The question of the evening was the psychology of waiting, borrowed from an essay by David H. Maister. We started with the familiar ones: the grocery line, the delayed train and the clinic where nobody will tell you how much longer.

A wait can be short and unbearable, or long and strangely fine. What we call a bad wait is rarely about the clock. It’s the gap between what we expected and what we got. Both live partly in the mind.

Both, it turns out, can be managed.

But going through my notes afterward, I kept circling back to the waits with no display board. Waiting for an answer. Waiting for pain to lift. Waiting for a decision that isn’t ready to be made yet. No one behind a counter to tell you that you haven’t been forgotten.

Someone in the room said the phone has quietly ended a certain kind of waiting: the idle, unstructured, faintly boring kind. The kind where you used to study the floor tiles, or think about nothing. Now the empty minute barely appears before we’ve filled it.

I’ve been leaning on my phone for those small hits more than I’d like to admit. In my own season of waiting, the unoccupied minute can feel unbearable. The phone doesn’t resolve anything. It just gives me somewhere else to look.

Then the room turned, the way a room turns when it touches something true. Not: how do we make people wait more comfortably. But: should we be eliminating waiting at all. Is the empty minute a defect, or a small, unglamorous good we keep confiscating from ourselves in the name of service and relief?

That was the real conversation, for me.

Then we built seven pitches. The ideas were good. Some shortened the wait. Some softened it. Some explained it. Some helped a person escape it.

Not one of them was the thing we’d just spent an hour admiring.

Nobody made something to help a person stay inside the empty minute. We praised boredom and built our way around it.

The easy version of this note is a scold. That we didn’t follow our own thinking. That our ideas were too practical, too eager for the exit.

That isn’t what I mean.

The practical ideas were the honest ones. Everyone in that room is an expert in their own waits, the clinic, the platform, the message that never gets answered. What surfaced first was what had actually cost them something. Of course the pitches ran toward relief. That isn’t a failure of imagination. It might be the most human thing.

We can admire discomfort in theory and still reach for our phones before the elevator arrives. We can defend boredom as a social good and still organize our lives around avoiding it.

I don’t think the gap needs closing.

Some waits are needless. Some are careless, built by systems that treat people’s time as though it belongs to no one. Those are worth fixing. But not every wait is a design problem. Sometimes it’s a transition, a holding space between the thing that no longer works and the thing that hasn’t arrived yet. There’s nothing to optimize because there’s nothing yet to know.

The future stays open. That’s frightening, when you can’t see what’s coming and when. It’s also the only source of possibility there is.

I’m trying to hold both. The part of me that wants relief. The part that suspects the empty minute has something to offer. To notice when I’m making a wait gentler, and when I’m just refusing to be present for it.

Maybe the work isn’t solving waiting. Maybe it’s getting more precise about which waits deserve fixing and which ones are asking something else of you.

Attention.

Or patience.

Or the nerve to let the future stay open one minute longer.

Leave a comment

It’s called the Opportunity Brief Agent, and the premise is simple: when a public-sector team says “we should use AI for X,” the tool refuses to answer right away. It separates what’s known from what’s assumed from what’s missing. Surfacing the risks before it offers a recommendation. Only then does it produce anything that looks like a brief.

Most AI tools work the other way. You type, you wait a few seconds, you get a clean answer in confident language. Speed and fluency are usually the selling points because fast, fluent answers don’t ask the user to think. They ask the user to approve and once the answer is on the page, approval is mostly inertia.

I designed the opposite of that on purpose for this project. A system that withholds. That makes you sit with the question a moment longer. That shows its reasoning before it hardens into a decision. The friction is the feature.

Decision support tools

We call these tools decision support. The human stays in charge, the AI just assists.

In practice, something stranger is happening. The tools I’ve been studying don’t really support decisions. They make them, and then they ask a human to sign off. An eligibility model decides who qualifies for a benefit. A procurement tool scores vendor risk. A planning system ranks capital projects. A generative tool drafts a council briefing in seconds.

There’s a button somewhere that says approve. The human is technically in the loop, but by the time anyone sees the output, the answer has already arrived. It’s sitting on the page in clean formatting with confident sentences. The human’s job is no longer to think. It’s to check.

Checking is not thinking. They use different muscles.

In the public sector the workload is heavy and the stakes are high and the deadlines never stop. So we can assume that the review can become a formality. The human stays accountable. The machine shapes the judgment. And yet, the responsibility and the authority drift apart.

What I noticed designing the OBA is how much of this happens before any decision gets made. It happens in the interface.

A confidence score makes the answer feel official. A ranked list tells you which option matters most before you’ve decided what matters. A summary smooths out everything that didn’t fit. An approve/reject button turns a complicated question into a click. An audit trail looks like accountability without doing the work of it.

As you can imagine, the more polished the output, the less anyone questions it. Fluency reads as intelligence. Speed reads as competence. Confidence reads as truth. None of those things are actually true, but the interface doesn’t care. By the time a human looks at the screen, the work of persuasion is already done.

Human in the loop gets used as proof that the system is responsible. But a human can be in the loop and not really be there.

It’s not that people are careless. Public-sector workers I’ve talked to are some of the most conscientious people I’ve met. The problem is that they’re drowning, and AI looks like a life raft. The tools reduce the load. That’s the whole pitch.

What I’m starting to suspect is that the same tools also reduce the engagement. The load and the thinking turn out to be harder to separate than we wanted them to be. Once you’ve taken away the friction, you’ve taken away the part where judgment happens.

Slowing a system down has a cost

Users may not want a tool that asks them to think before it gives them an answer. The whole reason they came to AI in the first place was to stop having to do that or reduce the amount.

But I keep coming back to the same thing: the interfaces we’re building right now are teaching people a particular relationship to their own judgment. Most of them are teaching deference. The question I haven’t answered yet, the one the OBA is really a test of, is whether a system can teach the opposite or whether the user has to bring that wanting with them.

That is the distinction the last two years of AI discourse keeps missing. We have talked about production the whole time. How fast we can make an image, draft a campaign, generate a layout, cut a video. And on production, the tools deliver. They reduce friction, they collapse the distance between idea and draft. The blank page stops being frightening because the first draft is now free (relatively)!

The hard part is knowing what is worth making, and what should be cut, and what is seductive but empty, and what is polished but already dead by the time it reaches anyone. AI can generate more options than any designer could produce by hand in a week. More options is not more direction. Often it is just a larger pile to sort.

Abundance has a shadow

When it becomes easy to make anything, it becomes easy to make things with no reason to exist. The internet is already full of them. Cinematic lighting attached to no point of view. Gradients arranged over no product logic. Visual systems that borrow the surface language of taste without earning the thing underneath.

This is the trap. AI can simulate the appearance of creative confidence. Restraint, polish, editorial mood, all the signals of taste, reproduced without the point of view that used to produce them. And a point of view is the whole difference between direction and decoration.

You can watch the industry correcting for this in real time. After the loud promises, that AI would replace agencies and personalise everything and make creative production effortless, the conversation has sobered into something simpler. Creativity still matters. Not creativity as an inspirational noun. Creativity as judgement under constraint.

Strong work has never been valuable because it exists

Strong work is valuable because it makes a decision. It chooses a frame, a tension, a way of seeing, and it excludes far more than it keeps. That is what the weak version of this argument gets wrong. It treats creativity as a volume problem. More ideas, more variants, more speed. But the good thing is almost never the result of more. It is the result of less. Less noise, less trend-chasing, less fear of leaving something out.

The tools are pattern machines, and this is exactly why they are dangerous. Ask for luxury minimal branding and you get something that looks like every other luxury minimal brand. Ask for cinematic editorial photography and you get a beautiful image that feels like a thousand beautiful images. The risk was never ugliness. The risk is competent sameness.

Bad taste in the AI era will not always look bad. It will look polished. It will look current. It might get likes. But it has no pressure inside it. No friction, no weirdness, no human stake. And design in particular is not first-glance work. A generated interface can look clean and still fail as a product. A generated brand can look refined and still be strategically empty. The tool does not remove the need for taste. It exposes the absence of it faster.

