Design in the Age of Infinite Generativity

Consider one of the strangest facts about human language. A finite vocabulary — a few thousand common words, a handful of grammatical rules — can produce an infinite number of sentences. Linguists call this property infinite generativity. Every child masters it before learning to multiply. You are doing it right now, reading a sentence that you have almost certainly never encountered before in exactly this form, and understanding it without effort.

For most of human history, design did not work this way. Design was constrained. A beautiful cloth had to be woven by a weaver. A building had to be drawn by an architect. A typeface had to be cut by a punchcutter. The vocabulary of design was finite, and the grammar of execution was expensive. Ideas came cheap; making them real was the hard part.

On the morning of March 21, 2026, at the Chennai Design Festival, a man named Anand — officially the world's first LLM Psychologist at Straive — walked onto a stage before an audience of designers and made a claim that should have sounded alarming but somehow felt liberating: AI is about to give design the same property that human language has always had. Infinite generativity.

The host had introduced him as someone who would "show us that as we design the future of AI, we need to stop being just programmers and start being philosophers." It was a grand billing. But Anand is not, by temperament, a philosopher. He is a tinkerer. What followed was not a philosophical treatise but a rapid-fire demonstration of what happens when an endlessly curious person with access to every AI tool on the market decides to see what's actually possible in design — and then shows the receipts.

For years, the standard model of creative work looked like a bell curve. Ideation — the spark, the concept — sat on the left, taking maybe ten percent of the effort. Execution occupied the fat middle: the 80% spent rendering, building, iterating, implementing. Evaluation sat on the right at another ten percent. The shape of that curve was a law of the creative universe, as stable as gravity.

Then something shifted. Slowly at first, then suddenly.

Anand put up a visualization — built entirely by Claude from a prompt he wrote — showing exactly what is happening to that curve:

Watch the visualization long enough and something uncomfortable sets in. The execution hump — the fat middle where most creative professionals spend most of their careers — is smoothly flattening. The ideation and evaluation dots, previously dwarfed, are now taller. The shape of work has inverted.

This is not a distant prediction. It is already happening, and the economics are accelerating it. The question that concerned Anand — and, by the end of his talk, the audience — was not "will this happen?" but "what do we do with the 75% of effort that just got freed up?"

His answer came in three acts: Ideation, Execution, and Evaluation. Not just as stages of design — but as new territories, each opened up by AI in ways designers hadn't fully imagined.

The Vocabulary Problem

Anand has no formal design education. He will tell you this cheerfully, without apology. He is a coder, a data scientist, someone who spent decades making things with software — not with brushes or fabric or light. And for most of his career, this meant his vocabulary for describing what he wanted to create was limited. You can't ask for what you don't have words for.

So he did what any curious person with access to an LLM does: he asked.

"Give me some new designs. Tell me what are the different kinds of styles, art techniques that I can use, and what are the different prompts that I can use for this."

The AI came back with a gallery of techniques Anand had never heard of. Decalcomania. Grattage. Sgraffito. Stippling Mixed Media. And then, buried in the list, the one that stopped him cold: the Xerox style.

"Effectively a style of art which is simply based on photocopies of photocopies of photocopies, and it has a very gritty and interesting effect."

Each of these was a door. And behind each door was a prompt — a set of words that could be passed to an image-generation model to produce something in that style. Anand had built a gallery:

LLM Art Style Explorer — techniques including Decalcomania, Grattage, Sgraffito, Stippling, and dozens more. Click to explore.

This is a more profound act than it first appears. The limitation in creative work is rarely ability — it is vocabulary. Designers who have spent years in formal education acquire hundreds of terms for things. Someone who hasn't spent those years is not less capable of recognizing beauty; they simply have fewer handles to grab onto. AI doesn't replace the human judgment. It hands you the vocabulary at speed.

The Outlier Map

But discovering what's possible in the abstract is only half the ideation challenge. The more interesting question is: who is doing things nobody else is doing? Where are the outliers — the designers whose vocabulary is genuinely different from everyone else's?

Anand's team approached this empirically. They took roughly 200 charts from ten major publications — the New York Times, the Wall Street Journal, The Economist, South China Morning Post, Reuters, Bloomberg, The Guardian, The Pudding, The Washington Post — and ran them through an image embedding model. Similar charts cluster together. Very different charts end up far apart.

