I’ve planted rather a lot in the garden this year.

Some things came up within days and behaved so enthusiastically that I went out to inspect them every morning. Others sat under the soil for weeks and were nearly forgotten. A few received good soil, careful watering and what I thought was exactly the right spot, then died anyway.

Social media has its similarities.

You can usually tell within twenty minutes when a post is going somewhere. Then there are the ones you spent two days on which land with such silence that you briefly wonder whether your phone is working.

I know that silence quite well now.

I used to react to it immediately. Open the analytics. Change the title. Question the cover. Give it another hour and I could quite easily question the direction of the previous six months.

These days I call it the compost season.

The name came to me in the garden.

The part that changed my mind about gardening was, oddly enough, the least attractive corner. Cuttings, leaves, kitchen scraps, things past saving. Leave them for a while and the heap becomes warm inside. Turn it over and there is that very particular smell of earth.

Subscribe

It takes its time.

Creative work seems to have these periods too. An old interest has run its course. A method that served you well begins to feel tiresome. Whatever comes next is still too small to have a name.

The analytics, naturally, have a less poetic interpretation.

Down 12%.

I try to leave it alone now.

Here are ten things that help.

1. Post it, then go and water something

I used to stay close to my phone after posting.

Twenty minutes. Check.

Make tea. Check again.

If the numbers were a little low, my brain immediately became very busy. Perhaps the title was wrong. Perhaps I chose the wrong first image. What time did I post last Tuesday?

Gardening has given me a wonderfully unsophisticated solution.

I post, then go outside with the hose.

Watering everything takes about twenty minutes. By the time I come back, my hands are wet, there is usually soil on my shoes, and the fate of one Instagram post feels considerably less urgent.

Plants are quite good company in this respect.

You water them today. They feel no obligation to grow three centimetres by dinner simply to reassure you.

2. Touch things less

When I first started planting, I interfered constantly.

I moved pots. Changed soil. Examined leaves. Decided one corner was too shady and moved something into the sun. The following day the sun looked rather aggressive, so back it went.

I suspect I fussed a few plants to death.

I recognise the same impulse online.

The moment the numbers soften, there is a tremendous urge to interfere. Change the style. Post more often. Reconsider the entire account. Something you liked perfectly well yesterday suddenly appears deeply questionable.

Sometimes my hands need a rest before my judgement improves.

I leave things for a few days now.

I may still be wrong, of course.

Constant fiddling has rarely made me more right.

3. Some things simply die

This took me a while.

When I’ve spent time on something, I have a rather childish expectation that the effort should count for something.

The garden has little interest in this arrangement.

A plant looked beautiful when you bought it. You chose its position carefully. You watered it properly. Then one morning you find the leaves hanging down and that is that.

Work can go the same way.

You make something you love and people are mildly interested. You feel certain another piece will do well and it disappears almost unnoticed.

I’ve become less interested in conducting a post-mortem every time.

Sometimes it didn’t grow.

Pull it out. There will be a small patch of bare soil afterwards.

I’ve learnt to tolerate the patch.

4. Be careful with garden centres

Gardening has introduced me to a new form of poor judgement: entering a garden centre for one bag of soil.

I leave with four plants.

Every one of them felt essential at the time.

The internet works on me in much the same way. There is always a new aesthetic, a new tool, a new format everybody suddenly appears to understand.

Spend long enough looking and your own work begins to feel strangely slow.

At that point I tend to go backwards.

Old books help. So do old houses, auction catalogues and magazines that have been sitting on a shelf for years.

I recently opened an interiors book I’ve owned for ages and became stuck on several pages I must have skipped the first time.

Your eye changes.

The book has been patiently waiting.

5. Leave a patch to grow badly

A garden becomes exhausting when every square metre has a plan.

I have one small area I interfere with very little. Things appear. I watch for a while. Some stay.

I think creative work needs a patch like that.

Make something that sits awkwardly with the rest of your work. Write about something you have no expertise in. Take a photograph you like even though it will make a mess of the grid.

And perhaps resist calling it a “new series” by Thursday.

A new interest is quite a fragile thing.

It doesn’t need a content strategy yet.

6. Have an ordinary Tuesday

Some of my favourite moments in the garden happen when I’m there for another reason.

I go outside to take the rubbish out and notice that something which was closed yesterday has opened. I’m collecting washing in the afternoon and find a fig that should probably have been picked two days ago.

None of this was scheduled.

I think people who spend a lot of time making content need ordinary days rather badly.

Go to the supermarket. Change the sheets. Make something unremarkable for lunch.

An entire day can pass without producing a photograph.

I’m increasingly fond of those days.

When life is documented too efficiently, your eye begins to recognise things according to whether they can be posted.

You miss quite a lot that way.

7. Cut the branch that is doing perfectly well

I still hesitate when pruning.

The branch looks healthy. It has grown. Why on earth would I cut it?

Then you begin to understand that healthy growth can still be taking energy in the wrong direction.

The hardest things to remove from creative work are rarely the disasters. I can delete those with enthusiasm.

It’s the things that are doing reasonably well.

The numbers are fine. People like them. You have simply become a little bored.

I ask myself a question now: if I came across this for the first time today, would I still want to make it?

The answer can be mildly embarrassing.

Share

8. Feed the soil for a while

When I started gardening, nearly all my attention was above ground.

Eventually I began looking at the soil.

Sometimes a struggling plant has had plenty of water. The soil itself is exhausted.

During a quiet period online, I try to look at what I’ve been putting into my own head.

If most of it has been Reels, posts and other people’s updates, I can hardly be surprised when my own ideas begin to feel familiar.

A long book usually helps me.

Going outside helps too.

Recently I left my phone indoors and spent nearly two hours in the garden. I came back with no revelation whatsoever.

It was lovely.

9. Let an idea stay underground

Gardening has made me aware of how impatient I am with things I cannot see.

You plant something and immediately want evidence.

Is it growing? Is anything happening down there?

I do this with ideas too.

Some arrive very quietly. I can sense an interest forming, though I have no useful language for it yet. Putting it online too early changes the way I look at it. Other people’s reactions arrive before my own opinion has settled.

Now I keep a few things in Notes.

Two weeks later, I look again.

Some are dreadful. Delete.

Others are still there and have acquired a little more shape.

Those are usually worth following.

Message me

10. When something finally flowers, spare us the comeback announcement

Eleven days away from Instagram rarely requires:

I’M BACK.

You went to the supermarket. It’s all right.

When there is something new, post it.

The explanation can stay private.

A flower eventually opens in the garden without publishing a retrospective on the difficult weeks underground. I’ve come to appreciate the efficiency of this.

People can see the flower.

They don’t need the compost report.

There are several parts of my garden this year that bear very little resemblance to what I planned.

One area is overcrowded. Another is still rather bare. I bought the wrong plants more than once. On several occasions I finished watering everything just before it rained and stood at the window watching half an hour of my life become completely unnecessary.

I quite like the process now.

When social media goes quiet, I interfere with that a little less too.

I still check the analytics when I said I wouldn’t. I still see someone gain thousands of followers overnight and feel a small, unpleasant jolt somewhere in my stomach.

Apparently gardening has not made me wise.

Give it another day

As the late afternoon sun slants across the yard, the drooping tomato plants in their pots are already losing their shapes in the gathering dusk.

I lean my mud-caked trowel against the brick wall and brush a dry clump of grass from the sole of my shoe. The analytics dashboard remains as quiet as stagnant water—no miracles, no notifications.

It doesn’t matter.

Whatever is meant to sprout will summon the strength to push through the hard shell down in the dark; whatever has bloomed will fade in its own time, clearing the branch for the next autumn rain.

Tomorrow morning, I will likely step out onto the balcony with my coffee once more, acting very much like a hopeless amateur, just to check on that patch of soil that still refuses to show any signs of life.

And then, I will wait another day.

Comment

^{Where other people see a house waiting to be saved, I can already see what it will become}

I have always been unreasonably optimistic about ruined houses.

Where other people see flaking plaster, sloping floors and a budget already slipping its leash, I am deciding where the first cup of coffee will go once the place is finished.

This probably has something to do with the way I have lived. Properties belonging to our family business have needed one renovation after another. I never formally decided to study buildings; the buildings simply kept presenting themselves until, by force of circumstance, I became half an expert in old houses.

Fifteenth-century houses are particularly unforgiving.

A drawing may be immaculate, but open one wall and the whole plan can change. Beams, pipework, fire regulations, heritage restrictions, costs and deadlines arrive in the room together. A contractor cancels. A material fails to turn up. What was meant to take a few days stays with you for an entire season.