Taste is not preference

People talk about it as though it means liking beige rooms and obscure films and quiet typography. But taste is not what you like. It is how you choose. It is looking at ten directions and knowing which one has life in it. It is sensing when an idea is almost good and not sharp enough, when a visual is compensating for a weak concept, when something attractive has to go because it does not serve the work.

Which is why taste is uncomfortable. It requires refusal. AI gives you more to say yes to. Taste is the thing that says no, and that no is not negativity. It is the pressure that gives the work its shape. Without it the output just expands. More images, more copy, more layouts, none of it stronger.

The creative person’s job is to interrupt the abundance. To keep asking what the idea actually is, where the tension lives, what is merely beautiful, what is being repeated from culture without being examined, what should be removed. Those questions are not adjacent to the work. They are the work.

Changing roles

So the role moves. Less maker, more editor. Craft does not disappear, it relocates: out of asset generation and into briefing, selecting, sequencing, rejecting, combining, judging.

The studio Evil Martians wrote about this recently, and their version is more honest than most because they were trying to automate their own work away and failed at it in public. They run a large devtools blog, every article carrying a custom illustration of their mascot, and once the image models got good they assumed they could one-shot the whole pipeline. They could get 90% of the way there. The last 10% broke them. The mascot drifted between generations. Small details fell apart on a second look. Each illustration ended up costing six or seven hundred dollars in their own time, cleaning up slop at seven on a Friday. What they landed on instead is what they call the last human mile: a person owns the visual thesis at the start, the model does the bulk, and a human illustrator owns the final pixels. Their own line is the one worth keeping. None of this is an argument against AI in the workflow. It is an argument against AI as the entire workflow.

That is the shape of it. Knowing when to use the tool for breadth and when to stop. Knowing when a result is good enough to provoke a thought but not good enough to ship. Knowing when the thing became more polished and less true.

This is where weak direction gets exposed. A generic brief produces generic work faster. A weak strategy gets decorated instead of fixed. An underdeveloped eye receives an endless supply of plausible options and no reason to choose one. The tool will not rescue the thinking. It amplifies whatever thinking is already there, which is both the opportunity and the threat. A team can generate three hundred options and still never ask what they are trying to say.

A job search makes this feeling easy to name. You send the application. You rewrite the sentence. You adjust the resume. You try to sound confident without sounding fake. You wait. You refresh. You hear nothing. Or you hear no. Or you hear almost, but not quite.

After a while, it starts to feel less like looking for work. You are standing outside a locked door, trying to convince yourself there is still a room on the other side.

The hardest part is not always the rejection. Sometimes it is the lack of visible progress. Nothing looks different from the outside. You are still in the same place, still explaining yourself, still trying, still hoping the next attempt will be the one that finally lands.

And because nothing has changed externally, it is easy to assume nothing has changed at all.

But that is rarely true. Dear reader, this is what I try to remind myself.

You are learning how to ask more clearly. You are learning what you actually want, not just what sounds impressive. You are learning how to recover faster from disappointment. You are learning which versions of yourself feel honest and which ones are just performance. You are learning how to keep showing up without turning every no into evidence against your whole future.

That kind of progress is hard to measure. It does not always arrive with a title, an offer, a clean announcement, or a neat little update. Sometimes it is just the fact that you are still here, less naive maybe, and not closed off.

There is a strange dignity in continuing before anything has worked.

Not in a motivational-poster way. Not because failure is secretly beautiful or because everything happens for a reason. Some things are just hard. Some seasons take too long. Some opportunities do not come when you think they should.

But still, something is being built in the repetition.

Every attempt teaches you where you are stronger than you thought. Every disappointment clarifies what you are not willing to give up. Every time you start again, you prove that the wanting has not disappeared just because the world has not rewarded it yet.

Keep failing as much as you need to.

Not recklessly. Not without reflection. But without deciding too early that the closed door is the final answer.

At some point, the sky opens up. Maybe not dramatically. Maybe not all at once. Maybe it is just one email, one conversation, one person who sees what others missed. But when it happens, it will not have come from nowhere.

It will have come from all the invisible progress you almost forgot to count.

The easy read is that this is a simply a celebrity endorsement. I think it is closer to a confession. It tells you what Meta now believes the hard problem is, and the hard problem is not intelligence because the models are already good enough. The hard problem is getting something onto your face that you do not want to take off.

Every interface that has mattered has worked by disappearing

The computer filled a room, then a desk, then a pocket. Each move made it less of an event. You stopped going to the machine. The machine came with you, and then it stopped announcing itself at all. The smartphone did not win because it folded a camera and a map and a library into one slab. It won because pulling it out in a restaurant stopped being rude. It became part of the furniture.

Don Norman has a line that designers repeat until it goes soft in the mouth: good design is invisible. We say it as a compliment, but it is also closer to a description of how attention works. A door handle you have to think about is a door handle that has failed. The chair holding you up right now is doing its whole job by staying out of your awareness. You only meet the object when it breaks.

Heidegger got there first, and more strangely. He watched a carpenter use a hammer and noticed that the hammer, in use, is not really there. The carpenter does not perceive a hammer. He perceives the nail, the wood, the work. The tool withdraws. It becomes ready-to-hand, an extension of intention so complete that it stops being a separate thing you are holding. The hammer only reappears as an object the moment it snaps, and suddenly you are standing there holding a useless piece of wood, thrown back into noticing.

This is the bar AI is now trying to clear. Not to be impressive. To withdraw.

Which is why Meta is not competing with Apple. It is competing with Ray-Ban

That sentence sounds wrong until you look at what they are actually selling. The marketing is not about processors or field of view or how many photos the things can take. It is about colourways. Collaborations. How the frames sit on a face that people already want to look like. The spec sheet has gone quiet. The question has changed underneath the product.

It used to be: can we put AI into a pair of glasses. That was a question for engineers, and the engineers answered it. The question now is whether anyone will wear them when no one is being paid to. That one does not belong to engineering at all. It belongs to design, and to something older than design, which is the plain fact that people put objects on their bodies to say who they are.

Kylie Jenner was not hired for her grasp of machine learning. She was hired because she understands a thing engineers are trained to find irrational: nobody chooses glasses from a list of features. They choose the pair that makes them feel like the person they are trying to be. This has always been true of watches and rings and the bags people carry. Utility gets you considered. Desire gets you worn.

And here is where my own argument starts to come apart in my hands

Because I have been talking about the disappearing interface as if disappearance were the whole problem, as if the company that makes the glasses vanish most completely wins. But there is a thing already on the table that disappeared years ago, and it is the phone. The smartphone is not an immature technology waiting for its successor. It is arguably the most successful consumer object ever made. It already withdrew. It already became furniture. Replacing something that works that well is a much harder act than inventing it was.

And glasses do not fit everyone the way the phone does. Not everyone wears them. Not everyone wants a thing on their face from morning to night. A phone slips into nearly any pocket and nearly any life. A wearable never will, which quietly caps how many people the product can ever reach. You are not handing someone a better version of a habit they already have. You are asking them to charge another device, carry another device, maintain another device, and wear it, and the payoff has to be enormous to justify all of that. For most people, right now, it isn’t.

This is the part Heidegger turns against me. The phone is ready-to-hand. It has receded so far into the texture of ordinary life that we reach for it without deciding to. A sixth device you have to remember to charge and choose to put on is, by his own logic, present-at-hand in a way the phone no longer is. It sits there as an object, a thing you are aware of carrying, a decision you have to make each morning. The very newness that makes it exciting is what keeps it from withdrawing. It cannot vanish the way the phone vanished, because the phone got a fifteen-year head start on becoming invisible.

So maybe the question was never whether glasses replace the phone. The history of computing rarely works by succession anyway. The laptop did not kill the desktop. The tablet did not kill the laptop. The watch did not kill the phone. Each new device found a narrow thing it did better than anything else and settled into the gap. The laptop took deep work, the tablet took the couch, the watch took the glance, the earbuds took the ear. Smart glasses may simply take context, the AI that sees what you see, and sit alongside everything else rather than above it.