The insight here is deceptively simple. AI needs inspiration just as much as humans do. You cannot prompt your way to genuine novelty with only generic descriptions. But if you can find the actual outliers — the practitioners at the edge of the distribution — you can feed those as examples and use AI to generate variations on genuinely unusual ideas. The outlier becomes the seed. AI becomes the germinator.

This is the new skill: not just prompting, but curating the inputs. Finding the humans who think differently. Building the community at the frontier. Figuring out who the outliers are. As Anand put it, "that's where our job is." And generative AI is making even that easier.

The talk was not purely theoretical. Anand had spent the hours before the session sitting with ideas from the designers around him — and running them through AI to see what would come back.

Take Senthil and Amutha, founders of Payir, a grassroots organization in Tamil Nadu. They had demonstrated at the festival a re-usable fabric calendar — a calendar that, when its year is done, converts into a bag. It was a small, beautiful example of what the founders believed: that good design can be for the masses and by the masses, reclaiming time, energy, and joy.

Anand fed this idea to Claude and asked for half a dozen ideas in the same spirit. What came back was a full conversation worth reading — Claude reasoning carefully about Payir's ethos before generating ideas. The results:

The Modular Kolam Mat

A fabric mat with a grid of snap-on/Velcro tiles in traditional kolam patterns. Each morning, rearrange the tiles to create new designs — teaching geometry, symmetry, and cultural patterns to children while serving as a doormat. When tiles wear out, they become cleaning rags. The base mat eventually becomes a shopping bag. It's meditation, math lesson, home decoration, and zero-waste design in one object.

The Growth Chart Kurta

A child's kurta with measurement markings embroidered along the seam. As the child grows, parents mark their height with simple stitches or fabric markers. When outgrown, the garment is unpicked and reconstructed into a smaller kurta for a younger sibling, or a pouch, or a doll. The measurements become a family record — a fabric memory that doesn't end up in landfill. Every stitch holds a story.

The Seed Library Sari Border

Sari borders designed with pockets holding indigenous seed varieties — millets, traditional vegetables, pulses. Women exchange border sections at community gatherings, literally trading seeds while trading stories. The border can be removed when the sari wears thin, becoming a seed-storage belt or headband. Agricultural preservation disguised as fashion; distributed knowledge disguised as decoration.

The Recipe Towel Chronicle

Kitchen towels with regional recipes printed in Tamil — not just ingredients, but the why behind each step (fermentation science, spice medicine, seasonal timing). As towels wear from washing, the faded sections reveal which recipes the family cooks most. Eventually the towel becomes produce bags or pot holders. Culinary knowledge literally woven into daily rhythm, aging gracefully into new utility.

What made these ideas remarkable was not just that they were good — though they were. It was that they were culturally specific in the way only good design is. AI, given examples from a specific community with a specific ethos, produced ideas that honored that community's constraints, materials, and values. Not generic solutions. Not Western solutions. Kolam patterns, Tamil recipes, Indian saris, village seed-trading networks. The vocabulary of the training data, selectively invoked.

Everyday Design Problems

Anand had taken the same approach with ideas from another designer at the festival: Narendra Ghate, whose ideas ran toward the elegant and behavioral — dimming lights to reduce office noise, tearable double shampoo sachets, stain band-aids.

He fed those to Claude and asked: give me half a dozen ideas that'll make us go "Wow! That's clever AND high impact." Read the full conversation. What came back:

Medication Packaging That Changes Color When Expired

pH-sensitive ink on blister packs that shifts from green to red when the drug degrades. Not a printed date you squint at — the whole pack turns orange when it's time to discard. Works because most medications release acids as they break down.

The "Ready to Order" Coaster

Flip the coaster from green to red when you've decided. Solves the eternal "hovering waiter vs. neglected table" problem without apps, buttons, or staff training. Gives customers agency while signaling intent clearly. Costs nothing, reduces frustration on both sides.

USB Cables That Glow When Drawing Phantom Power

A thin phosphorescent strip that lights up when current flows — even 0.1W. Surfaces the invisible vampire drain from chargers left plugged in. People don't unplug because they can't see the waste. Make it visible, behavior changes instantly. Also an instant diagnostic for "why isn't this charging?"