Old houses are not especially interested in obeying a designer.

It is not enough to think about how beautiful the finished room might be. You have to decide what deserves to remain, what must come out, and whether a new material can live honestly beside the old structure. Many good decisions disappear once the work is complete. A bad one returns for years.

Do this long enough and a kind of private knowledge settles in the body.

I notice the depth of a door lining. I see a floating worktop and think first about how it is supported. However seductive an interior image may be, if the beam is in the wrong place or the services have nowhere to go, I stop believing in it.

None of this appears in my biography. The knowledge lives in dust, rework, sudden surcharges and arguments on site. It also lives in the private relief of seeing a difficult room finally made right.

I have always thought of restoring an old house as the most literal form of manifestation.

First you see a room that does not yet exist. Then you spend years, a great deal of money and still more patience bringing it into the world. My vision board has never lived only on a wall. Sometimes it is a leaking house. Sometimes it is a floor plan revised so often that the paper has gone soft at the folds.

Later, I brought the same instinct to Instagram.

In three months, the account grew from 15,000 followers to nearly 50,000 and the work received more than six million views. One evening alone brought 6,000 new followers. Architects, designers, photographers, brands and established studios began appearing among the people watching.

From the outside, it looked like a sudden arrival.

I knew that long before the numbers began to move, something had been accumulating.

Scarcity makes a handful of departures feel larger than the thousands arriving

^{Late at night, the phone is face down and life is still going on}

It was after eleven. I was sitting in the kitchen in my pyjamas. The bottle had been open since the night before and the wine was no longer particularly good. I poured a little and meant to go upstairs when I had finished it.

My phone was still lit.

That day’s post was doing well. Follows and saves were climbing. Then I noticed that a few people had unfollowed.

Share

The number was tiny. I did not even know who they were, yet I looked at the post again. Was one image too cold? Had I said too much? Had I posted too often?

I used to do this all the time. Hundreds of people might arrive in a day, but my attention would settle on the few who had left. The newcomers became a total before I had even met them. A stranger leaving felt oddly personal.

That night I looked for a few minutes, turned the phone face down, finished my wine and went upstairs.

The next morning, I did not change the work I had already planned to publish.

There was no revelation. I had simply grown tired of auditing myself. If someone no longer liked what was here, one more agreeable image was not going to keep them for long.

I used to explain myself to hostile commenters as well. I would write a reply, decide the tone was not quite gracious enough, and rewrite it. A stranger could take half an evening from me with a few unpleasant words.

Then, one day, I blocked someone.

The page became quiet at once. Nothing else happened.

Since then, the matter has been simple. People who do not like the work may leave; malice gets blocked. Unkindness can still reach me, but it occupies less time. Once the phone is down, there is wine in the kitchen and an actual life upstairs. A stranger does not need to spend the whole evening with me.

Huddling together for warmth can be confidence outsourced

^{The coffee is still hot; the work has acquired a timetable of its own}

When I first began building the account, I loved being in touch with other emerging creators.

We were in roughly the same place. We looked at one another’s posts, left a few words, and talked about the numbers. When somebody’s work suddenly travelled, everyone was excited. When somebody had been posting into silence for weeks, a familiar person would go and sit beside them for a moment.

That companionship was good to me. I do not want to recast it as a calculated exchange. Many of those relationships were sincere and remain so.

But at the time I depended on the response.

After posting, I watched for familiar profile pictures. When they appeared, the work felt received. When they did not, I wondered what had gone wrong.

Then life became busier. There were old houses under construction, emails to answer and the usual procession of unforeseen problems. Once, after publishing something, I put my phone in my bag and left to deal with a building site. It was evening before I remembered that I had not checked the numbers all day.

The post had travelled perfectly well on its own.

As this happened more often, I grew accustomed to leaving after I published. For a few hours, sometimes half a day. If nothing happened at once, I could wait. Some images were quiet on the first day and were still being shared and saved weeks later.

New followers were no longer arriving only from the circle I already knew.

There were architects, photographers, material brands and design practices I had once followed myself. They came from different cities. Most never commented. They looked, disappeared, and returned a few days later.

I could not know where an image might end up. Perhaps it would be opened in a studio; perhaps it would appear on somebody’s morning commute. Once I understood that, I became less impatient.

The familiar people were still there. I simply no longer needed each of them to confirm every piece of work. At last, I had acquired a little patience with what I made.

Image-making grows easier; I grow more restrained

^{As image-making becomes easier, what I choose not to publish matters more}

The account changed when I stopped treating every image as a solitary piece of work.

I used to ask whether this particular image was beautiful enough. Later, I became more interested in whether it belonged beside everything that had come before it.

I can spend a long time looking at very small things. Where, exactly, is the light entering? Is the stone too busy? Could the legs of that chair truly support it? Why does the hand look as though it has been moulded from plastic?

Two versions may look almost identical to somebody else. I know they are not the same.

When images are no longer scarce, what becomes expensive is a personal history - and the discipline to choose.

Subscribe now

AI has given me speed. It has also produced a great deal that must be deleted. Sometimes nine images out of ten are perfectly complete, yet not one of them is right to publish. I often spend longer choosing than generating.

Gradually, kitchens, bathrooms, people, materials, history and the present began to speak to one another. They did not need to look alike. They simply needed to be governed by the same decisions. What I refuse has become as important as what I like.

A distinctive account is not made by repeating the same thing a hundred times. It is more like a person entering different rooms while keeping the same pace, the same eye and the same sense of proportion.

In three months, the following grew from 15,000 to nearly 50,000, with more than six million views. One evening brought 6,000 new followers. After that, some posts would quietly continue to bring in hundreds.

Of course I was pleased. I took screenshots. I refreshed the figures. There is no virtue in pretending not to care when something you have worked on for a long time is finally seen.

The next morning, however, my inbox was still full. Problems on site had not resolved themselves because 6,000 people had arrived. I finished my coffee and returned to the image I had not quite finished the day before.

The growth was real. So was the rest of my life. Both continued.

The names I knew only from the world’s leading magazines appeared together, as if by private arrangement

^{The names I once encountered in magazines were suddenly in my follower list}

The first time I recognised the name of a well-known practice among my followers, I opened the profile to make sure.

For a moment, I wondered whether I had misread it.

Designers and professional accounts I had once collected in reference folders were now looking at my work. Some noticed materials. Some studied the spatial details. Others moved through several groups of images in succession.

At first there were one or two. Then, over several days, they appeared in a cluster, as if they had made a private arrangement. There was no announcement and no noise. The familiar names were simply there.

At some point I realised that the room had changed. Design enthusiasts were still present, but so were people who worked seriously in architecture, interiors, art and branding. My page had become one of the places they came to look for references.

Had this happened earlier, I would probably have become terribly self-conscious. Every post would have been accompanied by the question of what they might think.

When it actually happened, I was calmer than I expected.

I still felt pleased when I saw the names. I allowed myself a moment, then went back to the chair in the image that still was not quite right.

Their presence confirmed that the work had travelled further than before. It did not make me change direction for the sake of one impressive follower. The account grew fastest during the period in which I looked back the least.

A few evenings ago, I was in the kitchen again, in my pyjamas, with a glass of wine.

The phone was beside me. Somebody followed; somebody else unfollowed. I looked once and did not open the list.

The wine was better this time.

If I had to locate the beginning of those three months, it would not be the post that brought 6,000 followers in one evening. Nor would it be an image later saved by an architect whose name I knew.

It began on an ordinary night.

I noticed that somebody had left, and the next morning I made precisely what I had intended to make.

Substack summary

In three months, my Instagram grew from 15,000 followers to nearly 50,000 and received more than six million views. The real catalyst was not a single viral post, but the private knowledge accumulated through years of restoring old houses - and the evening I stopped counting who had left.

First came an Italian design account. A few days later, I noticed an architect from Switzerland. Then came designers and studios from Germany, the United Kingdom, and the United States. Some of them were people I had already been following for years. I had saved their work, returned to it, and studied it long before they knew our account existed.

Seeing those names on our profile felt slightly surreal.

Of course, I was pleased. But I also felt exposed.

Architects and designers rarely look at an interior image and stop at whether it is beautiful. They notice proportion. They look at the way materials meet. They can tell whether a door could actually open, whether a piece of furniture is properly scaled, and whether the light has a believable source.

An ordinary viewer may first notice an impressive slab of marble. A professional may immediately see that the veining is too large for the surface, or that the stone could never have been installed in that way.

So I wondered what they would see in our work.

