Every feature in a modern feed was added by somebody, for a reason, and the reason is almost never that users requested it.

I know this because I spent the last year building a social platform and had to decide about each one individually. Some of these were genuinely hard to leave out. All of them work, in the sense that they do what they were built to do.

Here are six, what each is actually for, and what the app looks like without them.

If you want the short version: I am building it at kindredcontent.app, and it opens properly on 17 November.

1. The feed never ends

What it does: refills faster than you can read it, so there is no natural stopping point.

What it is for: a feed with a bottom is a feed that lets you leave, and every minute you are gone is unsold inventory. The ending was not a design flaw. It was a leak.

Without it: you scroll the people you follow, you reach the end, and you are caught up. Most people find this disorienting for about three days and then describe it as the thing they like most. Your thumb keeps going for a second out of habit the first few times.

It is also, by a distance, the most expensive thing I gave up in engagement terms.

2. Autoplay

What it does: starts the next thing before you decided you wanted it.

What it is for: closing the gap. Specifically the two seconds after something ends, when you are briefly unoccupied and might get up. That gap is the decision point and autoplay exists to remove it.

Without it: things end. You choose the next one or you do not. It costs watch time and the gap is the entire thing people say they want back.

3. Ranking your own follows

What it does: decides which of the accounts you deliberately chose are worth showing you today.

What it is for: engagement. A feed that can draw on everything ever posted beats one limited to the people you picked, by the only measure being used.

Without it: you post, and the people who follow you get it. All of them, in order. There is nothing to boost past because there is nothing in the way.

This is the part creators notice first, and it is why I started building in the first place. I had 3,525 followers on my photography account and a post of mine reached 407 of them.

4. Streaks

What it does: counts consecutive days and resets if you miss one.

What it is for: it is the most effective retention mechanic anyone has built, and it works by making you feel slightly bad. The thing bringing you back is not wanting to be there, it is a number you have been maintaining.

Without it: you open it when you want to. Nothing is keeping score and nothing resets.

5. Notifications that are not about you

What it does: someone you might know. Activity near you. Something happened in a group you are barely in.

What it is for: a badge is a reason to reopen the app, and reasons to reopen are the product.

Without it: you get told when somebody replied to you, followed you, or posted something, because those are events involving you. The test I used was whether I would send it if it cost me a dollar. Most of what lands on a phone fails that immediately.

6. Ads between your friends

What it does: the obvious thing.

What it is for: it is not a feature of the product, it is the product. Roughly 97% of Meta’s revenue is advertising, which means the feed is the delivery mechanism and your attention is the inventory.

Without it: there is nothing to sell, which means there is nothing pulling the feed away from you. Every other item on this list is downstream of this one. Take out the ad product and the reasons for infinite scroll, ranking and streaks mostly evaporate, because they were all there to grow a number that no longer matters.

What is left

Less than you would expect, and that is the point.

A chronological feed of the people you chose. Comments that are not ranked. Landscape video up to three hours, because there is no incentive to chop it into something shorter. Full resolution photographs. Private feeds for a small group when you do not want to broadcast. Family accounts with teen controls, free, because gating child safety behind a paywall is the kind of thing I would criticize someone else for.

It is noticeably less compelling, in the specific sense that it does not compel. That is not a side effect I am managing around. It is the specification.

The honest part

The pool creators earn from is small right now, because the platform is new. Free accounts get an hour of viewing a week, and more than that is $4.99 a month, which means I am selling screen time, which is awkward given everything above. I have written about that awkwardness at length rather than hoping nobody notices.

And a feed that ends is a harder product to grow than one that does not. That is the trade and I made it on purpose.

Kindred opens properly on 17 November. Chronological, no ads, no algorithm. Creators earn from a shared subscription pool rather than an ad auction. Family accounts are free.

Get on the list for launch day →

Free to join, no credit card.

Adrian Hallberg is a film photographer and the founder of Kindred Content. He lives full time in an RV, mostly in the western US.

There is a photograph I took in the Utah west desert of a white horse standing side on, looking directly into the lens, with the rest of the herd grazing out of focus behind him and the first light coming across the hills.

If you look at the edge of the frame you can read the markings. KODAK PORTRA 400. And the numbers: 35, then 36.

It is the second to last frame on the roll.

I think about that more than I think about the photograph.

What the number means in practice

Thirty six exposures. That is the whole roll. When you load it in the morning that is your entire budget for whatever the day turns out to be, and there is no way to check as you go.

So when the horse turned and looked at me, I had two frames left.

Not twenty. Not a burst. Two.

I took both. The first one is the one at the top of this piece. The second has his head slightly turned away and it is a worse photograph, and that is the end of the roll, and the moment was over about four seconds after it started.

I did not know I had it until the scans came back several weeks later.

The thing the limit actually does

It would be easy to turn this into a sermon about slowing down and I want to be careful, because the honest version is narrower than the romantic one.

The limit does not make me a better photographer. There are people shooting digital who are far better than I am and the format has nothing to do with it. Anyone claiming film confers taste is selling film.

What the limit does is much more specific: it moves the thinking to before the shutter.

With thirty six frames and no preview, you cannot resolve a composition by trying nine versions and picking the best one later. You have to resolve it standing there, in your head, before committing. Which means you look at the scene for longer, and you notice things during that looking that you would otherwise have discovered in an edit.