I still think about my own glasses, the ones I stopped feeling years ago. They withdrew completely, which is the only reason they work. Whether anything with a chip in it ever earns that same forgetting is the question I cannot answer, and the one the whole industry is betting its next decade on.

This week I kept returning to cloudscapes inside interiors. Not clouds through a window. Clouds brought into the room, painted across wallpaper, murals, printed onto surfaces that used to be walls. There is something strange in it. The sky becomes domestic. The wall stops acting like a wall.

One reference

Cloud imagery has always carried a softness, but indoors it gets complicated. It can read as romantic, theatrical, a little false. A cloudscape does not open the room. The walls stay where they are, the ceiling holds. What shifts is psychological. It hands the room a horizon where there should only be enclosure. The edges loosen. Escape, without anyone having to move.

One design implication

So a surface is never neutral. Wallpaper, colour, image: they change the mood of a room without changing what the room does. A cloudscape can make a space feel lighter, dreamier, larger than its dimensions. Used without restraint, it tips into decorative escapism, a horizon that asks nothing of you.

One question

When does an image in a room make atmosphere, and when does it become a backdrop?

Until next week.

Thomas is a German tableware brand known for functional porcelain, modern forms, and an unexpectedly bright use of colour. The brand’s lifestyle photography communicates this personality well, but many individual products are presented conventionally: isolated against white backgrounds, with little sense of how they might live inside a home.

This project explores an alternative campaign direction for the brand’s white porcelain espresso cup and saucer.

The central idea was simple. The objects of everyday life form a kind of domestic architecture. Rather than treating the cup and saucer as decorative props, I treated them as structural elements — repeated, stacked, aligned, cropped, positioned in relation to their surroundings.

The result is a campaign system that gives a familiar white object more visual presence while staying inside Thomas’s colourful, optimistic brand world.

White Porcelain Is Hard to Photograph

White porcelain presents a specific art-direction problem. Its simplicity is the appeal, but it disappears easily inside an image and reads as generic when photographed without context.

So the challenge wasn’t to make the product more elaborate. It was to create visual interest while preserving the qualities that define it: precise geometry, glossy white porcelain, functional simplicity, familiar domestic scale, and the exact shape of the cup, handle, rim, and saucer.

The first temptation was to lean into the stereotype of German design—restraint, coolness, rigid minimalism, industrial precision. That framing was incomplete. Thomas’s actual identity isn’t cold or monochromatic. It uses saturated surfaces, playful compositions, warm domestic settings. A purely austere campaign would have imposed an external stereotype instead of responding to the brand in front of me.

The stronger direction held German modernist order together with colour, optimism, and ordinary domestic life.

The Interest Lives in the Relationships, Not the Object

The project was built around one tension: how do you make a simple white cup feel distinctive without redesigning the product or burying it in styling?

The answer was to move the source of visual interest away from decoration and toward relationships. The relationship between one object and another. The rhythm of repeated handles. The contrast between white ceramic and coloured laminate. The interaction of scale, cropping, and negative space. The way cups and saucers can read as columns, platforms, façades, small structures.

That shift turned the tableware from a passive object placed inside a scene into an active part of the scene’s composition.

Creative Direction

Objects as Architecture

The campaign treats porcelain as part of the architecture of everyday life. Cups become small volumes. Saucers become platforms and planes. Handles create repeated circular openings. Stacks establish height and hierarchy. The table and wall act as horizontal and vertical fields, like the floor and façade of an interior.

The reference is subtle, not literal. The images don’t turn the products into miniature buildings. They borrow architectural principles: repetition, balance, proportion, structure, rhythm, spatial tension, and apply them to objects.

Domestic Modernism

The images needed to feel designed without feeling sterile. I built a visual language around domestic modernism: clean geometry and controlled compositions, set inside believable household environments. Coloured laminate surfaces, softly painted walls, windowsills, natural daylight. These give the campaign a lived-in quality and establish context without props or lifestyle clichés.

Colour as Structure

Because the product is white, colour became a structural device rather than a decorative one. The campaign uses pistachio green, powder blue, mint, warm beige, muted brown — enough contrast to define the porcelain silhouette, drawn from the bright character already present in Thomas’s identity.

The palette stays restrained inside each image. One or two broad colour fields frame the product. No competing colours fighting for the same frame.

Light and Material

Soft raking daylight reveals the glaze, the rim thickness, the curves, the small surface reflections. The material contrast carried as much weight as the light: glossy porcelain against matte laminate, hard ceramic against soft linen, precise geometry against diffused daylight. The lighting avoids luxury drama. It is quiet, directional, domestic, and it keeps the object’s material qualities visible.

Image-Making Approach

The campaign was made through an iterative, AI-assisted process: product reference photography I shot myself, detailed art-direction prompts, and repeated visual evaluation.

The goal wasn’t to generate attractive tableware imagery. The hard constraint was product fidelity across different compositions. Reference images defined the cup’s cylindrical body, the handle shape and attachment points, the proportions between cup and saucer, the rim thickness, the glossy white finish. Each image was then built around a single compositional principle.

Repetition. Multiple cups aligned in grids or repeated across a surface. Keeping every handle oriented the same way created a strong circular rhythm and made the product feel like part of a system.

Stacking. Saucers stacked or layered to build height, depth, and overlapping planes. Tight cropping let the repeated rims work as a graphic pattern.

Cropping and scale. Some compositions isolate a single handle, rim, or partial saucer. Enlarging a familiar detail makes the product feel monumental while keeping its real material qualities.

In-situ context. Early explorations placed the product against seamless editorial backdrops. Controlled, but detached — from the brand and from the product’s domestic purpose. The stronger images give the object a spatial relationship to everyday life: a laminate table, a windowsill, a wall, a linen surface. This was the turning point of the project. The product became more distinctive once it was given somewhere to live.

Prompt Writing as Production Design

Prompt writing functioned as production design. Each prompt separated four kinds of information.

Product constraints: the geometry, proportions, material, finish, and elements that could not change. Composition: camera angle, framing, object count, spacing, orientation, cropping, hierarchy. Environment: surface material, wall colour, domestic context, negative space. Lighting and tone: direction of daylight, shadow quality, material contrast, atmosphere.

Instead of asking for a general image of espresso cups on a table, a prompt specified two rows of three cups, identical handle orientation, an overhead viewpoint, a mint-green surface, no styling props. The precision reduced ambiguity and made the outputs easier to evaluate against a consistent system.

Evaluation and Iteration

I assessed each image against three criteria.

Product fidelity. Did the image hold the recognisable shape and proportions of the reference? The common failures were altered handle geometry, inconsistent rim thickness, distorted saucers, and a broken relationship between cup and base.

Compositional strength. Did the image have a clear visual idea? An image wasn’t successful just because it looked real. The strongest ones had deliberate spacing, a memorable silhouette, controlled repetition, a clear relationship between product and frame.

Brand fit. Was the result plausible for Thomas? Some images were sophisticated but too austere — they drifted from the brand and, more importantly, lost product fidelity. Others added props or food styling that pulled focus off the product. The final direction balanced fidelity with order and play: clean and precise, but using colour and context to avoid a cold, anonymous result.

Campaign System

The project produced a flexible set of image types, built to support several brand applications.

Hero images: wider in-situ compositions that establish the campaign world and the central idea of domestic objects behaving like architecture. Product stories: closer views of the cup and saucer in use, espresso with visible crema, holding the geometric language. Graphic crops: extreme close-ups of handles, rims, and stacked saucers, more abstract, suited to social, editorial layouts, and transitional moments. Repetition studies: overhead grids and repeated arrangements that introduce pattern and rhythm, usable for category pages, banners, or animation.

Together they read as a system, not a collection of unrelated AI experiments.