Would they immediately recognise it as AI? Would they think the images looked impressive on Instagram but fell apart under closer inspection? Would they notice the strange texture on a wall, the overly perfect light, or a spatial decision that would never work in reality?

The criticism I had imagined never came.

They did not question whether the images were made with AI, and they did not make the technology the subject of their comments. They simply responded to the design: the materials, the light, the atmosphere, and the feeling of the spaces.

That changed the way I understood the idea of removing the “AI look.”

It is not about hiding the technology or trying to fool people into believing an image is a photograph. We simply do not want the technology to be noticed before the design.

When someone looks at an image and first thinks about the stone, the light on the wall, or how it might feel to spend time in that room, AI has already done what we needed it to do.

The technology is present. It just does not have to stand in the centre of the picture.

Subscribe

The tool itself is not the secret

The image-generation tool we use and trust most is ChatGPT.

It helped us create a post that brought in more than 6,000 followers overnight. Many of our videos have also brought in more than 1,000 new followers from a single post.

Those numbers can easily create the impression that we have access to an unusual tool or a secret prompt that very few people know about.

We do not.

Anyone can open ChatGPT and type in phrases such as “natural stone,” “smoked oak,” “soft natural light,” or “quiet luxury.” Within seconds, almost anyone can generate an attractive interior.

Making one beautiful image is no longer particularly difficult.

The difficult part is making the next image, and the one after that, and still having the work feel as though it belongs to the same visual world several months later.

A single successful image can involve a degree of luck. A consistent visual identity cannot depend on luck alone.

AI can generate a room for you, but it cannot tell you which room is yours. It does not know whether your idea of calm means using fewer colours, removing furniture, softening the light, or leaving an entire wall empty.

It does not know whether “luxury” means expensive materials, perfect styling, or a quieter kind of order that does not need to announce itself.

Those decisions still belong to the person directing the work.

Step One: Build a master system, but do not worship it

Our prompts are never ready to use from the moment they are written.

At first, they looked more like a growing collection of working notes. They included materials, camera positions, lighting conditions, proportions, finishes, and many words that we later removed.

Some phrases sounded sophisticated but produced images that tried far too hard.

We once revised the same kitchen repeatedly. The first version felt spacious, but the marble looked like an enlarged print pasted onto the cabinetry. In the next version, the stone became quieter, but the entire kitchen suddenly became smaller. Later, we asked only for matte hardware, and the AI also changed the cabinet doors and the lighting.

While working on a powder room, I once asked it to remove a lamp on the left.

The lamp disappeared, but the vanity became wider. The mirror changed shape, and the wall finish no longer matched the previous version. In another image, the mirror reflected timber panelling that did not exist anywhere in the room.

I did not have a profound creative insight at that moment.

I was simply annoyed.

The previous version had been very close.

This happens often when working with AI. It solves the problem you have pointed out while changing five other things you never asked it to touch. You go back and explain again: keep everything else exactly the same and change only this one detail.

Sometimes it understands. Sometimes it does not.

That is why I have never experienced prompting as an effortless shortcut. Much of the work is manual. You generate, compare, return to an earlier version, adjust the language, and try to describe the problem more clearly.

Sometimes the improvement between two versions is almost invisible to anyone else. The wood grain is slightly better scaled. The stone has less contrast. A reflection feels more natural. The room is no longer trying quite so hard to look expensive.

The work can feel repetitive, even physical. You look for a long time. You compare one image against another. You notice that something feels wrong before you can explain what it is.

Over time, we developed what we call a master system.

It includes the camera language, materials, lighting, proportions, and level of styling we often use. It also contains a long record of what we do not want to see.

Once that system began to stabilise, our speed changed dramatically. We no longer had to rebuild the entire visual world for every new image. A kitchen, bathroom, living room, or bedroom could begin from a familiar foundation.

But the master system is not a spell.

The same words can behave very differently in another space. A description of stone that works beautifully in a kitchen may feel too heavy in a small bathroom. Lighting that feels natural in a living room may become theatrical in a narrow corridor.

The system gives us a place to begin. It does not remove the need to look.

When I think about what makes the system valuable, I cannot point to one perfect sentence in a prompt. Most of the useful knowledge is probably hidden in the versions we rejected, the mistakes we had to name, and the decisions we learned to repeat.

Step Two: Learn to explain what you dislike

At the beginning, I often said only one thing:

“It still looks too AI.”

That described my reaction accurately, but it was not useful to the model.

AI does not know what “AI-looking” means.

We had to break that feeling into smaller and more specific problems.

Sometimes it came from dirty, fragmented textures. Along the edges of stone, timber, plaster, and fabric, tiny particles would appear without belonging to any real material. It looked as though several surfaces had been broken apart and blended together.

At other times, there was simply too much visual noise.

Every surface had its own texture. Wood grain, marble veining, wall finishes, fabric, reflections, decorative objects, and lighting effects all competed in the same frame. From a distance, the image looked rich. Once enlarged, it felt restless.

AI has a natural enthusiasm for “more.”

Ask for luxury, and it may add a sculptural light. Tell it that a corner feels empty, and it may place a vase, a branch, or a piece of art there. Ask for natural stone, and it may give you an entire wall of veining that is impossible to ignore.

None of those things is automatically wrong.

The problem begins when they all appear at once.

Real design does not require every surface to perform or every object to be noticed. A material can be beautiful without becoming the subject of the entire room. A wall can remain empty without being unfinished.

A large part of our prompts eventually became a description of what the model should avoid.

No fragmented textures. No dirty digital noise. No materials that look like printed surfaces. No exaggerated marble veining. No oversized wood grain. No impossible reflections. No light arriving from several unrelated directions.

We also avoid rendering every object at exactly the same level of sharpness. Real photography has depth. Some details are clear, while others soften or disappear into shadow.

Then there is a more difficult problem: sometimes the image is simply too perfect.

The furniture sits in exactly the right place. Every section of stone is beautiful. The light falls only on the surfaces that deserve to be illuminated. Nothing is awkward, uneven, or unresolved.

There may be no obvious error, but after a while the image starts to feel less like a room and more like AI’s idea of a perfect interior.

Real spaces are less obedient.

Natural materials contain variation. Light disappears in certain corners. A plaster wall does not have precisely the same texture across its entire surface. Stone does not arrange its most beautiful veining in the centre of the composition for the benefit of the camera.

We are not trying to introduce artificial damage or make the rooms look worn. We simply do not want them to feel as though they have never been touched by reality.

Even now, I sometimes encounter an image that feels wrong before I can explain why.

I leave it overnight and open it again the next morning.

Many images that feel impressive in the evening reveal their problems in daylight. The stone suddenly looks like plastic. The furniture feels slightly out of scale. Or the whole space resembles a type of “AI luxury home” I have already seen on many other accounts.

Sometimes distance is more useful than another round of editing.

Step Three: Choose clarity over complexity

Over the past year, we have created many kitchens, bathrooms, and powder rooms.

The phrases I probably use most often are: remove the decoration, take away the lamp, simplify the mirror, make the stone quieter, reduce the wood grain, and leave the wall empty.

I regularly ask for rugs, vases, decorative branches, oversized pendant lights, and objects with no clear reason to be there to be removed.

AI seems uncomfortable with emptiness.

Whenever part of a room has not been arranged, it tries to fill it. An empty corner becomes a place for a sculpture. A bare wall becomes an invitation for art. A quiet surface suddenly develops texture.

But I often feel that a space needs exactly that area where nothing is happening.

We once revised a powder room through many versions. At first, it had a sculptural lamp, decorative objects, dramatic stone veining, and a complicated mirror. Each time we removed something, the room seemed to lose a little of its obvious “design.”

When we finally reduced it to stone, timber, a simple mirror, and light, the space became quieter.

It was less eager to attract attention, but it felt much more convincing.

I do not believe every interior should be minimal. Some rooms should be rich, colourful, decorative, and even theatrical. Restraint is not the only valid design language.

AI, however, can turn richness into crowding very quickly.

It does not always recognise the moment when enough becomes too much.

That decision still belongs to us.

Sometimes removing one piece of furniture makes the entire image better. At other times, the composition loses balance. Sometimes a bare wall gives the materials room to breathe. In another image, the same wall may feel strangely unfinished.

There is no fixed formula.

You have to keep looking.

For AI, complexity is easy because it can be created through addition. Clarity is more demanding. It requires hierarchy. It requires knowing what the image is really about and allowing the rest of the room to support it.

Sometimes the focus is the relationship between stone and timber. Sometimes it is a strip of light moving across a plaster wall. Sometimes it is simply the distance between a chair and the architecture around it.