That is not a virtue. It is a workflow consequence. But it produces a different relationship with the thing in front of you, and for a subject that is not going to cooperate, it happens to be useful.

Horses do not cooperate

This is the part people underestimate about photographing animals that do not care about you.

You cannot direct them. You cannot ask for another pass. You arrive before light, you park a long way off, you walk in slowly and then you stop walking and you wait, and the herd decides what happens next. Frequently what happens next is that they move away over a rise and you have driven three hours to photograph an empty hillside.

The actual skill is almost entirely waiting, and waiting is not a skill anyone describes as one because it looks exactly like doing nothing.

Most of the hours on any trip like this are spent sitting on cold ground, watching, with the camera on the knee rather than at the eye. The photograph takes a 250th of a second. Everything around it takes days.

And then on that particular morning the light did what it does for about twenty minutes, and a grey turned white by the sun looked up, and I had two frames.

Why I leave the border in

People ask about this and the answer is less philosophical than they expect.

I leave the sprocket holes and the stock markings and the frame numbers visible in a lot of what I post, and the first reason is simply that I like how it looks. It is a strong graphic element and it frames the image in a way a clean crop does not.

But there is a second reason that I only articulated later, which is that the border is evidence.

It says what this is. Portra 400, frame 35, exposed in a camera, developed, scanned. It is not a style applied afterwards to look like something. In a period where any image can be generated or filtered into resembling anything, a visible film edge is a small, checkable claim about how the picture came to exist.

I am not pretending that is unfalsifiable. Somebody could fake it in an afternoon. But I know what it means on mine, and increasingly that feels worth keeping in the frame.

The part that is not transferable

I am suspicious of essays that take a craft practice and extract a life lesson, so I will be careful here too.

The patience thing does not generalize as neatly as it sounds. I am not a patient person in most of my life. I refresh things. I check whether anything happened in the last four minutes, repeatedly, like everyone else. Sitting on a hillside for six hours has not made me calm in any portable way.

What it has done is give me a reliable example of what the other mode feels like. Not a philosophy, a reference point. I know what it is to be somewhere for a long time with no input, waiting for something that might not happen, because I do it a few times a year on purpose.

That is useful mostly as a comparison. It lets me notice when I am in the other state, which is the one where something arrives every two seconds and none of it was asked for.

The frames you do not get

Of the roll that white horse is on, several frames are nothing. Grass, an empty ridge, a horse facing the wrong way, a composition I misjudged in the moment and could not check.

That is normal and it is most of the work. The ratio between what you shoot and what is worth anything is not flattering on any format, and film just makes you look at the bad ones because they are physically there in the strip between the good ones.

I have come to like that. A contact sheet is honest in a way a curated grid is not. The hillside you drove three hours for and got nothing from is sitting right there, two frames before the one that worked, which is roughly the correct proportion.

The photograph on the last two frames is not evidence that I am good at this. It is evidence that I was there, for long enough, on a morning when it happened.

Which is most of it.

Adrian Hallberg is a film photographer and the founder of Kindred Content. He lives full time in an RV, mostly in the western US.

I mention that not to establish credibility but to make the following more embarrassing.

Last year I was recruiting creators to a platform I had built. I sent personal messages, carefully written, to people whose work I genuinely admired. Nineteen of them replied yes. Not polite yes, enthusiastic yes. Several asked follow up questions. A couple said it sounded like exactly what they had been waiting for.

Zero of them signed up.

Not a low conversion rate. Zero.

What I did about it, which was wrong

I rewrote the messaging.

Of course I did. That is the reflex, and it is the reflex because copy is the thing a marketer controls directly. Something is not converting, so you examine the words, because the words are yours.

I rewrote it three times over about three weeks. Shorter version. Version with more specifics about the product. Version that led with the problem rather than the solution. I tested different openings. I tried a version that asked a question and a version that did not.

Reply rates moved a little. Signups stayed at zero.

Which should have told me something much sooner than it did, because when an intervention produces no movement at all, you are usually not intervening on the thing that is broken.

The gap between the yes and the action

Here is what was actually happening, and it took me far too long to see.

The creators were saying yes to an idea. A platform with no algorithm, where their work reaches the people who follow them. That is an easy thing to agree with and agreeing with it costs nothing.

The action I needed from them was completely different. It was: invite your audience to a new platform. Put your name on it. Ask people who trust you to go somewhere unproven.

And at the time, the platform required a subscription to join. So the specific thing I was asking a creator to do was send their audience to a payment screen before those people had seen a single thing.

Nobody wants to do that. Nobody should want to do that. A creator’s relationship with their audience is the asset, and I was asking them to spend some of it on my behalf, at a door that asked for a credit card.

So they said yes to the idea, looked at what the yes actually entailed, and quietly did not do it. Which is not dishonesty. It is the normal gap between agreeing with something and acting on it, and my job was to notice that gap and I did not.

The test I should have run first

What does yes cost them?

That is the whole thing. One question, and I now run it on everything.

A yes that costs nothing predicts almost nothing. Email signups, survey responses, nodding along in a discovery call, anyone saying “that sounds great.” All free, all common, all nearly useless as signal.

A yes that costs something predicts well. Putting their name on it. Spending their credibility. Changing a workflow. Paying, obviously, but also anything that burns time, reputation or social capital.