Reverse Parking Sensor for Shopping Carts

When a shopping cart is pushed backward, it emits a gentle beep — like a truck reversing. Solves the epidemic of heel-clipping in grocery aisles. Simple piezo buzzer triggered by wheel direction. Cost: pennies. Impact: massive goodwill.

What all of these shared, Anand noted, was a quality Claude's response explicitly named: "They surface invisible problems... they leverage existing human instincts... they feel obvious in hindsight — the mark of great design."

AI had not just generated ideas. It had generated ideas with a theory of why they worked.

The Tool Problem

Here is a confession that Anand made with no embarrassment whatsoever: "I've been using ImageMagick as a tool for the last what, thirty years or something like that, ever since it was invented probably."

Thirty years. And yet — until he asked an LLM to "go crazy, create all kinds of ImageMagick filters and give me the code for each one" — he had no idea what the tool was truly capable of.

Gallery of ImageMagick filters discovered by asking an LLM to "go crazy." Each thumbnail links to the code that produces it.

The AI cataloged the possibilities, generated the outputs, and produced a gallery of how to use the tool. An encyclopedia of its own capabilities, written by a machine that had read every StackOverflow answer, every documentation page, every blog post that had ever mentioned the word "ImageMagick."

This is a subtle and important point. The gallery is not for Anand. It's for the next AI agent. He is building institutional memory for a system that learns — a vocabulary, a prior, a reference library. When you ask an agent to create an image effect, having a concrete gallery of possibilities dramatically improves the output. You can say "go towards the esoteric end" or "try something from the bottom half of this." The human role becomes steering, not rowing.

The Blender Intern

And then there was the architecture story, which is perhaps the most revealing.

Anand had absolutely no experience with Blender. Neither did the intern he assigned this to. Here was the task: take a single building, and using AI with a Blender MCP, create a campus.

The intern spent one hour installing Blender. One hour prompting. The AI looked at the building, inspected it, created a copy, looked at the result from different angles, said "okay, let me create two more," created four sites, zoomed out, zoomed in, made adjustments — and finished with a campus in roughly five prompts.

But Anand's most provocative line came after this story:

Scale. This is the word that kept surfacing. Not just that AI makes individual tasks faster. But that it enables a kind of parallel, distributed execution that was structurally impossible for individual humans. You cannot personally explore ten different design directions simultaneously. You can direct ten AI agents to do exactly that, in the time you'd spend on one.

The Humble Critic

This was the section of the talk that felt most personal — and most honest. Anand had been contacted by an ex-colleague who sent an image of a chart and asked for feedback. Anand did what he now does reflexively: he uploaded it to Claude and asked for a critique.

Anand paused. "It went on to a series of review suggestions which I, who have several years — in fact, decades of experience in data visualization — am not able to beat."

And then, unprompted, a small act of intellectual honesty that landed in the room: "Of course, I'm lazy, that's another point. But it is actually better than me. So why would I hang on to my experience, my skills, whatever? Let me learn from it. What is the harm? Some humility always helps."

The evaluation space, Anand argued, was perhaps the most valuable and least explored of the three. Ideation and execution get the attention. But systematic evaluation — the kind of rigorous, experienced critique that most design work never receives — is where AI can add the most immediate value with the least risk.

Leave the Math to the Machines

The talk had been moving fast. Slides, demos, images, videos, AI conversations — each one a data point in Anand's broader argument. And then, at the end, he stopped. He put up a single image.

It was a perfect ending — funny, deflating, and somehow true. If AI is going to handle the execution, optimize the variations, critique the outputs, catalog the tools, and generate the ideas — what remains?

Play. Curiosity. Wandering into territory you don't understand. Finding the outliers. Building the community. Asking the questions that the machine doesn't know to ask because no one has yet told it what matters.

"But if AI is going to be doing all of this, what are we going to do?" Anand had asked. His answer, wrapped in a six-year-old's logic, was actually quite serious: the human job, in the age of infinite generativity, is to play. To explore. To bring the domain knowledge, the cultural context, the why that no training dataset has fully captured. To sit in a room at the Chennai Design Festival and look at a man converting a calendar into a bag and say: what else could this be?

The machine amplifies from there. But the question has to come from somewhere. The question has to come from you.

"Go have fun, everyone."