When that central idea is clear, the room does not need every object to compete for attention.

The images we are most proud of are rarely the ones containing the most design. They are often the ones that seem comfortable enough not to prove anything.

Step Four: Do not turn other people’s AI images into your shortcut

We rarely feed another creator’s AI image directly into the model and ask it to make something similar.

That does not mean we reject references.

I look at architecture, interiors, photography, materials, and construction details every day. No one develops taste in isolation. We are all shaped by what we have seen.

The more important question is what we choose to learn from it.

I do not want to learn only from the finished surface of another AI image.

Many AI interiors online are becoming increasingly similar: the same low camera angle, the same monumental stone wall, the same sculptural furniture, the same soft light with no clear source, and the same decorative objects placed in exactly the right position.

The images are often beautiful.

The problem is that they can appear to come from one anonymous studio.

When those images are repeatedly fed back into AI as references, the work becomes even more alike. The visual shortcuts and small artificial details are copied along with the style.

We would rather allow our own visual language to develop slowly.

Style is not created by selecting one type of marble or applying the same visual filter. It grows from many small habits: how much empty wall we are comfortable leaving, how we understand the weight of a material, whether we prefer morning or evening light, and the point at which we feel a room has become too much.

These choices need time to become recognisable.

If we continually borrow another creator’s completed result, our own visual language begins to drift. We may inherit someone else’s compositions, material combinations, and sense of atmosphere without even noticing.

We do not want to become a refined copy of somebody else.

We want the work to become recognisable as ours, even before someone sees the account name.

There is another reason we avoid this shortcut.

We want to practise communicating with AI, not only train it to imitate.

Giving the model a reference image and asking it to “make something like this” is a very short conversation. The image has already explained the composition, colours, materials, lighting, and atmosphere.

Without that image, we have to explain why we want a space to feel calm without feeling empty. We have to describe why a rare stone should have presence without becoming the only subject of the room.

We have to explain why an image may already be beautiful and still not belong to our world.

Those questions cannot always be answered with a few adjectives.

Sometimes, while trying to explain them to AI, I realise that I had not fully understood my own preference. I knew I disliked something, but I had never needed to say why.

AI has not made me think less.

Quite often, it forces me to find language for decisions I previously understood only through instinct.

Step Five: Develop AI leadership

I currently use the term “AI leadership” to describe this ability.

It may sound larger than it is. What I mean is practical: can you guide a tool that is extremely capable and fast, but has no fixed point of view, towards the direction you genuinely want?

AI can produce many answers.

Some are beautiful. Some are unexpected. Some move far away from the original idea. It will not stop because a result no longer feels like your work. It will not ask whether a room has already been overdesigned.

Those decisions still have to come from a person.

In practice, our relationship with AI is rarely a matter of making one request and accepting the result. It is usually a repetitive conversation.

The model changes one version, and we point out another problem. The new version fixes the mirror but damages the vanity. It improves the material but changes the proportions of the room. Sometimes every revision becomes better. Sometimes the work moves further away from the original direction.

I have also lost images that were almost finished because of one final, unnecessary change.

That experience did not teach me how to be right every time.

It taught me to recognise when to stop.

When should we continue? When should we return to an earlier version? When has the direction become impossible to recover, making it better to begin again?

AI leadership does not mean controlling every result perfectly. It means maintaining a clear direction while working with a system that can produce almost unlimited possibilities.

It also means deciding where AI is allowed to surprise you and where the boundaries need to remain firm.

Some of its unexpected ideas are useful. A slightly different composition, an unusual material relationship, or a softer quality of light may lead somewhere we had not considered.

But surprise is not the same as direction.

Without judgment, the model’s possibilities can quickly pull the work towards whatever is most visually dramatic, familiar, or immediately attractive.

As models become more capable, this ability will matter more, not less.

Beautiful images will become increasingly common. Speed will no longer be rare. The more output AI can produce, the more valuable it becomes to know what deserves to remain.

Over the past year, we have been teaching AI to understand our visual preferences.

At the same time, it has forced us to examine those preferences more closely.

Why do we like this light? Why do we dislike that stone? Why are we unwilling to continue in a direction that may perform well on social media but no longer feels like us?

Those answers did not appear all at once.

They slowly entered the workflow, one correction at a time.

We are not trying to teach AI how to design in general. We are teaching it to understand how we design.

That, to me, is the essence of AI leadership: directing the technology without allowing it to take control of your taste.

Step Six: Do not keep an image simply because it is beautiful

AI can produce a large number of images very quickly. At first, that speed is exciting. Later, it can become a distraction.

When dozens of attractive images appear in front of you, it is tempting to believe they should all be used.

We now delete many images that are objectively beautiful.

Share

I once created a kitchen that I loved in the evening. I immediately began imagining how I would edit it into a video. The next morning, I opened it again and suddenly realised that the stone looked like a layer of plastic printed onto the cabinetry.

On a small screen, it had enough drama to work.

Once enlarged, it had no weight.

We did not use the image.

Other images have no obvious mistakes. They may even have strong viral potential. But they feel too familiar, as though we have already seen the same room many times elsewhere.

A beautiful image is not automatically useful to a visual identity.

Sometimes an image performs well because it follows a language the audience already recognises. That does not always mean it moves the work forward.

This is one of the least visible parts of the process.

People see the final image or a fifteen-second video. They do not see the versions we abandoned. They do not know how many times we changed one wall, one reflection, one piece of hardware, or one area of wood grain.

Generation is fast.

Choosing is not.

I increasingly believe that style is formed not only by what a creator publishes, but also by what they decide not to publish.

The images we reject draw the boundaries of the work. They help us understand what may be beautiful, effective, or popular, but still does not belong to us.

When people stop discussing AI

Comment

Looking back, I do not think our advantage comes from a secret tool.

We use ChatGPT, and we are completely comfortable saying so. New models and new interfaces will continue to appear. Some of the prompts that work today may become irrelevant very quickly.

Tools will change.

What remains is the way a person looks at space, understands materials, and responds to an image that is already beautiful but still does not feel right.

That judgment does not appear automatically when the software improves.

So I no longer consider “This does not look like AI” the highest compliment our work can receive.

That sentence is still about the technology.

I would rather hear someone talk about the stone, the light on the wall, or the feeling that a room makes them want to sit down and stay for a while.

When people stop asking how an image was made and begin imagining how the space would feel, I know we have reached the result we were looking for.

The technology is still there, of course.

It is simply no longer the most interesting thing in the room.

Instagram creates the first impression. Substack builds the relationship. The most resilient creator businesses begin where the two meet.

Since March, I have been using AI to help manage and shape my Instagram workflow.

My audience grew from 15,000 to 45,000 followers. Strong posts still bring in more than 500 new followers at a time.

But the more those numbers grew, the clearer something became:

Being seen is not the same as being remembered. And having followers is not the same as having a relationship with them.

Instagram can place a piece of work in front of millions of people overnight. It cannot promise that their attention will still belong to you tomorrow.

The algorithm gives creators extraordinary speed. It also introduces permanent uncertainty.

You can build visibility there. You should not build your entire future there.

So I began constructing a second layer:

Instagram makes people stop. Substack gives them a reason to stay.

If I had to describe the relationship between the two, I would think of Instagram as an *ukiyo-e* print displayed in a public gallery: vivid, immediate and designed to capture a passing gaze.

Substack is the private notebook sent to a smaller circle: slower, quieter and meant to be read more than once.

This is not simply cross-platform promotion. It is a way to turn passing attention into an enduring relationship—and aesthetic instinct into intellectual property that compounds over time.

I. Instagram is the public gallery

One of the great misconceptions of the attention economy is that being noticed requires becoming louder.

More text. More stimulation. More information. More exaggerated promises.

But when everyone raises the volume, nobody becomes clearer. We simply produce more noise.

Powerful visual work does not chase the viewer.

It makes the viewer pause.

On Instagram, an image, Reel or carousel does not need to explain your entire world. Its first task is to communicate something that cannot be easily replicated: your judgment.

How do you choose a material? How do you use light? What do you remove from the frame? How do you turn scattered references into a coherent visual language?

The most valuable asset on Instagram is therefore not volume.

It is recognition.

When someone can identify your work before seeing your name—when a palette, composition or atmosphere begins to feel unmistakably yours—a personal brand is no longer merely being promoted.

It is being formed.

The strongest content rarely reveals everything at once. It offers a compelling entrance while leaving one important door open. It moves the viewer beyond *This looks beautiful* and towards a more valuable set of questions:

- Why did she make that choice?

- What system sits behind this image?