What I Learned

AI image generation doesn’t remove the need for art direction. It makes the art-direction decisions more visible. A weak concept produces generic outputs no matter how polished the technique. The quality of this project came from the visual thesis, the constraints, the willingness to read the failures, and the consistency held across many iterations.

The most important thing I learned was that context strengthened the product. Removing the cup from its environment looked, at first, like the more editorial move. In practice it weakened the brand connection. Placing the product inside a carefully designed domestic space made the images more specific, more believable, more ownable.

I also learned to separate realism from accuracy. An image can be photographically convincing and still misrepresent the product. For commercial work, product geometry isn’t a secondary detail. It’s the requirement.

Outcome

The final direction shows how an understated white porcelain product can carry a rich visual identity without being redesigned or over-styled. Through repetition, colour, architecture, and domestic context, the work pushes the product past white-background photography and presents it as part of a way of living.

The system it produced is recognisably Thomas, adaptable across formats, colourful without being decorative, architectural without being cold, and AI-assisted without letting the technology become the concept.

Project type: Self-initiated speculative campaign
Role: Creative direction, art direction, visual research, prompt development, AI image generation, image evaluation
Focus: Product photography, campaign systems, brand interpretation, generative AI
Brand studied: Thomas
Deliverables: Campaign concept, visual language, product imagery, image series and potential applications

Created independently as a speculative brand study. Not commissioned or endorsed by Thomas.

AI can generate images, summarize documents, produce interfaces, write drafts, organize research, and make rough ideas feel finished before they have been fully thought through. A lot of the work now moves faster. Some of it feels easier. Some of it feels stranger.

But speed has not made judgement less necessary. If anything, it has made judgement more visible.

Workingnotes is where I write about that space: the gap between output and meaning, between something that technically works and something that has been properly noticed, chosen, shaped, and understood.

The central question I keep returning to is:

What are we outsourcing too quickly and what is still worth noticing for ourselves?

That question shows up in different ways here.

Sometimes I write about AI and authorship: what tools can accelerate, what they distort, and what still belongs to human judgement.

Sometimes I write about interfaces and attention: the small design choices that guide, interrupt, reward, confuse, or quietly train the way we move through the world.

Sometimes I write about taste and culture: images, objects, media, aesthetics, and the visual systems that shape meaning, power, belonging, and desire.

And sometimes I write shorter notes called Cues: fragments from books, images, tools, rooms, conversations, and everyday frictions that reveal something about attention, taste, or judgement.

This is not a newsletter about rejecting technology. It is also not a newsletter about treating every new tool as inevitable progress. It is about paying closer attention.

To the interface.
To the image.
To the sentence that sounds right but says very little.
To the object that carries more history than its branding admits.
To the AI-generated thing that feels almost finished, but not quite.
To the small moment where a person still has to decide what matters.

If you are new here, these are good places to begin:

What 657 tasks taught me about the AI conversation
An essay on what using AI repeatedly reveals about the gap between hype, usefulness, and actual creative work.

Not quite: AI and the work that stays yours
A reflection on using AI to build a product, and why faster execution does not remove the need for design judgement.

The colour plants don’t use
A piece about fluency, coherence, and why polished AI output can feel like understanding without actually being understanding.

On looking down
An essay on phones, waiting rooms, boredom, and how interfaces train our attention.

Who gets to profit from tradition?
A critique of taste, branding, authorship, and what happens when cultural tradition becomes a marketable aesthetic.

Workingnotes is for designers, writers, builders, and curious people trying to stay attentive while the tools around us keep getting faster.

The work is changing.

The need to notice has not gone away.

I know where it comes from. There is a small room in my waking life waiting to be painted, fans of paint chips spread on the table, all of them blue, none of them this blue. The chips have names like Borrowed Light and Stiffkey Blue. The dream does not bother with names.

Miró wrote it in cursive under a small blue cloud on bare canvas: ceci est la couleur de mes rêves. A painting that is mostly a confession. I didn’t understand that before.

Because the blue in the dream is exact. It knows what it is. It is the colour of distance that has decided to come close, the far hill, the hour after sunset, moved indoors and made personal. Every chip on the table is an approximation, and the dream keeps correcting me, night after night, patient as a teacher with one lesson.

Tonight I will go back and I will stand inside it and try again to remember it exactly, knowing the morning takes most of it, knowing I will spend the day with the fans of paper, looking for the part it lets me keep.

Until next time.

At some point, I looked up.

Everyone else was doing the same thing.

Rows of people, all folded into the same posture. Necks bent. Shoulders rounded. Faces lit from below. Strange, though it is not strange anymore. That might be the stranger part. How quickly a room full of people looking down becomes ordinary.

I noticed the posture first. How bad it looks on us. Not just physically, though there is that too, the collapsed neck, the inward curve, the body slowly shaped by whatever it keeps returning to. But something else bothered me more.

No one was just waiting.

No one was staring at the wall. No one was letting their mind wander. No one was being bored in the old way, the useless way, the way that sometimes turns out not to be useless at all.

I do this too. I was not outside the scene. I was part of it until I happened to look up.

Dopamine nation

Anna Lembke, in Dopamine Nation, describes the brain’s pleasure-pain balance as a seesaw. Every hit of dopamine tilts it one way, and the system corrects by tilting back harder. The pain side. The restlessness. The reaching for the phone again, not because you want anything in particular, but because the baseline has shifted and neutral now feels like lack.

She is writing about addiction. But the mechanism does not require a substance. It requires only repetition and a reward that arrives faster than thought.

Simone Weil wrote that attention is the rarest and purest form of generosity. She meant something precise: not concentration, not effort, but a receptive emptying. A making-space. You do not grab at the thing you are attending to. You wait for it to arrive.

What strikes me about that formulation is how poorly it describes anything I do with my phone. Scrolling is the opposite of waiting. It is the continuous foreclosure of arrival—something comes, and before it can settle, something else is already there. The feed does not want you to attend to anything. It wants you to keep moving. That is a different act entirely, with a different effect on the mind.

Weil thought attention could be trained. That it was, in fact, the thing worth training. Most of what we call education, she wrote, is really practice in the quality of attention we can bring to a problem we cannot yet solve.

I think about that in waiting rooms. A hospital is already structured around problems you cannot solve. You are there because something has exceeded your control. The waiting is not incidental to that; it is part of it. You are meant to sit with not-knowing, with the particular discomfort of a door that has not opened yet.

Maybe looking down is how we avoid that. Not the boredom. The uncertainty.

Staring into nothing

My best thoughts almost never arrive when I am trying to have them. They come when the phone is silent and away from me. When I am brushing my teeth, staring at nothing. When I am walking without headphones. When I am waiting for water to boil. When there is no input, no small machine offering me something easier than my own mind.

A thought needs room to embarrass itself. It needs to start badly, drift, contradict itself, circle something it does not understand yet. It needs the kind of time that looks like nothing from the outside.

The phone is too good at filling that space. Not because it entertains us, exactly. Not because it distracts us. Because it interrupts the empty interval where thoughts begin to form before we know what they are. Lembke calls this the central problem of modern life: we have engineered away the discomfort that used to be productive. The gap that itched. The restlessness that eventually turned into something.

I too have been there

I do not think every empty moment needs to become meaningful. Sometimes scrolling is just scrolling. Sometimes distraction is mercy, I too have been there. But I wonder what happens when we lose the habit of looking up. When we stop trusting a blank wall, a quiet minute, a boring stretch of time, to do anything for us.

Weil’s attention is not a productivity practice. It is closer to a posture toward the world. A willingness to be present to something without immediately converting it into use.

A waiting room asks for that. So does most of a life, if you stay in it long enough.

I did not look down again and I got to see the room clearly: all of us waiting, none of us quite there.

Leave a comment

There is a kind of growth that only happens in private, in the dark, in the time between when something is planted and when anyone can see that it took. I cannot point to it yet. I cannot say here, look, this is the new thing. I can only report the conditions: that I have been weeping more, the trees are still bare, and something in me is moving the way roots move, which is slowly, and downward first.