The mistake I made is the one I had watched clients make repeatedly, which is treating stated intent as a leading indicator when it is mostly a politeness measurement. People tell you what they think you want to hear, especially when they like you, and especially when the thing you built is obviously the product of a lot of work.

Nineteen enthusiastic replies looked like validation. They were nineteen people being nice to someone who had sent them a thoughtful message.

Why marketers reach for copy

I want to defend the reflex a little, because it is not stupid.

Copy is fast. You can rewrite a message in an hour and send it that afternoon. Structural problems take weeks, usually involve engineering, and frequently require convincing somebody who does not report to you.

Copy is also measurable in a satisfying way. Open rates move. Reply rates move. You can show a chart.

And copy is yours. The pricing is not yours, the onboarding flow is not yours, the thing that happens after the click is frequently somebody else’s department entirely. So the marketer optimizes the one surface they own, and sometimes that works, and sometimes it means three weeks of polishing a message that was never the problem.

In my case I owned all of it. There was nobody else to blame, and I still went for the copy first, which tells you how strong the reflex is.

What to check before you touch the words

The order I use now, which would have saved me those three weeks.

One. Follow the actual ask to its end. Not what you are asking in the message, what the person has to do if they say yes. Write out every step. The number of steps is usually higher than you think and the costly one is usually not where you expected.

Two. Find the step that costs the most. Not the hardest step, the most expensive one, where expense includes money, time, risk and reputation. That step is where the drop is, almost always.

Three. Ask whether that cost is removable. Sometimes it is not, and then you need a different audience or a different offer. In my case it was removable, and removing it meant rebuilding the pricing model, which took a significant chunk of a quarter and was obviously correct.

Four. Then look at the copy. It might be fine. Mine was.

The thing that is easy to miss

The nineteen yes replies were not worthless. They were strong evidence that the positioning was right, which is genuinely useful information and the reason I kept going.

They just were not evidence of anything about conversion, and I was reading them as both.

That is the specific error worth taking from this. Different signals tell you different things, and enthusiasm about a concept tells you about the concept. It tells you nothing whatsoever about whether anyone will do the thing, because doing the thing has a price and agreeing does not.

The product now lets people join free. The creator asks their audience to look at something rather than buy something, and that is a completely different request even though the message describing it barely changed.

Same words, roughly. Different door.

Adrian Hallberg is a film photographer and the founder of Kindred Content. He lives full time in an RV, mostly in the western US.

The office is a 45 foot fifth wheel, usually parked somewhere in the western United States with no hookups and no neighbours.

The power comes from about a thousand watts of panels on the roof, through a charge controller, into lithium batteries, out through an inverter. That is the whole system. There is no wall socket behind any of it. When the batteries are empty the laptop does not charge, and no amount of wanting it to does anything.

I have been working this way for a while now and it has changed how I think about building things, in ways that took me a long time to notice.

Power stops being invisible

In a normal office, electricity is infinite. You plug in a thing and the thing works, and the cost is a number on a bill once a month that nobody connects to any particular action.

Out here it is a quantity I can watch.

Running the air conditioner on a hot afternoon is a decision. So is a long compile. So is leaving three monitors on while I make lunch. Every one of those draws against a number that I know, that I can see falling, and that refills only when the sun does.

The effect is not frugality exactly. It is that I developed an intuition for cost where previously there was nothing. I know roughly what things take. And having that intuition for one resource makes you suspicious of every other resource you have been treating as free.

Which turns out to be most of them, in software. Nobody watching a loading spinner is being billed in a way they notice, so nobody optimizes it. The cost is real and it is paid by somebody who cannot see the meter.

The bottleneck is never where you think

Early on I had a heat problem. Things got hot under load in a way that worried me, and my first instinct was that I had too much draw, that the inverter was undersized, that I needed more capacity.

It was the wire.

The cable running between the batteries and the inverter was too thin for the current going through it, so it was dissipating energy as heat instead of delivering it. I replaced it with 4/0 welding cable, which is absurdly thick and feels like overkill in your hands, and the problem went away entirely. Same panels, same batteries, same everything else.

I had been looking at the components. The failure was in the connection between them.

I have since found that to be true of nearly every system I have debugged, including the software ones. The slow part is rarely the thing doing the work. It is the handoff. The query is fine and the serialization is killing you. The function is fast and it is being called four hundred times. The feature works and the thing that tells the user it worked takes six seconds.

Look at the wires first. They are boring and they are usually the answer.

Build for the circuit you have

This one cost me a water pump and about three weeks.

I replaced the fresh water pump with a bigger one. Seven gallons a minute, seventy PSI, far more capable than the old unit. The pressure was excellent. For about three weeks everything was great.

Then it blew a fuse. Then it blew two more in one night.

The problem was not the pump. The pump was fine and the pump was doing exactly what it said on the box. The problem was that the manufacturer specs a 40 amp fuse for it, and the circuit it was wired into is a factory 15 amp run on 14 gauge wire. I had installed a component that was better than the system it was living in, and the system did what systems do when you exceed them.

I could have rewired the circuit. I thought about it for a while. In the end I returned the pump and bought a smaller one sized to the run that already existed, and it has been boring and correct ever since.

That decision is the one I think about most, because the instinct in both hardware and software is always to upgrade the component. More capacity, better specs, the thing with the higher number. And quite often the right move is the less satisfying one, which is to pick the part that fits the system you actually have rather than the system you wish you had built.