- Where can I find the complete idea?

Instagram does not need to provide every answer.

Its role is to create the curiosity that makes a deeper answer desirable.

II. The transition is part of the work

Many creators build significant reach on Instagram but struggle to translate that reach into a deeper audience relationship.

The problem is often not the content. It is the transition.

A lonely “Subscribe to my newsletter” in a bio rarely gives people a meaningful reason to act. Nobody needs another email simply because another newsletter exists.

People need to understand what becomes available when they leave the feed and give you more of their attention.

The invitation should feel like a continuation of the work, not an interruption by an advertisement.

For example:

This series shows the final visual decisions. I wrote a deeper Substack note about the material logic behind them, the alternatives I rejected and the complete system that shaped the result. If you would like to see what exists beneath the surface of the work, we can continue there.

There is no pressure in this invitation. No manufactured urgency.

It simply tells the reader: this is the entrance; the fuller world exists on the other side.

That transition also performs a valuable act of selection.

Some people want a few seconds of visual pleasure. Others are willing to spend ten minutes understanding your method, judgment and point of view.

The first group creates reach.

The second can become a real readership.

A creator does not need to take every passer-by home. You only need to show the people already moved by your work where the next door is.

III. Substack is the private notebook

When someone leaves Instagram, opens your link and shares an email address, the relationship changes.

That person is no longer merely a view distributed by a platform.

They have given you permission to reach them again.

This is where Substack becomes valuable. It should not function as an archive for longer Instagram captions. Its role is different: to reveal the structure of thought behind the visible work.

Instagram presents the result. Substack opens the process.

Instagram shows what you made. Substack explains why you made it that way.

Instagram introduces your aesthetic. Substack reveals your judgment.

The first letter should fulfil the promise that brought the reader there.

Do not spend their initial curiosity on a ceremonial autobiography. Give them the method, framework, resource or insight they were promised. Trust begins when an invitation is honoured.

The second letter should reveal another dimension of your work.

Show not only the polished outcome, but the decisions, failures, rejected directions and independent beliefs beneath it. Visual language creates attraction; intellectual substance determines whether the relationship can continue.

The third letter should open the room.

Ask a question worth answering. Invite readers to share their experience, disagreement or uncertainty.

A living newsletter is not a stage from which one person speaks indefinitely. It is a room in which a distinctive group gradually recognises one another.

Over time, the best newsletters create a rare form of familiarity.

Readers do not simply follow you. They begin to understand how you think—how you read a room, judge a material, interpret a trend or dismantle a business problem.

That familiarity cannot be produced by a single viral post.

It is built through reliable arrival, one letter at a time.

IV. Followers are rented. Relationships are built.

Success on a social platform is, at its core, success on rented land.

You may build an exquisite storefront and attract enormous traffic, but you do not own the street. An algorithm change, an account restriction or the decline of a particular content format can alter your reach without warning.

This is not an argument for abandoning platforms.

It is an argument for using them with clarity.

Platforms are exceptionally good at discovery. Email is better suited to continuity.

An email address is not merely another row in a database. It represents a person who has actively chosen to hear from you again. As long as the work remains worth opening, you no longer need to reapply to the algorithm every day for access to that relationship.

Email is not a perfectly sovereign territory. Deliverability, open rates and platform policies still matter. But compared with a social-media follow, it is more direct, more portable and closer to a relationship the creator can genuinely maintain.

This is why 45,000 Instagram followers may be exciting, while a smaller group of readers who finish your essays, reply to your questions and anticipate your next letter may ultimately be more consequential.

The first gives you visibility.

The second gives you a future.

V. Monetisation is trust taking shape

Creators often treat monetisation as a sales event that appears abruptly after the content.

They share freely for months, then suddenly begin promoting. Readers feel the break in tone, and the creator often feels uncomfortable making the transition.

A more coherent model allows the product to emerge naturally from the body of work.

When readers have discovered your aesthetic through Instagram and understood your method through Substack, they are no longer buying only a PDF, course or guide.

They are buying organised judgment.

They are buying a shorter path through a problem.

They are buying access to a system that has already demonstrated its value through the consistency of your work.

The strongest products are often already hidden inside the newsletter:

- A question readers ask repeatedly can become a guide.

- A widely saved framework can become a course.

- A recurring debate can develop into a research report, private white paper or membership community.

This is also where AI becomes genuinely useful—not as the author of your point of view, but as a system for recognising patterns.

It can help identify recurring questions, reveal which ideas readers save or respond to, and organise a growing archive of work. It can accelerate production.

But it cannot decide what is worth saying.

That remains the creator’s work.

Selling, then, no longer needs to feel like an unexpected act of persuasion. It becomes the process of turning recognised value into a more complete and useful form.

The most resilient creator businesses do not rely on endless urgency. They rely on a quieter form of confidence:

When the work, the ideas and the product all belong to the same world, the offer does not feel like an interruption. It feels like the natural next step.

VI. Instagram creates speed. Substack creates compounding value.

Instagram and Substack should not be treated as two unrelated channels that demand two separate identities.

They are two parts of the same creative architecture.

Instagram lowers the threshold of the first encounter. Through visual language, a stranger can understand your sensibility in seconds.

Substack extends that encounter. It creates space for the thinking, experience and method beneath the aesthetic to unfold.

One creates breadth; the other creates depth.

One generates discovery; the other builds memory.

One belongs to the public square; the other to the private study.

When the two are properly connected, creators no longer have to choose between reach and substance.

Visual work can become an entrance to ideas. Essays can become the roots beneath the images. Short-form content can bring in new readers, while long-form work gives them a reason to stay.

Attention continues to move across the platform.

Your relationships, knowledge and body of work continue to accumulate.

That is the real promise of Instagram and Substack together.

The goal is not to escape the algorithm, but to stop being entirely dependent on it. It is not to reject reach, but to give reach somewhere more meaningful to arrive.

---

From being seen to being remembered

In a world where content is becoming cheaper and images more abundant, production is no longer the rarest advantage.

Judgment is.

So is the ability to choose, to articulate a distinctive point of view and to earn trust over time.

Instagram can make you visible in a crowded room.

Substack can make you memorable to the people who matter.

One is the public print, catching a moment in a fast-moving world. The other is the private notebook, preserving thought in slower time.

For an independent creator, the strongest future is not simply a larger following on one platform. It is an unbroken path:

chance encounter becomes a voluntary subscription.

A subscription becomes a reading habit.

A reading habit becomes trust.

Instagram gives you speed.

Substack gives you compounding value.

The business that truly belongs to you begins where the two meet.

*If this way of thinking about design, AI and creator-led business resonates with you, subscribe. I write about the systems beneath beautiful work—and how aesthetic judgment can become lasting intellectual and commercial value.*

*And if someone in your circle is building an audience on borrowed land, share this essay with them.*

Share

I deleted the archive first. Years of posts — the sediment every account mistakes for equity.

We started this cycle at 10,000 followers. Six months later: 45,000. Roughly half the posts brought in somewhere between 500 and 5,000 followers each. The account runs 700,000 to 800,000 views a month, with peaks above 1.3 million.

What follows is the operating system, not the inspiration. Ten steps, and the exact line between what the machine does and what I do.

Share

Who is actually in the 45,000

The number is the least interesting thing about it.

Among them: principals of studios whose work you have seen without knowing whose it was. Designers who appear in the top magazines with some regularity. A number of Michelin-starred chefs. Several of the top yacht companies. Several world champions. Billionaires. Founders of businesses that genuinely work. And — the group I care most about — creators and small-business owners with real taste.

None of them followed because of a hook. They followed because the room looked like somewhere they already spend time.

This is the part that gets missed. A narrow, consistent, unaccommodating aesthetic does not grow an audience. It sorts one. Broad content collects volume; specific content collects the right people, and it does it slowly, and then all at once. Every decision in the workflow below exists to protect that narrowness — which is also why chasing the follower count would have destroyed the only thing worth having.

My method is the 80/20 rule, inverted

Everything below runs on one principle, and the principle is a reversal.

In thinking, I do 80%. The premise, the argument, the judgment about what is worth making. AI gets the other 20%, and that 20% is brainstorming — widening the field, throwing twenty options at me, taking the opposite position so I can test mine against it. It expands. It does not decide.

In repetition, AI does 80%. The variations, the drafts, the reformats, the fortieth version of a thing I will publish once. All the work that used to eat a week and never made anything better, only finished.

Most people run this exactly backwards. They let the machine think, then labour by hand over the execution — and end up with a competent, well-lit, entirely anonymous account. The output is the giveaway. You can always tell which end of the ratio someone gave away.