A friend sent me a photograph this week of a pink cake with a single tulip laid across the top. The tulip was already past its peak, the petals slightly furled inward, the green of the stem darkening at the edge. Beautiful, but not in the way fresh flowers are beautiful. Beautiful because it had been chosen at exactly the moment it was beginning to give itself away.

I have been thinking about that photograph for days.

Maybe what looks like decay from one angle is generosity from another. The tulip is not wilting on the cake. It is bending down to meet it.

Patience, patience.

I am letting myself believe, just for the length of one slow inhale, that something underground is drinking.

01 - Don’t be afraid to weep

Don't be ashamed to weep; 'tis right to grieve. Tears are only water, and flowers, trees, and fruit cannot grow without water.

Brian Jacques, Taggerung

02 - Flowers

Until next time.

I like the exercise. It pushes past the abstract dread of AI will change everything toward something concrete enough to use. Anyone who sits down with their calendar and sorts honestly will know more in an hour than they did before. The framework deserves its uptake.

But sitting with it longer, I keep finding the same three soft spots. They matter enough that the exercise can leave you with a tidy chart and the wrong conclusion.

What the bucket test gets right

Most AI-and-work conversation happens at the altitude of job titles. Marketing will change. Accounting will change. True, and useless. You can’t act on a job title. You can act on tasks.

The bucket test moves the conversation down a level. It makes the invisible structure of your work legible to yourself, sometimes for the first time. That alone is worth fifteen minutes.

It is also honest about Bucket 3. Some judgment can’t be compressed into a process, not because we’re sentimental but because the work is irreducibly entangled with situation, relationship, body. Naming that without flinching is a useful corrective to the doomier corners of the discourse.

So the framework is doing something real. The question is what it leaves out.

Blind spot one: tasks are not hours

Sort your tasks and you get three counts. Five in Bucket 1, eight in Bucket 2, four in Bucket 3. The piles look balanced. The chart looks calm.

But tasks are not hours.

Two roles from WorkMap, side by side. A copywriter has roughly a dozen Core tasks: developing campaign concepts, adapting tone across media, generating headline variations, drafting first-draft copy, reviewing trends, presenting to clients. Most score high on AI fit. A few sit in the middle. Client presentations and creative negotiation stay human-led. Sorted into buckets, the spread looks reasonable. Something in each pile.

A dental hygienist has a similar number of tasks: patient assessment, cleaning, X-rays, recall scheduling, history review, equipment sterilisation, patient education during treatment. Most are physical, hands-on, irreducibly human. Only recall scheduling and history review drift into Bucket 1 or 2. Sorted into buckets, this spread also looks reasonable. Bucket 3 heavier, but the others have things in them too.

Both roles end up with a chart. Both charts read some exposure, some safe work. They look like the same situation.

Same buckets, different weeks: Two roles. Identical task distribution. Opposite exposure.

They are not. The copywriter spends maybe 5% of the week on the human-only tasks and 70% on the AI-fit ones. The hygienist spends 90% of the week on the human-only tasks and 10% on the AI-fit ones. The bucket test, written as it is, treats these as comparable.

Counts without hours are decorative. The audit only starts working when you weight by time.

Blind spot two: you can’t redesign a plate you don’t control

The framework’s advice when most of your work sits in Bucket 1 is to deliberately move tasks toward Bucket 2. Document your process. Train AI. Free up time for higher-judgment work.

For some people, this is good advice. For many more, it is wishful.

Consider an administrative assistant. The Core task list reads almost like a definition of Bucket 1 itself: drafting routine email, maintaining filing systems, formatting documents, data entry, mail triage, calendar coordination. AI handles most of it well. The framework would say move toward Bucket 2. Become the person who orchestrates the AI doing the routine work, and reclaim the time for higher-leverage tasks.

But administrative assistants do not usually get to redesign their own plate. The work that comes to them comes assigned. The manager handing over a forty-page document to format is not asking whether AI could do it faster. They are asking for it formatted. The shift from doing the task to orchestrating the task requires permission from someone above you, time that is not already booked, and often a job description rewrite. None of those things appear in a bucket chart.

This is the harder structural fact the framework softens. Task allocation is rarely yours to choose. Some roles are positioned to redesign their work: senior individual contributors, managers, consultants. Others are positioned to receive work. The exercise is most actionable for the people who least need it, and least actionable for the people who most do.

A useful audit names this directly. Of the tasks on your plate, how many do you actually control? The path from Bucket 1 to Bucket 2 is not a personal choice. It is a negotiation.

Blind spot three: the conveyor belt cuts both ways

Ashisha’s most evocative image is the conveyor belt. Bucket 2 work slides into Bucket 1 over time, and new Bucket 3 work opens up as AI advances. This is true. But the belt does not move in one direction for every role.

For some work, the belt does what the framework describes. Routine tasks get automated, judgment work expands, and the human keeps moving up the value stack. The trajectory is clean.

For other work, the floor rises and the ceiling drops at the same time. The routine tasks get automated, yes. But the senior judgment tasks above them are also being chipped at. A junior copywriter loses drafting work to AI. A senior copywriter loses some editorial judgment as AI gets better at brand voice. The hollow grows in both directions.

The bucket test cannot tell these apart. It gives you a snapshot — three piles, today. It cannot show you whether your Bucket 3 is growing or shrinking, whether the work that anchors your role’s existence is on stable ground or being quietly redefined underneath you.

A snapshot is not a trajectory. An audit that doesn’t ask which way your floor is moving will let you feel safe in a role being hollowed out from the top.

A sharper exercise: three questions, not three piles

If you’re going to audit, audit better. The same fifteen minutes, but with three questions running underneath the sorting.

What share of your week is AI-exposed? Not how many tasks. How many hours. Open the calendar, weight the tasks by time, and find the percentage of your actual working week sitting in Bucket 1 or Bucket 2 territory. That number is what tells you whether you have an urgent problem or a manageable one.

What share of that exposure is yours to redesign? Of the tasks you could move from Bucket 1 to Bucket 2, how many do you actually control? Which require a manager’s sign-off, a team norm shift, a job description change? The honest answer separates the work you can act on this week from the work you can only advocate for.

Where is your role’s floor and ceiling moving? Look at the top of your work: the judgment calls, the senior decisions, the things only you do. Is that surface area growing or shrinking? Look at the bottom: the routine, the entry-level rungs. Is that being automated faster than new judgment work opens up above? Two roles can have identical bucket distributions today and be on opposite trajectories.

These three questions don’t replace the bucket test. They do what an audit is supposed to do. They turn a snapshot into a position, and a position into a direction.

What this means for the audit

The bucket exercise is a good first cut. The right shape, the right altitude, fifteen minutes of honest sorting. But it stops one layer short of the answer most people are actually looking for, which is not where do my tasks sit but what does this mean for my role, and what do I do about it.

That second question needs time-weighting, control-mapping, and a read on which direction the floor is moving. The buckets, on their own, are too quiet to tell you any of that.

This is the gap I keep running into when I look at the WorkMap data — that role-level distributions and trajectory signals reveal things task-counting alone can’t. But the audit you can do manually, with your calendar open and an hour to spare, is still real. It is still worth doing.

Just keep going after the buckets. Where the tasks sit is the beginning. What it means for you is somewhere past it.

The faces from the 1840s look out at me unsmiling, still from long exposures that asked people to hold for half a minute, twenty seconds, ten. I’ve been at it an hour when something obvious occurs to me.

Most of these people had never seen themselves clearly, not in the way I can see them now.

A face other people know

For most of human history, your face was something other people knew. They recognized you. Loved you. Judged you. Your face was theirs to remember, theirs to describe back to you if you asked. Your own direct relationship to it was looser. More occasional. Glimpsed in water, in polished brass, in a square of smoked glass if you were rich enough to own one. You had a face. You just didn’t have it the way you had your hands.

Narcissus saw himself in a pool and didn’t know it was him. He fell in love with what he thought was a stranger. The oldest story we have about someone looking at their own reflection is a story about misrecognition. The surface didn’t give him back to himself. It made him a face he couldn’t place.