Most of what goes wrong in a product is not an underpowered component. It is a well specified part exceeding the thing it was dropped into.

Shipping when connectivity is a variable

The other constraint is the one people ask about more, which is the internet.

It is usually fine and occasionally it is not, and the uncertainty is the actual problem rather than the speed. You cannot plan around a connection that is good eighty percent of the time if you do not know which eighty.

So the workflow changed. Long tasks get queued for when I know I have signal. Anything that needs a reliable connection gets batched rather than done whenever it occurs to me. I pull things down before I need them, because the moment I need them is frequently the moment I do not have them.

None of this is clever. It is just planning around a resource I cannot assume, which is what everyone building for mobile is supposed to be doing anyway and mostly does not, because they are testing on office wifi.

I would say I build more carefully for bad connections than I would if I worked somewhere with fibre. That is probably the clearest product effect of the whole arrangement.

What the view actually changes

Here is the part I expected to be able to say something tidy about, and cannot quite.

I thought the landscape would make me more creative. I do not think that is true, or at least I cannot demonstrate it. What it does is more specific and less romantic.

It makes me bored in a useful way. There is a stretch most evenings where there is nothing to do and nothing on, and that is where most of the thinking I value actually happens. Not at the desk. Standing outside, doing nothing in particular.

Which is, I realize, exactly the state a feed is engineered to eliminate. The gap is the point, and I spend my working hours building something designed to leave the gap alone, which is not a coincidence I planned but is probably not an accident either.

And it gives me a sense of scale that I find clarifying. Being somewhere very large and very indifferent for long stretches does something to how urgent your product decisions feel. Most of them are not urgent. Almost none of them are.

The unromantic part

I should say clearly that a lot of this is maintenance.

Jacks, shocks, fuses, pumps, a solar array that needs looking at, a truck that needs looking at more. There are weeks where the systems take more of my attention than the software does, and the systems are not optional in the way a feature is.

Anyone picturing a laptop and a view should add several hours a week of fixing things that broke because the whole house drives down highways.

I would not trade it. But the version of this life that gets photographed is about four percent of it, which is roughly the same ratio as everything else that gets photographed.

Adrian Hallberg is a film photographer and the founder of Kindred Content. He lives full time in an RV, mostly in the western US.

I found another one last week. Scrolling Threads, of all places.

A European team, chronological feed, no ads, no algorithm, funded by memberships rather than advertising. They had just launched their paid tier. Their marketing was better than mine. Their user count was considerably higher than mine.

My first reaction was the obvious one, a small sinking feeling, and I want to talk about why that reaction was wrong.

Who else is doing this

It is not two of us.

There is Bluesky, which is the large one, built on an open protocol with the explicit promise that you can leave and take your followers. There is Mastodon and the wider fediverse, which has been at this for years and solved the infrastructure problem while never solving the money one. There is Pixelfed for photographs. There are several newsletter platforms that have quietly become social products. There is the European team I found last week. There are, I would guess, a few dozen more that neither of us has heard of, being built in evenings and weekends by people with day jobs.

And there is me, writing code in a fifth wheel somewhere in the western United States.

None of us coordinated. Most of us did not know about each other when we started.

What simultaneous invention usually means

When the same idea occurs to a lot of people at roughly the same time, it is almost never because everyone independently had the same flash of genius. It is because the conditions changed and a number of people noticed.

The conditions here are not subtle.

On a 2024 earnings call, Mark Zuckerberg told investors that for the first time more than half of what people see on Instagram is AI recommended. Not leaked. Said out loud, to shareholders, as a growth metric.

Creators across every platform have spent years watching work get limited or demonetized for being about anything difficult, and have adjusted what they make accordingly, mostly without deciding to.

And the ordinary experience of opening a feed became, for a lot of people, mildly unpleasant in a way that is hard to articulate and easy to feel.

So a dozen people, independently, in different countries, with different skills, arrived at roughly the same conclusion: the thing is broken in a specific way, the fix is structural rather than cosmetic, and nobody large is going to do it because the incentives run the other direction.

That is not a coincidence. That is a lot of people reading the same situation correctly.

Why most of us will fail

I want to be realistic about this, including about myself.

The cold start problem is brutal and it is not a quality filter. A social product needs people in order to be worth joining, which means the first year is spent asking people to join something that is not yet worth joining. Most attempts die there, and they die regardless of whether the software was good.

Bluesky is the instructive case, because it got past the first part and the second part is still hard. They had enormous surges, millions of registrations in days, driven by people leaving somewhere else. By last year their daily active number was a small fraction of their registered one, and had fallen meaningfully from six months earlier.

Arriving and staying are different problems. Solving the first tells you almost nothing about the second.

So the honest expectation is that most of the dozen do not make it. Mine included, possibly. That is what the base rate says and I have no particular reason to think I am exempt.

The failure I actually worry about

Not that I fail. That all of us fail, and the conclusion drawn is that people do not want this.

Because that conclusion would be wrong, and it is the one that gets drawn. A dozen attempts, all small, all eventually folding, and the lesson recorded is that the market spoke and the market likes algorithmic feeds. The next person with the idea gets told it has been tried.

What would actually have happened is that a dozen undercapitalized people each tried to solve a network effects problem alone, which is the hardest problem in consumer software, and none of them had the distribution to get past it.