The second principle: put the weight at the front

The 80/20 split is about dividing effort. This one is about where in the sequence you spend it.

When something works here, it doesn’t stay a single post — it becomes a series. Same structure, new subject, run until people recognise the format before they read the name. Which means the first post in that series was never one post. It is the mother of every version that comes after it, and all of them inherit whatever was inside it: the structure, the pacing, the restraint, the identity. They inherit the laziness too, if it’s there.

So the first one gets a disproportionate share of me. Hand-worked. Argued over. Rewritten well past the point that looks rational for a single upload. Not because that post needs it — because the twenty descended from it will need it, and they have no way of acquiring it later. A series cannot become more original than its first instance. It can only get further from it.

The more of yourself you put into version one, the cheaper every version after it becomes. Front-load, and the rest of the series is assembly work the machine can largely do. Front-load nothing, and you pay the same effort twenty separate times, and none of it compounds.

Heavy at the start. Light forever after. That is the whole economics of it.

The stack

I don’t run one model. I run several, deliberately, because they disagree — and the disagreement is the useful part.

The workflow: 10 steps, 4 phases

The phases map onto the ratio above. Phase 1 is where I do 80%. Phases 2 and 3 are where the machine does. Phase 4 is where I take it back.

Every step is written the same way: what AI does, what I do, what comes out. If a step has no clear output, it isn’t a step.

Phase 1 — Research and direction

Three steps, two different clocks. Research is monthly — about a day of it. The brief and the pillars are set once a quarter and left alone. Everything downstream inherits all three.

1. Research

• Me: Three sources, three different jobs. YouTube for depth — long-form is where someone actually elaborates a topic instead of compressing it, and I want the full understanding before I have an opinion. Instagram for the market — I go through my own niche and pull the posts that outperformed, especially anything over a million views. Perplexity for sourcing — when I need a fact, a figure or a claim I intend to stand behind, it comes back with where it came from, and I go read that.

• AI: Summarises, cross-reads, holds the notes. It does not tell me what any of it means.

• Output: A running research file. Every entry logs mechanism, never subject. What structure carried it, what was withheld, why the best frame was held back to slide three. Copy a subject and the ceiling is a slightly worse version, slightly late.

2. The brief

• Me: One page, written by hand. Who this is for, what sits behind it, what it will never say.

• AI: Argues with it. Stress-tests the positioning against competitors. Never writes it.

• Output: A document pasted at the top of every session that follows, unchanged, all quarter.

3. Four pillars

• AI: Proposes eight directions off the brief and the research file.

• Me: Cut to four. One test: can this pillar hold twenty posts without repeating itself? If not, it’s a topic, not a pillar.

• Output: Four pillars, fixed for the quarter. I’ve run six. The account becomes a bit of everything and not much of anything.

Phase 2 — Production

Four steps. This is the 80% I don’t touch.

4. The month

• AI: A slot for every publishing day, distributed across the four pillars and weighted toward whatever performed last month. I don’t publish to a quota — the cadence is whatever the pillars can carry without thinning out.

• Me: Move two or three. Approve.

• Output: A full calendar. No slot is ever filled by something that occurred to me that morning.

5. Concepts — the three-model brainstorm

• AI: The same brief goes to ChatGPT, Gemini and Claude at once. Three models, three angles on an identical question. One reaches for structure, one for volume, one for the counter-argument. Twenty concepts between them.

• Me: Keep five. I’m not looking for the best answer — I’m looking for the place where the three of them don’t agree, because that’s usually the only place an unobvious idea lives.

• Output: Five production briefs.

6. Copy

• Me: I handwrite the thinking first. Always. The premise, the position, the thing I actually want to say — on paper, before a model sees a word of it.

• AI: Then I debate it. I hand the handwritten version to Gemini and Claude and make them push back: what’s weak, what’s already been said a thousand times, what I’m assuming. Several rounds. Then they draft against the argument that survived.

• Me, again: Rewrite the opening line of every post myself. Competent captions are the problem — nobody remembers them, and that first line is usually the only sentence anyone finishes.

• Output: Final copy, ready for the audit.

7. Visuals

• AI: Stills in Midjourney and Nano Banana. Carousels in ChatGPT — that one I hand over whole; with a precise enough description it builds the full set, and it’s the only place in this workflow where I’ve stopped intervening. Motion in Veo, Kling or Runway, and only on frames that already earned it.

• Me: Kill most of them. What survives is whatever matches the brief — not whatever took longest.

• Output: Five finished sets.

Phase 3 — The audit

One step, two passes. The step most people skip, and the one that changed my output the most.

8. Audit

• Pass one — the machines. The finished piece goes back out to all four: ChatGPT, Gemini, Claude, Manus. Not “make this better” — each one gets a role. One reads as the target client. One hunts for the weakest sentence and the unearned claim. One checks it against the brief from step 2, line by line. Manus reviews the end product as a finished object rather than a draft. Four readers, four blind spots, and they don’t share mine.

• Pass two — the grid. Nine posts on screen together before anything is queued. A new visitor meets them as one room, not a stream — so they get judged as a room.

• Me: Take what survives both passes. Most of the notes get ignored. The point isn’t to obey four models; it’s that the thing I can no longer see after four hours is obvious to a reader who has never seen it, and I have four of those on demand at two in the morning.

• Output: One batch, scheduled in a single sitting. Nothing goes out on impulse; post on impulse and the strategy slowly turns back into a mood.

Phase 4 — Distribution and learning

Two steps. Where the account either compounds or leaks.

9. Conversation

• AI: ManyChat is an assistant. That is the whole of its job — sort, flag, filter, hand over. It has never closed anything and was never asked to.

• Me: This is the one part of the system that stays labour-intensive on purpose, and I’ve stopped trying to make it cheaper. Every real conversation, personally. Anyone who’s been answered by a bot knows exactly how that lands.

• The rule: the bigger the contract, the closer the contact. A DM opens a conversation. A serious project gets a voice, and then a room. I have never signed anything significant through a chat window, and I’ve stopped expecting to — at that size, people are not buying the work, they are deciding about the person, and no interface does that for you.

• It runs both ways. This isn’t an inbox I empty. I start as many of these as I answer, and I don’t set a frequency for it. Anything you can schedule here isn’t the thing that works.

• Output: Qualified conversations, off-platform — and for the ones that matter, in person.

10. Review

• Me: Three questions, none of them a number. What won, and by what mechanism. What we believed in that lost. What’s worth doing again.

• AI: Pulls the data, proposes the reweighting.

• Output: Rewritten pillars, reweighted calendar — and, when something works, a series built off it as the mother post. Same structure, new subject, run until people recognise the format before they read the name, then a few more times after that. Most people get a hit, congratulate themselves, and go hunting for the next idea. The next idea was the one they just had.

I’m deliberately not putting a number on any of this — not hours a week, not posts a month. The moment a workflow gets expressed as a quota it starts being run for the quota. What matters is the sequence and the division of labour. The hours fall out of that on their own, and they fall a long way.

The three things I never automated

The premise. What the account is for. AI is extraordinary at execution inside a frame and has no opinion whatsoever about which frame.

The first line of every caption. Non-negotiable.

Every conversation that could become a client.

Everything else — every variation, every draft, every scheduling decision, every analysis pass — belongs to the machine, and I’m not sentimental about it.

Not because machines can’t operate there. Because that’s the only part anyone was ever paying for.

Nobody has ever called it out

This is the part I did not expect.

The people in that audience are professional designers, architects and photographers. They look at images for a living. They are the single hardest room to fool — the ones who notice the reflection that can’t happen, the ceiling detail that doesn’t resolve, the light arriving from a window that isn’t there.

Not one of them has ever accused me of hiding behind a machine. Not once, in six months of publishing at volume, has anyone written to say this is AI, as an objection.

I don’t think that’s because it’s undetectable. I think it’s because the question stops being interesting when the answer is both. Most of what I publish genuinely is a mixture, and often I couldn’t cleanly separate it myself. The render is machine. The decision about which render lives is not. The frame is machine. The judgment that the frame is one degree too polite is not. What people are responding to isn’t the origin of the pixels — it’s whether a human eye passed over the thing and refused most of it.

The workflow runs on AI. The result still reads as natural, and that combination is the whole of the differentiation. Automation on its own converges: everyone using the same tools with the same taste arrives at the same well-lit nowhere. What doesn’t converge is what you were willing to throw away.

