequence of all this is that the aesthetic of the Venetian salon now lives in digital entertainment in a peculiar new way: as a statistical pattern in a model's weights rather than as a deliberate choice by a designer. The chain of citation has become, in places, fully automated. A developer in 2026 can ship a casino interface without ever having looked at a real casino, without having read a design system manual, without having made an explicit aesthetic decision. The AI does it. The AI does it well, in the sense that the result is recognisably a casino interface. Whether the resulting screen is a homepage, a registration flow, or a familiar entry point such as a DicePalace login page, the visual language increasingly emerges from accumulated training data rather than from a designer consciously selecting references. The AI does it without understanding what any of the elements were originally for.

This is not a complaint. It is an observation about how cultural transmission has changed. For most of the last three hundred years, the migration of the Venetian aesthetic happened through human acts of citation: a Belle Époque architect citing a Settecento salon, a 1990s designer citing a Belle Époque casino, a 2018 developer citing a 1990s design system. Each act was performed by a human who held some image of the source in mind, even if the image was second- or third-hand. The new generation of vibe-coded and AI-generated interfaces breaks that chain. The citation is being made by a system that has no image, no source, no intent — only a high-dimensional statistical average of every act of citation that preceded it.

What this means for the future is unclear. One possibility is that the aesthetic ossifies — that the AI-driven default becomes so pervasive that no new design language can compete with the gravitational pull of the centroid, and digital entertainment platforms in 2040 will look approximately like digital entertainment platforms in 2026. Another possibility is that the very ease of generating the default frees designers to deviate from it deliberately, producing a wave of intentional anti-citation work that resembles nothing the Venetian salon ever produced. A third possibility — historically the most common outcome in design — is that something between these two happens. The default stabilises for the mass market; a small avant-garde produces variations; the variations slowly migrate into the default; the cycle continues.

What is certain is that the eighteenth-century Venetian salon is still in the room. Three hundred years after its inhabitants put on their masks and sat down at their green baize tables, their visual choices continue to shape the screens of every digital entertainment platform their successors interact with. The mechanism of transmission has changed three times. The transmission has not stopped. Whether anyone designing those interfaces today knows that the chandelier in the corner is a reference to a Murano original in a palazzo on the Grand Canal is, in 2026, an increasingly philosophical question — the kind that the chain of citation, having become invisible to the people performing it, may have made permanently unanswerable.