Those are completely different findings and they look identical from outside.

Which is why I changed my mind about competitors

Here is the part that took me a while.

If somebody else in this category succeeds, that is good for me even if I fail. It establishes that people will pay for a feed, that the subscription math works, that a chronological product can hold an audience. Every one of those is currently an open question, and whoever answers it answers it for all of us.

And if I succeed, every other attempt helped, because collectively we moved the idea from strange to familiar. The second person to explain a chronological ad free feed has an easier job than the first.

We are not competing for a fixed pool. The pool is everyone who is tired, and it is very large, and almost none of them have tried any of us.

The actual competition is not the European team with better marketing. It is inertia, which is undefeated.

What I would want from the others

Mostly one thing: be honest about the numbers.

The temptation in a small category is to inflate, because everyone is looking for signs of life and a good number is the easiest thing to produce. But inflated numbers from a dozen small players produce a false picture of a thriving alternative scene, and when it does not materialize the whole idea takes the reputational hit.

Real numbers, even small and unflattering ones, are more useful to everyone building in this. I have tried to publish mine that way and I intend to keep doing it, including when the launch I am currently planning does not go how I hope.

The honest ending

I would prefer it to be me. Obviously. I have put a year into this and I am not pretending to be above wanting it to work.

But I would take any of us over none of us, and I did not think that two weeks ago. Two weeks ago I found a competitor and felt the small sinking feeling, and the small sinking feeling was a misreading of what I was looking at.

What I was actually looking at was evidence that I am not wrong about the problem.

That is worth more than being the only one who noticed.

Adrian Hallberg is a film photographer and the founder of Kindred Content. He lives full time in an RV, mostly in the western US.

Here is a fact about the thing I built that I have not written about, mostly because it is uncomfortable.

Kindred has a free tier. The free tier is capped at one hour of viewing a week. If you want more than that, you pay.

Read that back slowly against the rest of my output. I have written about attention being extracted, about feeds engineered to hold you, about an industry optimized for time spent. And my own business model is: here is an hour, and if you want more hours, that will be five dollars a month.

I sell screen time. That is what the paid tier is.

I want to deal with that properly rather than hope nobody notices.

The thing that forces the decision

Somebody has to pay for the thing to exist, and there are only two options.

Advertising is off the table, for all the reasons I have spent a year writing about. The moment ad money arrives the customer changes, and every internal question gets answered differently from that day forward.

Which leaves subscriptions, which means there has to be a reason to subscribe. A free tier that is fully unlimited has no upgrade path, and a product with no revenue is a hobby that ends when the founder gets tired.

So the question becomes what to limit, and every answer is bad in a different way.

The options I rejected, and why

Limit who you can follow. Free users get twenty follows, paid users get unlimited. This is the worst one. It directly damages the thing the product is for, which is seeing the people you chose, and it punishes exactly the behaviour I need people to adopt.

Limit posting. Free users can post three times a week. Terrible for a platform desperate for content, and it makes the creators I am recruiting worse off for being new.

Limit quality. Free users get lower resolution. This is the standard move and it is defensible, and I use a version of it for video. But as the primary gate it means the free experience is a degraded version of the work, which for a photography-heavy platform undermines the whole pitch.

Limit features. Private feeds, family accounts, the things that cost real engineering. Except I decided those should be free, because gating family safety behind a paywall is exactly the thing I would criticize somebody else for.

Which left time, because time is the only axis where the limit does not damage the core promise. You still see everyone you follow, in order, at full quality, with everything you posted intact. You just see it for an hour a week unless you pay.

The lie I could tell here

There is a version of this essay where I frame the hour as a wellness feature.

“We cap viewing because we care about your attention.” It writes itself. It would be well received. It fits neatly with everything else on my Medium page and I suspect a reasonable number of people would nod along.

It would also be mostly untrue, and I think the small dishonesty would do more damage than the awkward fact it was covering.

The hour exists because the business needs a reason for people to subscribe. If I were funded another way I would not have put it there. It is a monetization mechanic, and dressing a monetization mechanic in wellness language is precisely the move I have spent a year complaining about when larger companies do it.

So: it is a paywall. It is a paywall placed on the axis that does the least damage, chosen after rejecting several that would have done more. That is the honest description and I would rather give you that one.

What I think actually defends it

Not the framing. The structure underneath.

The paid product still ends. This is the part that matters and it is easy to miss. Paying removes the hour. It does not turn the feed into an infinite one. You still see the people you follow, in order, and then you reach the bottom and you are done.

So the thing being sold is more of a finite thing, not access to an endless one. Compare that to the alternative model, where the free product is unlimited precisely because unlimited is how the revenue works. There, your time is the product. Here it is the price.

That is a real structural difference and it is the whole reason I can sleep at night about this.

The incentive points the right way, or at least not the wrong way. My revenue does not increase if you use the app more. A subscriber at four hours a week and a subscriber at forty pay me the same amount, and the one at forty costs me more in bandwidth. There is no version of my spreadsheet where your attention is an asset to be grown.

That is not virtue. It is just what happens when the money comes from a flat fee rather than an auction, and it is worth more than any promise I could make.

The objection I cannot fully answer

Subscriptions price people out.

An ad funded product is available to somebody with no money. Mine is available for an hour a week, and then it is not. That is a real cost, it falls on people who can least afford it, and the free tier is a partial answer rather than a complete one.