A medieval farmer probably knew his hands better than his face. He knew his gait. The sound of his own voice from inside his skull. The ache in his back at dusk. He could describe his wife’s face, his children’s, the saints’ on the church wall, his enemies’. His own might come back to him as a flicker. A dark shape in water. Something glimpsed and then lost.

Then photography arrives, and the self becomes fixed. Observable. Something you can hold in your hand and look at from the outside. The face stops being occasional. It becomes a noun.

Self-recognition

We tend to assume self-recognition is natural and constant, as if every human in every period had roughly the same relationship to their own appearance. But maybe that relationship intensified as reflective technologies became more accurate, more available, and eventually impossible to avoid.

From there it only gets worse. Or more, anyway. Mirrors get cheap. Cameras get small. Phones get cameras. I catch myself now in a black phone screen, the laptop’s webcam preview, a tagged photo I didn’t know was being taken (early Facebook days), a Zoom window I open for a meeting and spend the next hour avoiding. The face is no longer occasional. It’s always nearby. Some part of it is asking to be managed.

Elena Ferrante wrote once about finding a photograph of herself at seventeen that she liked so much she framed it. Friends and relatives, even people who had known her for decades, kept saying some version of the same thing back to her: the woman in the frame didn’t match the one they remembered. Ferrante eventually admitted she liked the picture precisely because it didn’t match her own self-image either.

This is the part I keep thinking about. We have more access to our own faces than any humans in history, and we still don’t recognize them. We see them constantly. We don’t know them better. The girl at seventeen was her, and she didn’t recognize her. The image was real, and so was the disagreement.

Earlier humans may have experienced themselves more from the inside out. Through labour, sensation, voice, memory, the body as it moved through weather and fatigue. Modern people experience themselves from the outside in. We see ourselves seeing ourselves. We watch ourselves perform being ourselves. Even alone, some part of us has learned to imagine the camera.

Vanity is too easy

But Ferrante’s photograph suggests the loop isn’t closed. The image keeps coming back, and we keep not recognizing it. We get more practice without getting more accuracy. The face stays slightly stranger than we expect, the way a recording of your own voice does.

There is a strange intimacy in that, but also a kind of violence.

We know ourselves now through endless visual correction. Is my hair weird? Do I look tired? Is that how I smile? Is that what I look like when I talk? Is my face symmetrical? Is my skin worse than yesterday? Do I look like the version of myself I think I am?

This is not only vanity. I think calling it vanity is too easy.

It is also infrastructure.

When the world gives you constant access to your own image, you develop a different kind of self-consciousness. You become both subject and spectator. Person and object. Face and watcher of the face.

There is a quiet grief in that.

Not because mirrors or photographs are bad. I love photographs. I am moved by them. They are one of the most beautiful technologies humans have ever made. They let us keep the dead close. They let us study time. They let us notice gestures that would otherwise disappear.

But every technology that preserves something also rearranges our relationship to it.

Photography preserved the face.

It also made the face newly available for inspection.

And maybe that is the part I keep thinking about. Not the invention itself, but the psychological afterlife of being able to look at yourself too clearly, too often, too easily.

What did it feel like to live before your face was familiar to you? I don’t know. Maybe people have always been self-conscious in some form. Maybe the village gaze was just as powerful as the camera. Being known by others is probably its own mirror.

But still, I wonder. I wonder what kind of self forms when your face is not constantly being handed back to you. I wonder whether a person could feel ugly or beautiful in the same way without the visual confirmation. I wonder whether the face, before it was fixed, had more room to remain mysterious.

Because now, we know our faces almost too well.

Or maybe that isn’t quite right.

Maybe we don’t know them better.

Maybe we just see them more.

And those aren’t the same thing.

Earlier I wrote about it as a process. Idea, prompt, revision, push until the output gets closer to what I had in mind. That still interests me, the part where a vague instinct becomes specific. But lately I’m more drawn to what happens before any of that. The wandering. The strange almosts. The images that don’t belong to a project yet and keep pulling at me anyway.

I’ve started using image generation more like a sketchbook than a production tool. Less how do I get the final output, more why do I like this.

The shift has made the work feel lighter. I’m not always trying to land somewhere. Sometimes I stay with an atmosphere, a colour, a face, a room, and see what the model does with it. Sometimes the image is wrong in the exact way that makes it interesting. Sometimes it misses the brief and finds something else. Sometimes it gives me a version of a thing I didn’t know I wanted to see.

What I like about this kind of playing is that it reveals taste almost accidentally. You start to notice what you keep asking for. The kinds of spaces you return to. The light you trust. How much softness or strangeness you’ll allow before it tips into something you don’t want. The difference between an image that is technically impressive and one that catches you.

That difference is the whole thing.

A polished image can feel dead. An imperfect one can have a pulse.

I’ve been paying attention to the pulse more than the polish. Not because craft doesn’t matter, but because the early stage is often where the energy is. Before everything gets cleaned up. Before the idea has to justify itself. Before the image has to become an output.

These aren’t case studies. They aren’t a clean demonstration of a workflow. They’re closer to visual notes. Things I made while testing, drifting, repeating myself, contradicting myself, getting bored, getting curious again.

Looking at them together, I can see patterns. I’m drawn to images that feel slightly suspended, as if something has just happened or is about to. Scenes that feel designed but not over-explained. A tension between artificiality and warmth.

The word I keep coming back to is almost. Almost editorial. Almost cinematic. Almost remembered. That’s probably the part I’m most interested in, because a finished image asks to be judged. An almost-image asks to be followed.

I don’t have a conclusion about any of this. Mostly I’m interested in what happens when image generation is allowed to be casual without being careless. When the point isn’t always to produce the perfect thing, but to notice what keeps appearing.

Maybe that’s the value of the sketchbook, digital or otherwise. It gives your taste somewhere to show up before you know what to do with it.

For now, this is what mine looks like.

Water demands a particular quality of attention, it’s the kind that happens when you’re looking, and then you’re just there. Held. In a trance.

Some of the images are almost luminous. Flat, reflective, infinite. A single tree, or a structure cutting cleanly through the horizon, and everything feels composed, still. I think of Tadao Ando's Honpukuji Temple here. His planes of concrete meeting water with devotional restraint. The water doesn't decorate the space, it is the devotion. A stillness that isn't neutral ast it subverts scale, removes context, and removes the human as reference point. And in that absence something creeps in, quiet and unhurried. An act of devotion that arrives unannounced. And lingers.

Other times, the water darkens. Folds into itself, textured and restless. No subject to anchor you, just depth without edges.

What draws me across these works is how each artist lets water destabilize certainty differently. In Klimt’s Water Serpents II, bodies dissolve into ornament and current, suspended somewhere between intimacy and myth. In Rawles’ A Dream for My Lilith, the surface fractures into light and ripple, holding the figure while subtly distorting it. The woman’s presence made fragile through water’s constant motion. In Hamilton’s Ophelia Plate 0124, objects submerged in near-black water feel suspended outside time, as if the image itself is holding its breath.

Across all of them, water refracts, distorts, absorbs. It holds form while quietly undoing it.

I’ve been sitting with that. The way these artworks don’t tell you what to feel or where to land. They just keep the surface moving. And you stay there, looking, aware that something is happening just below where you can see.

Until next time.

Subscribe now

The fantasy is of being made obsolete, but peacefully. Not crushed by the machine, released from the race. Not everyone followed Prometheus into the age of fire. Some stayed with Pan in the meadow.

I understand the appeal, but most of us are not outside it. We are inside it, negotiating, deciding what to accept, what to resist, what to keep doing by hand. That last part is where my own relationship with AI has become more hopeful, and more genuinely collaborative.

AI can build the interface. It cannot decide what matters.

Using AI to build a product makes the whole process easier. It does not make it easy to design one. That distinction feels worth holding onto, especially if you are a UX or product designer. AI coding tools have shortened the distance between an idea and a working product. They can help us move faster, they make development feel less intimidating. But speed is not the same as clarity, and a working interface is not the same thing as a well-designed one.

