A century ago, a well-off man had a valet. Not a driver, not a personal assistant, a valet: the person who looked after his wardrobe. He knew every suit, knew what was pressed and what was at the tailor's, laid out the right things for the weather and the day, and when his employer reached for the wrong shoes to wear with tweed, he quietly steered him away. One thing a valet never did was dictate taste. That belonged to the man he dressed. The valet carried only the invisible, laborious layer underneath it: memory, order, preparation.
The valet disappeared with the twentieth century, a luxury of a class and an era that ended. Now, oddly, he's coming back, just not as a person. My valet is a folder of text files, and I carry it in my pocket.
I should say where I'm coming from. My companies have been built on software for over ten years, I invest in digital technology, and I work with AI every day. I wanted to see whether the same kind of system I use for work could handle the seemingly simplest decision of all, what to put on in the morning. It took one afternoon. Since then I buy less and dress better, and that led me to a thesis about the whole industry.
Why we buy too much
Fast fashion doesn't run on clothes being cheap. It runs on an imbalance. Choosing well means knowing your measurements, what you already own, what goes with what and what will last. That takes thinking, and thinking, as we know, hurts. Buying is one click and returns are free. When one side of the equation is heavy and the other trivial, people behave rationally and buy volume as insurance against not knowing. Thirty pieces a month, a few of them worn, the rest hanging in the wardrobe with the tags still on.
Now hand that thinking, what suits me and whether I need it at all, to a tool that knows your wardrobe, your measurements and your taste. The cost of deciding drops close to zero. And once deciding is free, the question at the till turns around.
What it looks like inside
There's no magic. It's a few text files and one recurring task.
Photograph the wardrobe
One afternoon with a phone. Every piece I own, one photo at a time.
The base is an inventory. Every piece I actually own has its own row and its own ID: a T-shirt is tee-014, a leather biker jacket is jacket-006. Each row says more than "black jacket". It records the exact silhouette: volume, length, cut. A cropped, tight hoodie and an oversized one play opposite roles in an outfit. One is a base under a jacket, the other is the statement the whole look stands on. I didn't write the list by hand. I photographed the pieces in my wardrobe, uploaded the photos, and Claude pulled out the silhouette, colour and material on its own, with roughly eighty percent accuracy.
Above the inventory sits a layer of taste: named worlds I dress in. The Uniform is the base. A T-shirt, denim and one playful detail, a bracelet or a pendant. It looks like nothing and is really what the Swedes call lagom, not too much, not too little. At the same time I take a cue from John Mayer, who never tied himself to a single genre. Wearing cargo trousers and a kimono with visvim FBT Lhamo-Folk can be a perfectly reasonable choice. Rules are there to be broken.
A valet was never about choosing for his employer. It was about freeing his hands.
That's the point. AI takes on the parametric part, what I own, what fits and what goes with what, which leaves more room for the creative part, playing with genres. If you already have taste, this won't turn you into a cliché. It gives you more freedom.
Then you just ask. "I like these shoes, does it make sense to buy them?" gets an honest answer set against what you already own, not against a shop's conversion target. "My techwear outfit needs an interesting accessory" first gets you shown whether you already have one at home.
The last part of the valet is an automated agent that keeps the wardrobe up to date with what I've ordered, and it runs without me seeing it. Once a week it goes through my email, pulls out new orders and returns, and keeps the inventory current. It works in two stages, because I send back about a third of what I order. A new piece goes into staging first and only enters the permanent inventory once it has survived the return window. So the memory of what I own only ever holds things I actually kept.
The heart of it
For the last ten years, Instagram and Pinterest have taught us one simple equation: you like it on someone else, so buy it. It doesn't add up. The photo shows the piece on a different body, in a different wardrobe, in a different mood. An advisor built on your measurements and your preferences doesn't save you money by forbidding what you like. It saves it by catching the other case, the thing you like on its own that ends up unused at home.
Go back to the person who runs through thirty fast-fashion pieces a month. They don't do it because they're stupid or vain. They do it because deciding is expensive and buying is cheap, so they buy certainty in bulk. Put a valet in their pocket and they'll buy less but better, and in the end maybe for the same money as the original thirty. Today that takes time and know-how. Tomorrow it will be one question in a chat on your phone.
Where the thesis breaks
First, it will take a few years, not a few months, although the technology is ready. Second, it will never reach everyone. If you don't care how you look, no tool will give you taste. Third, the same technology can end up doing the exact opposite: hyper-personalised fast fashion that serves you even more things you'll buy and never wear. Which way it goes depends on one question, whether the tool optimises your wardrobe or someone else's conversion. Mine is built to stay on my side. Once shops start building these tools, they'll have every reason to turn that around.
How to build your own
You don't need to be technical, and you don't need to use Claude. I put my valet together as a skill in Claude's Cowork mode, but any capable AI tool would do. It took one afternoon, four hours at most, including photographing the wardrobe and fixing what the AI read wrong.
The easiest way to get one is to take a photo of this article, upload it to your AI app and say you want the same system, and that it should ask about everything it needs to know. Its questions will draw out the same context I went through: what's in your wardrobe, your measurements, the styles you wear.
Or start from mine. Valet de chambre is on GitHub. It ships empty, walks you through setup the first time, and keeps your closet, measurements and photos on your machine.
For now my valet is a folder of text files that took over the tedious part: keeping track of what I own and how it combines. I've always enjoyed thinking about clothes. This only took away the work and left me the room to think creatively. A strange direction for technology to push us in. Or, to borrow a question from King Gizzard & the Lizard Wizard, is this really changing for the better? Maybe.