I do not have a clever response. It is a genuine tradeoff and I picked a side. The side I picked means some people get less than they would elsewhere, and in exchange nobody’s attention is being sold to anyone.

Whether that is the right trade probably depends on who you are, which is an unsatisfying conclusion and also, I think, the accurate one.

What I would ask you to do with this

Not to approve of my pricing. It is five dollars and nobody needs an essay to decide about five dollars.

What I would ask is that you run the same test on the products that give you everything for free, because the awkwardness I just walked through is the awkwardness every company in this space has, and most of them resolve it by not mentioning it.

Who pays, and what are they buying?

If the answer is you, the limit is the price and you can see it. If the answer is somebody else, the limit is still there. It is just applied somewhere you cannot see, to something you did not know was for sale.

I would rather have the version you can argue with.

Adrian Hallberg is a film photographer and the founder of Kindred Content. He lives full time in an RV, mostly in the western US.

Everyone reads privacy policies wrong, on the rare occasions they read them at all.

The instinct is to hunt for the bad clause. The buried sentence about third party partners, the vague bit about affiliates, the line that says we may share aggregated information. People screenshot those and post them, and the screenshots are usually accurate.

But they are a symptom, and treating them as the disease leads nowhere. Companies do not write permissive privacy policies because their lawyers are devious. They write them because the policy has to permit whatever the business actually does, and the business was decided years before anyone drafted a word of it.

The policy is downstream. Always.

The two possible answers

Any company serving you content has answered one question: who pays?

There are exactly two answers. Either you pay, or somebody pays for access to you. There has never been a third, and the word “free” just means the second one.

If the revenue is advertising, then the product being sold is targeted access to a person. Which requires knowing things about that person. Which requires collecting, retaining, inferring, segmenting and sharing. Not as an abuse of the model, as the model. A privacy policy for an ad funded product has to permit all of that, because without it there is no company.

If the revenue is subscription, then the product is the thing itself. Knowing your politics or your late night browsing has no revenue attached. It is a cost, a liability, and a breach risk with no upside.

So the privacy difference between those two companies is not a difference in ethics. It is a difference in what the finance team needs in order to make the quarter.

Why this is more useful than reading the policy

Policies change. The income statement predicts the policy.

If you know a company funds itself with ads, you can predict the shape of its data practices without reading anything, and you will be approximately right. If a company funded by ads publishes a privacy focused announcement, you can predict that the announcement describes a narrowing rather than a reversal, and you will be approximately right about that too.

The useful question is not what does the policy say. It is what would this company have to stop doing to keep the promise it is making?

If the answer is “nothing,” the promise is cheap and probably real. If the answer is “most of its revenue,” the promise has a shelf life measured in however long the current growth story holds.

The honest complications

Three, and they all cut against the neat version of this argument.

Subscription companies have sold data. Plenty of them. Paying does not create an obligation that a company cannot decide to ignore when a new revenue line looks attractive. The structural incentive is better, which is not the same as a guarantee, and anyone claiming otherwise is selling something.

Ad funded does not mean unprincipled. There are people inside those companies fighting hard for better defaults, and some of them win. The incentive points one way, it does not fully determine every outcome, and pretending it does is lazy.

Subscription has its own failure mode. It prices people out. An ad funded product is available to someone with no money, and a subscription one is not. That is a real cost and a real tradeoff, and the version of this argument that ignores it is incomplete. Free tiers help. They do not fully solve it.

So the claim is narrower than it first sounds: the business model tells you which direction the pressure points. It does not tell you what any particular company will do on any particular Tuesday.

What this looks like in practice

Try it on something you use.

Ask what the company sells. Then ask what it would need to know about you in order to sell that thing well. Then ask whether the privacy policy permits exactly that, which it will.

The exercise usually takes about ninety seconds and it is more informative than reading twelve pages of legal text, because the legal text is the output of the process you just reconstructed.

It also explains why the settings never quite fix it. A privacy setting can only change things that are not load bearing. Anything the revenue genuinely requires will be either absent from the settings, on by default, or framed as something you would be strange to turn off.

My own version of this, including the uncomfortable part

I built a platform, so it is only fair to run the same test on it.

It is called Kindred. The revenue is subscriptions. Which means the honest version of what gets collected is short: an email address so you can log in, a date of birth for the age gate, payments handled by Stripe, and what you watched and for how long.

No advertising profile, because there is no ad product. Nothing sold to anyone, because there is nobody buying. No behavioural model of you, because it would cost money to build and earn nothing.

And here is the part I would rather not have to include, which is exactly why it belongs: watch time is measured. How long you spent on a given video. That is real data about a real behaviour, and I am not going to describe this as a zero data product, because it is not one.

The reason it exists is that creator payouts are calculated from it. A pool of subscription revenue has to get split somehow, and the split is based on what people actually watched. So the thing I measure is the thing that decides who gets paid.

That is a justification, not an absolution. A reasonable person can look at that and decide it is more than they want measured, and they would not be wrong to. What I can say is that it is used for one purpose, it is sold to nobody, and there is no commercial reason for it to ever become anything else.

Which is the whole argument, applied to me: ask what I would have to stop doing to keep the promise. The answer is paying creators accurately. That is a constraint I am stuck with rather than a virtue I chose, and that is generally the more reliable kind.