I learned this building my Opportunity Brief Agent, a small AI workflow that turns vague product questions into evidence-backed briefs for product and design teams.

I started where a lot of people seem to start now: with the assumption that you can take an idea, spend a weekend with it, and ship something good. There is some truth in that because you can get further, faster, than before. But the build does not erase the design work, it just moves it.

My project did not begin in code. It began the way most design projects do, with research, a messy sketch, a product question, and a stack of decisions I had to make before AI could be useful: what should this agent actually do, what does the user need to input. What kind of output would feel useful instead of generic, how should the workflow build trust, where does the interface need to feel guided, and where should it stay flexible?

These had to be answered before the tool could become anything coherent.

I started on paper, then moved into Figma to refine the structure. Layout, hierarchy, spacing, colour, typography, content structure, tone. Only after that did I bring it into Claude Code.

Building

AI was useful because I had already made enough decisions for it to respond to. It had something to interpret. A direction, a prototype, a visual system, a product logic. Without that, I would have been asking the AI to guess.

And the thing about AI-generated interfaces is that they get you close, but not all the way there. The structure works, the page renders, the interaction functions, the product technically exists. But there is usually a gap between something that works and something that feels designed.

For this project, the gap showed up in the details. Colours that were not quite right. Font sizes that needed adjustment. Spacing that did not match the rhythm I had built in Figma. The interface had the bones, but it did not yet carry the judgment behind it.

So I went in manually because I wanted to stay close to the material. The interface existed and behaved somewhat correctly, it did not have my judgment in it yet.

This is the part of using AI in design that I find most interesting. It gives you speed, but it also forces you to confront what your actual design judgment is. When the tool gets you 60 percent of the way there, the remaining 40 becomes more visible. That last stretch is where taste, clarity, restraint, and product thinking matter most.

It is also where agency matters. I could have kept asking the AI to fix every detail. I could have prompted my way through the colour system and the type scale, but not every task should be optimized away.

Some parts of the work are valuable because they keep your eye sharp. They force you to notice what feels off. They remind you that craft is a way of thinking through the product, not a phase you complete and leave behind.

Too much delegation dulls those instincts. If you never touch the details yourself, you lose the ability to see them and I think that is part of why we become designers in the first place.

Designers should be experimenting with these tools seriously. They change what is possible for a solo designer or a small team. They make it easier to prototype real interactions, test ideas faster, move beyond static mockups. But they do not remove the need for design principles. They make those principles more important.

Using AI in design rewards clarity. If the direction is vague, the output is vague faster. If the product logic is weak, AI will not make it strong. If the visual system is inconsistent, the build reproduces the inconsistency. AI can accelerate execution but it cannot replace the responsibility of knowing what you are trying to make, and why.

Hephaestus was the only Olympian who actually made things. The others ruled, schemed, fell in love, went to war. He worked in the forge, lame from a fall, hammering metal into shields and chariots and locks. The gods kept coming to him because nothing they could conjure replaced what he could shape with his hands.

That is the figure I keep returning to. Not Pan in the meadow, not Prometheus with the stolen fire, but the maker at the anvil, deciding where to strike.

Easier does not mean easy.

The product still needs structure, hierarchy, clarity, refinement, and taste. It still needs someone to look at the thing and say: not quite.

Then go back to the forge.

“Hey, congratulations. It’s been a year since you signed up.”

“God damn,” I said. “It’s been a year.”

A pause, just long enough for me to worry. Then everyone around us laughed. Nothing I said was clever, it was just the timing, the disbelief in my voice, and the fact of being there at all. The kind of thing you don’t plan and can’t repeat.

It lasted maybe four seconds. Then everything moved on.

I think about that moment a lot now, sitting in front of my computer in an online AI class, talking to people through a grid of small windows. The class is great, the system works. It’s efficient, engaging, and above all, incredibly convenient.

But the form of the interaction and the meaning of it don’t quite line up for me.

The absurd

I keep coming back to Camus, the gap between our desire for meaning and the structure of reality. The space between what we expect life to feel like and what it actually is. That gap feels like it’s widening. Not because technology is failing, but because it’s succeeding at exactly what it was designed to do.

Every tool I use now, AI tools especially, removes friction. That’s the point. Drafting an email used to require thinking through what you wanted to say. Now a model offers three versions before you’ve finished the question. Research used to involve sitting with not knowing. Now the answer arrives, formatted, before the not knowing has time to do anything.

This is presented as a gift. Mostly, it is. I use these tools every day. They make me faster, sharper, more capable. But friction is where certain things live and they don’t survive its removal.

That gym moment couldn’t have happened over Zoom. It depended on being physically caught off guard, on a stranger’s voice landing in a room with other strangers, on the slight lag of real laughter spreading through a space. There was no transcript. No optimization. It happened because the conditions were rough enough to allow it.

A lot of what I value turns out to be like this. Running into someone you weren’t looking for. The pause before someone answers a hard question. The walk home where an idea finally lands because you stopped trying to force it. These things don’t translate well to a screen. And they don’t survive being optimized for.

The friction remover

AI is the most powerful friction-remover we’ve ever built. It’s already collapsing the distance between intention and output, between question and answer, between draft and finished thing. That collapse is genuinely useful.

But it’s also changing the texture of attention in ways we haven’t fully named.

I notice it in small ways online. People selling image prompt guides to others when they could do the research themselves—test, struggle, iterate, arrive. The struggle was the thinking. Without it, you get an output, but not a position. You know what works, but not why you believe in it.

The gap in work

A similar pattern shows up in how people talk about work. There’s a growing restlessness, even among people who are doing everything right. A kind of exhaustion—investing time, energy, identity into systems that feel increasingly abstract. Work that lives in documents and dashboards. Collaboration that exists mostly through interfaces.

People are leaving. Going freelance. Building portfolio careers. Framing it as a rejection of corporate paths. I think it’s something else. I think it’s a response to the same gap.

The structure has been optimized so thoroughly that the parts of work that used to feel like life have been smoothed out of it.

Noticing with precision

The irony of writing this in a Substack post isn’t lost on me. There’s no clean outside. I’m inside the structure I’m describing, and so are you.

What I’m trying to do, what I think is worth trying to do, is simply notice to not reject, not romanticize the analog, not pretend the 90s were better in any absolute sense, even if I have a soft spot for them.

Just notice, with some precision, what’s being removed when friction is removed.

Some friction is waste. Loading screens, bad UX, paperwork—get rid of it. And some friction is the medium through which a thing becomes real. The slow part of learning. The discomfort of a hard conversation. The four seconds in a gym lobby. These don’t scale. They aren’t features. They’re where a particular kind of life happens.

The work now isn’t to close the gap between what we want and what we have. The gap is the condition. Our work perhaps is to understand it well enough to choose (deliberately) what we’re willing to give up to make things easier.

And what we’re not.

I don’t, yet. I just sit with it for a second.

I’ve been noticing these spaces. Not the impressive ones, not the ones built to capture attention. The quiet ones. The interfaces that seem to understand something most digital spaces forget, which is that the person arriving doesn’t always arrive with clarity. Sometimes they arrive with a half-formed question. A vague frustration. A sentence that doesn’t yet know what it wants to be.

01. Google

Google’s homepage is the obvious ancestor. Logo, field, white space. It waits. There is confidence in the restraint, but the contract is transactional: convert your need into a query, reduce the mess of your thinking into searchable language, leave.

02. Prompt boxes

The prompt box looks almost identical. Field, cursor, a few suggested openings. The grammar is the same. The contract is not.

The prompt box can receive a fragment. It can receive a paragraph that begins, I don’t know if this makes sense, but…

The search bar asks what you are looking for. The prompt box asks what you are trying to work through. One is a counter, one is a threshold. You do not have to arrive finished. You can begin messy.

A quiet room does not tell you what to think. It changes the conditions under which thinking becomes possible.