Two refusals

No chasing tools. The tools change every eleven weeks, and I don’t run after any of them. They serve a standard that was set before they existed and will outlast them — the brief decides, the tool executes, and I drop any of them the week something better arrives. A practice built on tool fluency is built on a surface being resurfaced beneath you. The tool works for me. I have never worked for a tool.

No chasing followers. The number is a byproduct, and not one of these 45,000 was bought. The moment the number becomes the objective, the system bends to serve it, and the account dies from the inside while the chart still points up. Gravity, not force.

What this does not solve

The steady state is 700,000 to 800,000 views a month. Some months clear 1.3 million; some months don’t come close. Volume is weather — it answers to format shifts, to whatever the platform is quietly promoting this quarter, to luck. The workflow makes the floor higher. It does not make the sky stop moving.

And 45,000 people watching a room is not one person commissioning one. That gap doesn’t close with better images or a tighter workflow. It closes by having something real on the other side of the attention — a different discipline entirely, and mostly not one AI can help with.

Message me

The house

I’m restoring a building from 1437, in Bavaria, into a hotel that doesn’t exist yet. Six centuries of settling, walls out of plumb, brought back with the machine doing every part a machine can do — and the hand doing the last course.

Wabi-sabi holds that the seam should stay visible. The repair isn’t hidden. It’s the most honest part of the object: proof that something was worth keeping.

That’s the whole method. The machine hauls the stone. The hand sets the final course, and leaves the seam where you can see it.

Taste stays human. AI does the heavy lifting.

Comment

Subscribe

Antidote to Algorithmic Perfection: Living in a hyper-rational world driven by code, metrics, and relentless optimization, tech leaders face severe cognitive fatigue. Wabi-Sabi offers a healing release through its embrace of asymmetry, roughness, and unscripted natural flaws.

The Evolution of Quiet Luxury: The ultra-wealthy have largely moved away from loud, ostentatious displays of wealth toward subtle sophistication. Wabi-Sabi appears austere and minimalist, yet it relies on astronomically expensive custom craftsmanship, rare natural materials (like fumed oak, lime plaster, and hand-forged iron), and one-of-a-kind artisan pieces—a ultimate form of understated status.

Digital Detox and Spatial Silence: In an era of acute attention scarcity, physical spaces heavy with visual noise cause chronic stress. Wabi-Sabi champions the power of negative space, neutral earth tones, and pure light-and-shadow dynamics, turning a residence into a sanctuary for deep mental reset.

Philosophical Resonance with Time: Tech and business titans intimately understand the volatility and rapid obsolescence of the modern world. Wabi-Sabi honors aging, patina, and the traces of time, providing a grounding perspective that gives wealth a sense of history, permanence, and authentic weight.

A New Cultural Capital: As traditional religions exert less influence, eastern Zen philosophies fill the spiritual vacuum. Appreciating Wabi-Sabi functions as an exclusive cultural currency and a subtle code of distinction among the global elite.

In the tea rooms of sixteenth-century Japan, a quiet coup took place in the history of taste.

The supreme luxury of the age was porcelain shipped from China: official-kiln ware, flawless in glaze, immaculate in form, worth small fortunes. Yet the tea masters, led by Sen no Rikyū, set it aside and reached instead for bowls from humble local and Korean kilns—warped in the firing, crazed with fine cracks, mottled where the glaze had misbehaved. A rough Korean rice bowl of the type called Ido came to outrank entire shiploads of imperial porcelain; the most celebrated examples are now treasured as cultural masterpieces in Japan.

At the time, the choice bordered on insult. Its logic, though, was calm: a perfect thing seems to contain no time. It does not age, does not wear, remembers nothing. Inside the warped bowl, by contrast, remains the instant the kiln fire slipped—and the craftsman’s generosity in letting that instant stand. It is an object in which something happened.

I. The Material of Time

The root of this aesthetic is Zen. For centuries, monks had taught impermanence. Everyone knew the words; the words rarely reached the body. The tea masters tried another method: they stopped explaining and handed you a cracked bowl.

Impermanence, in the sutras, is a concept. On the bowl, it is the crazing of the glaze—impermanence you can hold. That is wabi-sabi: one tradition speaking, the other showing, both saying the same thing. The place where time enters an object is the place where it begins to be beautiful.

Pushed to its conclusion, the idea became a craft in its own right: kintsugi. A broken bowl is not discarded. The shards are joined with lacquer, and the seams dusted with gold. The damage is not concealed but gilded. The repaired bowl becomes more precious because it has acquired a history. It broke, and it was cherished.

Four hundred years later, this aesthetic has met the interlocutor it could never have imagined.

AI is the first intelligence we have built that appears to contain no time. It does not age; it does not wear. When it errs, it does not repair—it regenerates. The failed answer disappears, and seconds later a new one stands in its place, pristine, as though the error had never occurred.

It carries no scars because it never has to heal. It is permanently new, and therefore permanently without a history. A thing that cannot break is a thing beyond the reach of kintsugi.

It can discuss wabi-sabi, and discuss it well. It has read the literature of crazed glazes and can explain them as kiln chemistry or Zen history, often more fluently than most of us can. But between the machine and the thing it describes lies the one raw material it cannot generate: years.

Ask it for a crack, and it will produce a perfect one—well placed, plausibly veined. Pixel for pixel, that crack may be almost indistinguishable from the real crack in the wall of an old house.

The difference is singular and absolute: the real crack has centuries inside it. The settling of foundations, the contraction of one hard winter, the tolerance of generations who saw it and let it be. It is something that happened. Inside the generated crack, nothing happened at all. It merely looks as though it did.

Here wabi-sabi reveals its severity. It was never a style. A style can be imitated; wabi-sabi cannot—because its raw material is not form, but time.

II. When Perfection Became Cheap

What, then, does the arrival of AI mean for a four-hundred-year-old aesthetic?

The answer runs against instinct: it means its hour has finally come.

The reasoning is the plainest economics there is: scarcity sets value. In the sixteenth century, perfection was scarce. A flawless bowl could consume half a craftsman’s working life. Perfection was therefore expensive; the world pursued it, and that is precisely why Rikyū’s choice could scandalize.

But perfection has now entered the age of infinite supply. Perfect images arrive by the hundred; polished prose in whatever quantity anyone cares to request. Perfection was a business humanity ran for millennia, and AI has turned it into tap water within a few years.

For the problem of too much, Rikyū left a demonstration. The story comes down through the tea chronicles, told and retold for centuries.

Toyotomi Hideyoshi, hearing that the morning glories in Rikyū’s garden were in magnificent bloom, announced a visit. At dawn, Rikyū had every flower in the garden cut down. Hideyoshi arrived, found nothing, and entered the tea room in rising anger. There, in the dim alcove, stood a single stem: the one surviving bloom. It is said to have been the most magnificent morning glory he ever saw.

Faced with the infinite, Rikyū’s answer was to delete down to one.

Four centuries on, that answer lands on us unaltered: when a machine supplies a hundred perfections a second, the human task shifts from acquiring to selecting—and, more often, to deleting.

The scale of scarcity tips, without anyone announcing it, the other way. When perfection is available to everyone, preciousness migrates toward what cannot be generated: real years, real wear, a crack that has actually stood in a wall for five centuries, a bowl worn smooth by three generations of hands.

The luxury industry sensed this direction long ago. Archive, vintage and patina are all, at bottom, ways of pricing time. AI has merely driven the logic to its conclusion. It has not killed wabi-sabi; it has made synthetic perfection cheap, and in doing so has cleared the field for the real.

The same revaluation applies to people. Our ledgers—the ledgers of creatures made of flesh—are full of assets the machine can never hold: every failure repaired and carried forward, every ending slowly digested, every wrong decision owned and walked with.

Lacquer, all of it, and gold dust.

Wabi-sabi appraises people and bowls with the same eye: what is precious was never flawlessness. It is the trace of mending.

The machine may already surpass us in many forms of intelligence. But we are older than it is. And on the ledger of wabi-sabi, age is an asset.

III. A Wall at Maison 1437

As for how this philosophy behaves in practice, one example comes from my own building site.

In the renovation of a fifteenth-century house, there is a step I have found almost impossible to explain to visitors: when a freshly plastered wall comes out too mechanically even, the plasterer is asked to return to it—not to make it worse, but to restore the small irregularities of the hand.

The wall must be sound, but it need not pretend that time will never touch it.

AI has carried a good share of this project—the generating, the simulating, the testing of possibilities—and carried it fast. But the final judgment on every wall is made to reserve space for the one ingredient it cannot supply.

In the evenings, on site, the new lime plaster meets the five-hundred-year-old stone, and the seam is still raw. No matter. Seams age.