The thing to take away

Not that subscriptions are good and advertising is evil. That is a slogan and the world is messier.

The thing worth taking is the test itself, because it works on anything and it costs you ninety seconds.

Who pays, what do they buy, and what does the company need to know about you to sell it?

Everything downstream, the policy, the settings, the announcements, the apology posts, follows from the answer. And you will usually find that nobody lied to you anywhere. They just told you, in twelve pages, what the business model had already decided.

Adrian Hallberg is a film photographer and the founder of Kindred Content. He lives full time in an RV, mostly in the western US.

I spent a week in the Faroe Islands with a medium format camera and came home with something like seventeen rolls.

I posted four frames.

That is not a complaint about anything and it is not unusual. Ask any photographer what ratio of their work has ever been seen by another person and the honest answer is dismal. We make far more than we show, and we have good reasons, and most of those reasons dissolve the moment you look at them properly.

The arithmetic of posting to everyone

Here is the thing that governs the decision and almost nobody says out loud.

When you post to a public feed, you are posting to a group with no shared context. Some of them know you. Most do not. Some of them were there. Almost none were. Some care about the specific thing, and the overwhelming majority are passing through on their way somewhere else.

So you edit for the median member of that group, which means you edit for a stranger with no context and four seconds of attention.

That edit kills certain photographs immediately. The one that only means something if you know what the weather had been doing for three days. The sequence of six nearly identical frames where the difference is the whole point. The portrait of the person you were travelling with, doing nothing in particular. The failure that is interesting because it failed.

All of those get cut, not because they are bad, but because they do not survive contact with an audience that does not know you.

What gets posted instead

The three or four that work cold.

The dramatic one. The one where the light is obviously good. The one that reads at thumbnail size, which is the actual constraint even though nobody writes it down.

I am not above this and I have the receipts. The four Faroes frames I posted were the four most immediately legible ones, and I knew that while choosing them. The roll has better photographs on it. They are just worse at being seen by strangers, which is a different quality entirely and one we have quietly started treating as the same thing.

The caption problem

Then you write the caption, and the caption has the same disease.

You are explaining context to people who were not there, which means the people who were there read a description of a thing they experienced, written for an audience of strangers, slightly performed.

Anyone who has posted about something that mattered knows that particular discomfort. You want to say what actually happened and instead you write something that works for everyone, which means it works properly for nobody. The twelve people who were present get the public version, the same as the three thousand who were not.

That is a strange thing to have accepted.

Broadcasting replaced sharing

These were two different activities and the distinction collapsed so gradually that most of us never clocked it.

Sharing is showing a specific thing to specific people who have reason to care. A photograph to the person in it. A roll to the friend who was on the trip. Twenty frames to your parents, unedited, because they want to see the trip rather than the portfolio.

Broadcasting is publishing to an audience, which requires editing for the audience.

Both are legitimate. The problem is that the tools for the second got extremely good and the tools for the first mostly did not, so everything migrated. Now the default move with any photograph is to put it in front of everyone or nobody, and the middle option, the twelve people, fell out of use.

You can see the gap in how people improvise around it. The group chat with four relatives in it. The shared album that someone sets up once and nobody ever opens. The text message with six photographs attached that lands as six separate notifications. Those are all workarounds for a thing that should have been a product.

What small actually changes

I started sending whole rolls to a handful of people and the difference was immediate and not subtle.

I stopped editing for strangers. The mediocre frames went in. The six near identical ones went in, in sequence, which is how you actually see what a photographer was doing. The failures went in and the failures got the most conversation.

The captions became sentences rather than copy. “This is the morning it would not stop raining” is useless publicly and perfect for someone who was standing next to you.

The replies changed shape. Not likes. Actual remarks, from people with context, sometimes a week later. Somebody noticing the thing in the corner of frame nine.

None of that is available in a public feed and it is not the feed’s fault. It is just what happens when the audience is twelve people who know what they are looking at.

The obvious objection

Why not just use a group chat.

Honestly, for a lot of this, you should. A group chat is a perfectly good small-group sharing tool and I am not going to pretend otherwise.

Where it falls down is scale and shape. Forty photographs in a chat is a wall of notifications and a thread nobody can navigate a month later. Video is worse. There is no sense of a body of work, no way to come back and look at the trip as a trip, and anything older than a few weeks is effectively gone.

The group chat is great for one photograph and bad for a roll. That gap is the whole opportunity, and most of what gets built in it is aimed at families with babies, which is a real use case but a narrow reading of the problem.

The disclosure

I built a platform, so discount what follows.

It is called Kindred, and one of the things in it is private feeds. Invite only, a small number of people, no ads, nothing ranked, nothing public. You post the whole roll to eleven people and that is where it stays.

I built it because I wanted it, which is the only honest reason I have. Seventeen rolls from the Faroes and four public frames was the thing that bothered me into doing something.

But the reason I wrote this is not the feature. It is the question underneath, which applies whether or not you ever use anything I made.

What have you not shown anyone, because the only options were everyone or nobody?

Most photographers have an entire body of work that answers that, sitting on a drive. So do most people who have ever taken a phone out at something that mattered.

Adrian Hallberg is a film photographer and the founder of Kindred Content. He lives full time in an RV, mostly in the western US.

There is a shelf at the front of every decent record shop with handwritten cards on it. Staff picks. Someone who works there listened to a thing, liked it, and wrote two sentences about why.