I keep coming back to that. The cursor blinking, the field waiting, the small grey suggestion. Nothing demanded yet. Just the door, slightly open.

Until next week.

That’s what led me to build WorkMap. Breaking roles down into tasks. 657 of them, across 70+ roles, using O*NET data. Each task gets a score, a plain explanation of what AI actually does there, and how the work changes.

The idea wasn’t to convince skeptics or impress early adopters. It was for people in the middle. People who’ve been told to “use AI” but haven’t been told how.

And once you start looking at work this way, it’s hard to go back.

The pattern no one names

Across all those tasks, one thing kept showing up: the variation inside a role is often bigger than the variation between roles.

The nurse example has become shorthand in this space, so I’ll use it. Documentation tasks score high. Patient assessment scores medium. Care coordination scores medium. Emotional support work scores low. That’s the entire range of exposure, inside a single shift, for one person with one job title. It’s not a just quirk of healthcare, it’s the pattern.

An operations manager: drafting status reports, reformatting data between systems, summarizing meeting notes. All score high. The same person’s vendor negotiations, conflict resolution, operational judgment under ambiguous conditions. All score low. The role-level framing tells that person almost nothing useful about their actual work.

A lawyer: document review, contract summarization, precedent research. High. Client strategy, courtroom judgment, ethical reasoning in novel situations. Low. The role label papers over a split with real implications for where someone should be developing skills and where they should be defending depth.

So when headlines say things like “knowledge workers are at risk,” it sounds meaningful, but it doesn’t actually help anyone understand their work. Task-level data makes the difference visible again. That’s what’s been missing.

What high exposure actually looks like

The common thread across high-exposure tasks is not sector, seniority, or credential. It’s actually the nature of the work: structured, repeatable, transferable to language.

Documentation. Report generation. Data formatting. Scheduling. Summarization. These appear as high-exposure tasks in nursing, in operations management, in legal work, in design, in financial analysis. They cross every sector boundary we use to organize the conversation. “AI is coming for white-collar work” only makes sense once you understand a lot of white-collar work is made up of document and data tasks.

What AI contributes to these tasks is worth naming precisely: the compress of time. A nurse who spent forty minutes on end-of-shift documentation may spend fifteen. A paralegal who spent two hours summarizing discovery documents may spend thirty minutes reviewing a draft. The task still requires the professional for accuracy checking, for contextual judgment, for the decisions that sit adjacent to the documented output. But the time profile changes. And once time changes, everything else follows, like cost, expectations, staffing.

That level of specificity is usually missing from the conversation and it is what matters in how individuals plan and how organizations resource.

The inversion

The part I find most interesting is this: the tasks people most identify with tend to be the least exposed.

Judgment calls. Relationship-based work. Physical presence. Safety-critical decisions. Creative direction. These cluster at low exposure because they don’t translate cleanly into language. They carry responsibility that can’t be handed off.

The implication is uncomfortable. Most of the anxiety is about losing “the core of the job,” that’s actually often the part least affected. The real shift is in everything around it.

A designer’s production work is exposed. Their creative direction isn’t.
A doctor’s admin work is exposed. Their clinical reasoning isn’t.

That doesn’t mean there’s no risk. If a big portion of your work gets compressed, roles change, headcount also changes. It also changes the question, from, “is my job safe?” to “which parts of my work are changing, and what happens if those get cheaper?”

That’s a harder question. But it’s the one that actually matters.

A Canadian footnote

The AI/work conversation is built almost entirely on American occupational data. O*NET is a US Department of Labor product. So when Canadian workers try to interpret these insights, they’re doing it through a system that doesn’t fully map to their reality. WorkMap includes Canadian roles mapped through NOC 2021. It’s a small correction, but it matters.

If your data model doesn’t reflect your labour market, your conclusions adds noise to decisions that matter.

No comparable tool I’ve found surfaces Canadian occupational classification data alongside AI exposure scores. That absence suggests the conversation has been globalized without being localized. It’s a pattern worth naming wherever it appears.

What the data can’t tell you

WorkMap has a confidence column; however, it doesn’t appear in the product.

Not because confidence is unimportant. Because the variation across tasks wasn’t meaningful enough to surface without creating false precision. When confidence scores cluster in a narrow band, displaying them implies differentiation that isn’t really there. A score of 72% versus 68% looks like a finding. It isn’t. Showing it would mislead people making real decisions.

This is one of the most underwritten problems in designing with data: what you choose not to show matters as much as what you show. More data doesn’t automatically translate to more clarity.

The graphic designer’s production tasks still sit in the same band as the data entry clerk’s and it points to something more precise. That’s the first move toward doing something useful with it.

Her audience didn’t just follow her skin. They followed her thinking about skin health and in turn created awareness around acne and its impact on mental health. So when her skincare line launched, it didn’t enter a neutral market. It entered into a relationship that had already been doing work for years.

I find this genuinely interesting from a UX perspective because this one inverts the sequence I’m trained to trust.

The sequence that got skipped

In product design, the logic runs in one direction: find the problem, build toward it, test whether the solution landed. It keeps you from falling in love with a solution before you’ve confirmed what you’re solving.

What the Earle launch did was arrive from the other direction. The trust preceded the product by years. And once trust is that established, it compresses everything downstream: the consideration phase, the skepticism, the need to be persuaded. People bought the products because the relationship had already made the case for its rigour and place.

The question that interests me is what happens next.

What the reviews are actually saying

Early coverage of Reale Actives describes it as “thoughtful,” “approachable,” “effective without being aggressive.” One review calls it “skincare that supports your skin without taking over it.”

These are positive reviews and they’re also soft ones. “Doesn’t irritate” and “won’t take over” are real positions, but they’re defined by what the product avoids rather than what it delivers. In UX terms, that’s a usability baseline, not a value proposition.

This isn’t unusual for a first launch, and it’s not necessarily a problem. But it does tell you something about where the work is still to be done.

When branding and product point the same direction

Rhode is the counterexample worth looking at. It gets dismissed as minimal, aesthetic, even vibe-driven. But underneath that surface, it’s tightly scoped in a way most influencer brands aren’t: a small product set, innovative phone case, a specific and nameable outcome (hydration and the glazed donut look). The brand doesn’t try to solve acne at a clinical level, it solves for predictability and feel.

It also turned out to be a successful model. e.l.f. Beauty acquired Rhode for approximately $1 billion in 2025. Problem-solution fit at a different layer, confirmed at scale.

Glossier built its early success by reframing the problem entirely. Not “fix your skin” but “enhance what’s already there.” The brilliance was that this framing made every product launch a confirmation of the same thesis: the Balm Dotcom, the Boy Brow, the Cloud Paint all solved for addition, not correction. Product and positioning became self-reinforcing because every new product proved the original claim.

In both cases, the brand made a specific promise the product could keep.

When it doesn’t hold

Lori Harvey’s skincare line entered the market with real visibility. Strong positioning, credible founder, genuine audience. Then, gradually, it lost traction. The brand’s Instagram has been quiet since December 2024. Everything is out of stock in the way that means clearance rather than demand.

What was missing isn’t hard to identify in hindsight: unclear differentiation, a product narrative that didn’t extend much beyond the founder’s profile, no visible feedback loop between what users experienced and how the brand responded. Without those, the brand stayed dependent on attention.

Attention is the least renewable resource in this market. It runs out. Or it moves on. Usually both.

What repeat purchase actually measures

You can generate demand without solving a strong problem (we can also see this with a lot of vibe-coded projects). The launch proves that, the conversion spikes, the press lands, the first sell-through, all looks like success. But repeat purchase is a different signal entirely. It measures whether the product kept the promise the brand made at launch, quietly, without the founder’s face in the frame, on a Tuesday when no one is paying particular attention.

That’s the gap this kind of brand has to close. Not in the campaign. After it.

Products don’t get to be right once. They have to keep being right, in the ordinary way, over time. The trust that got there first can open the door. It can’t hold it open indefinitely.