In fifty years it will have softened; in a hundred, no one may be able to tell where the old work ends and the new begins. Only then will the part of this house that passed through my hands be truly finished.

So this was never really a renovation. It is a seam of gold lacquer laid across a five-hundred-year-old bowl.

The most extravagant ingredient in that procedure cannot be manufactured—not by the machine, and not by me.

It can only be waited for.

Fortunately, waiting is a course this house has already taught me.

And it does not, in any meaningful sense, exist yet.

Thirty-seven posts, fifty thousand people, one hotel that isn’t there — to explain how this happened, I have to start with a word that German doesn’t have.

German has no word for yuánfèn.

I have been quietly grieving this for over twenty years. Germans have Schicksal — fate — but Schicksal is something that falls on your head, like weather, like a tax notice, like war. Yuánfèn is different. Yuánfèn is when two things that have nothing to do with each other, separated by mountains and oceans and sometimes by centuries, each take the long way around and arrive, at last, at the same door. In Chinese we grant this word to strangers who end up sharing a table at a crowded restaurant. Twenty years into the German language, I can confirm: they truly do not have it.

So when I try to explain how a woman born in China came to be the keeper of a house built in 1437 in Bavaria, waiting on a construction permit that would not come, waiting so long that what finally arrived instead was artificial intelligence — I find that one of my languages cannot produce the word, and the other needs only the one.

When this house was finished, Gutenberg had not yet printed his first Bible, and “Germany,” as a country, would go on wandering outside the map for another four hundred years. Since then the house has seen a few things: soldiers of the Thirty Years’ War burned down half the town nearby, Napoleon came and went, three empires rose and dissolved, two currencies died in its lifetime — and every generation was certain it was living at the end of history, until the next generation corrected it, gently.

The walls offer no opinions. Bavarian walls possess a patience that can only be called Catholic.

Meanwhile, on the other side of the world and five hundred years later, a small girl was listening to the grown-ups talk. Hurry slowly, the grown-ups said. And: what is yours will find you, no matter how far it has to travel. The girl didn’t understand. The words hung over the dinner table like steam — warm, impossible to hold.

She grew up and moved to the German-speaking world, and stayed for more than twenty years, long enough that she sometimes dreams in Konjunktiv II. She studied German literature in Germany and business in Switzerland — literature taught her to read slowly, business school taught her to count quickly, and she had no idea, at the time, that both would one day be needed inside the same house. She assumed she had left the dinner-table steam behind on the other hemisphere.

Until she met this house.

Or, to put it in the other grammar: until this house, having taken 589 years to make the trip, came looking for her.

The plan was a respectable one: break ground, document everything, turn the whole renovation into a public restoration diary. The camera was ready. The account was ready. Five centuries of stories were ready to be told.

The Bavarian administration replied with admirable economy: Received. Please wait.

Warten. The German word for waiting is a cold word — a waiting-room word, the word of a machine that spits out a numbered ticket. But the Chinese character for waiting, 等, is built from two smaller pictures: bamboo, above a temple. Some ancient hand, inventing the writing system, chose to house the act of waiting inside a temple in a bamboo grove — as if to warn us, several thousand years in advance, that waiting was always meant to be a practice.

I have done both kinds. At first I waited in German: clutching the ticket, glaring at the number board, furious at every letter announcing that one more document was still missing — one more document is always still missing. Then, at some point I cannot name, my waiting changed languages. The site sat silent as a church on Sunday; the scaffolding stood there with such gravity that it began to seem like a protected monument itself. History offers many models for people trapped in a long wait: monks copied scripture, nobles went hunting, Thomas Mann wrote seven hundred pages.

I opened an AI.

What happened next, in medieval language, can only be called a vision: the house did not move one stone in reality — and quietly finished itself in another dimension.

The breakfast room at first light. A blade of sun crossing the stone wall. A chair at the end of a corridor, with no one in it yet. Room by room, I imagined the finished house, generated it, and published it. This is where the thirty-seven posts and the fifty thousand people came from: the hotel they gathered to watch does not exist, and it drew more devotion than many hotels that indisputably do. A Bavarian reader will smile here and think of the Gospel: before the Word became flesh, there was the Word. In the language I grew up with, it is said more plainly: what comes first is always the longing.

And the image-making itself turned out, unexpectedly, to be an old craft. The machine produces pictures a hundred at a time, most of them mediocre in the most sincere way, and choosing the three that can stand still requires an old-fashioned pair of eyes — eyes trained by years of looking at stone, at timber, at light. The speed is new; the judgment is old. A medieval workshop master would understand the arrangement at a glance: the apprentice does the fast work, the master answers for what is right.

As for the delay — looking back, it hardly deserves the name. Had everything gone according to plan, this would be one more house finished on schedule, and if there is one thing Bavaria does not lack, it is that. It was precisely the waiting that reversed the natural order of things: the story arrived before the house, the drawings before the masonry, fifty thousand witnesses before the first guest ever touches the door. A 589-year-old building ended up waiting for a technology not yet six, which now rehearses the rest of its life. On paper it is a terrible match. But old houses have always been more open-minded about newcomers than people are. This one admitted glass, admitted electricity, admitted central heating — and each time, someone declared with total confidence that the walls would lose their dignity. The walls’ dignity is intact. The walls have learned that the average lifespan of an objection is about thirty years, and they themselves intend to stand for another six hundred.

In German, you would call all this a fortunate coincidence. In my mother tongue, two syllables are enough.

The day the permit finally came through, I was so calm it surprised me. Three years earlier I would have opened a bottle, posted a grid of nine photos, and personally notified everyone who had ever said soon, soon. That day I simply stood in the courtyard, looked at the wall for a while, and thought: good — then let us begin.

Only later did I understand that this calm was not maturity. It was what three years of waiting had left in the body. The Japanese found a name for it long ago: wabi-sabi — the beauty of the imperfect, the impermanent, the incomplete. A tea bowl with a crack in it. A wall gone mottled with age. An object that has been used so long it has become more itself. The word sounds Eastern, but the condition it names is everywhere in Bavaria: the old women here have never heard of wabi-sabi, yet they have spent whole lives using their grandmothers’ cupboards and sitting on church pews polished dark by centuries of sitting. Wabi-sabi was never a style. It is the state that everything arrives at, if it is allowed to live long enough.

It is also, it turns out, the state that a person arrives at, if she is made to wait long enough. Imperfection: one document is always missing, the way an old wall always has its crack — that is simply the true face of things. Impermanence: the schedule changed and changed again, and I made my way from “a delay is a catastrophe” to “a delay is weather” — you do not lose your temper at rain; you remember an umbrella. Incompleteness, the hardest course of the three: a house standing in scaffolding and a house cutting its ribbon hold equal dignity, and the first is arguably more honest, because it has not yet begun to perform. Hurry, in the end, is a form of distrust toward time. And a person keeping a 589-year-old house — a person raised on what is yours will find you, no matter how far it has to travel — has no standing whatsoever to distrust time.

So this will likely become the most wabi-sabi boutique hotel apartment in Bavaria — not the kind achieved with a coat of lime plaster and a rustic ceramic bowl, but the kind that is wabi-sabi in its very origins: an imperfect plan, an impermanent schedule, one long incompleteness unfolding first in AI and then in stone. The cracks will stay, because they remember more years than any decoration could. The plaster will show its age, because at 589, pretending to be young is the only true way to look dated. And everything new that enters this house — the artificial intelligence included — will first have to learn the house’s rate of speech: slow, but never stopping.

In an age when everything competes to be newer and faster, the old and the slow have become the last luxuries.

The renovation has begun. This time, at last, there is real dust to photograph.

I think sometimes of the mason of 1437, and what he would make of all this — an Asian woman, a machine that paints, fifty thousand people watching his wall through small glowing rectangles. He would understand none of it. But if I told him, in the slowest speech I could manage, that this house waited 589 years, that it waited until we arrived, and that all of us intend to treat it well — I believe he would nod. Between people who work with their hands, that sentence translates into every language.

Bavaria produces two things: beer, and patience. Three years in, I am pleasantly drunk on the second.

As for you — you, reading this on a screen somewhere. We have never met. There are languages between us, time zones, possibly half the planet. These words took a long road to reach you, and if they happened to land on you, and happened, even slightly, to move you —

German still has no word for that.

Neither, it occurs to me, does English.

But you and I both know what to call it.

If you’d like to stay: this is where I’ll keep writing — about AI and Eastern philosophy and where the two keep meeting, about architecture, about design, about waiting and craft, about the questions that get more interesting to me the longer I turn them over, which is daily.

The house is being built slowly. So are the words