Mubi built a streaming business on roughly that idea. A small number of films, chosen by people, with the explicit position that this is the opposite of an algorithm. Criterion does a version of it. So does any good bookshop, any magazine with a decent reviews section, any museum deciding what to hang.

All of those are somebody else choosing what you see.

So is a recommendation feed.

The question worth asking, and I do not think it gets asked carefully enough, is what the actual difference is. Because people who loathe algorithmic feeds will happily take a staff pick, and that is not hypocrisy. There is a real distinction underneath it.

Four differences that actually matter

Who it is optimized for. The record shop employee is trying to make you like the record. A ranking system is trying to extend a session. Those produce different selections, and the divergence compounds over years. One of them will eventually recommend something difficult that takes three listens. The other, structurally, cannot.

Whether you can see the reasoning. The staff pick has two sentences on a card. You can read them, disagree, and calibrate how much you trust that person’s taste going forward. A feed gives you no reasoning at all, and the modelled version of you doing the choosing is not available for inspection.

Whether anyone is accountable. The card has a name on it. If that person recommends four duds in a row you stop reading their cards. There is a feedback loop and a reputation at stake. Nobody is accountable for a ranking decision, and there is no mechanism by which bad taste costs the system anything.

Whether it is the same for everyone. This one is underrated. The staff pick is public, singular, and identical for every person who walks in. Which means it can be argued about. Two people can disagree about it because they saw the same thing. Personalized feeds eliminate the shared object entirely, so there is nothing to discuss.

What we actually threw out

Somewhere in the last fifteen years, curation got filed under gatekeeping and discarded along with it.

There was a reasonable version of that argument. Gatekeeping was real. A small number of people decided which records got pressed, which books got published, which films got distributed, and their taste was narrow in ways that excluded enormous amounts of good work and whole categories of people. Getting past them was genuinely good.

But gatekeeping and curation are different operations that happened to be performed by the same people.

Gatekeeping is deciding what is allowed to exist. Curation is deciding what to point at, after everything already exists. The first is a barrier. The second is a service.

We removed the barrier, which was correct, and we lost the service in the same motion, which was not. Then, with no editors left, the volume became unmanageable, and we handed the pointing job to a system that was never trying to point at good things.

The honest case against curation

I do not want to be sentimental about this, because curation has real failure modes.

Taste monoculture. A single curator with a strong aesthetic will produce a narrow catalogue, and narrow in a different way than an algorithm is narrow. Every Mubi-style service eventually gets accused of having one mood.

Who gets to pick. The reason gatekeeping was bad is that the gatekeepers were drawn from a thin slice of the population. A curator is a gatekeeper who has given up the ability to block, but they still have the ability to ignore, and the composition of who curates still shapes what surfaces.

It does not scale, and that is not a bug but it is a limit. A person can meaningfully curate dozens of things, maybe hundreds. Not millions. Any system covering everything will be automated, which means curation can only ever be a layer, never the whole answer.

Anyone selling human curation as a complete replacement for ranking is overselling. It is a different tool with different failure modes, and the honest pitch is that those failure modes are more legible and more arguable than the ones we have now.

What I would actually want

A clean separation between two things that currently get conflated.

The feed should be yours. Whatever you subscribed to, followed, or chose, in the order it arrived, with nobody in between. No ranking, no insertion, no suggestions. This is your mail, not a magazine.

Curation should be a different room you walk into. Clearly labelled, obviously edited, with a name attached. A page that says here are twelve things worth your time this month and here is why. You go there on purpose, and you go there knowing somebody chose.

The failure of the modern feed is not that it contains recommendations. It is that the recommendations were mixed into the mail and the two became indistinguishable. You can no longer tell what you asked for from what you were handed, which means you cannot evaluate either.

Keeping them in separate rooms solves most of it. One is yours. The other is somebody’s opinion, offered as an opinion.

The disclosure, and the tension in it

I built a platform, so discount accordingly. It is called Kindred, the feed is chronological, there is no ranking step anywhere in it.

And that creates a genuine tension with everything I just wrote, which I would rather name than paper over.

My whole pitch is that nobody chooses for you. If I then add an editorial section where I choose things, I have introduced exactly the thing I said I removed, with myself as the ranking function. A reasonable person could call that hypocrisy.

My answer, for now, is the separate rooms rule. The feed never gets touched. Nothing is inserted, nothing is ranked, nothing I like gets a boost. If curation exists, it lives somewhere you have to deliberately navigate to, with my name on it, and you can ignore it permanently with no cost.

Whether I hold that line is a fair thing to watch rather than take on trust. Every platform that ended up ranking everything started by promising the recommendations would stay in their own section.

So do not take my word for it. The useful habit, with any product including mine, is to ask which room you are standing in, and whether anyone told you that you had walked into a different one.

Adrian Hallberg is a film photographer and the founder of Kindred Content. He lives full time in an RV, mostly in the western US.

Ask any YouTuber who has been at it more than a couple of years about the yellow icon.

They will know immediately what you mean. It appears next to a video in the dashboard and it means limited or no ads. Sometimes it resolves after a review. Often it does not. There is an appeals process, and the appeals process works roughly as well as you would expect for a system operating at that scale.

What matters is not any individual decision. It is what happens to a person who has seen it a few dozen times.

The list, roughly

YouTube publishes advertiser friendly guidelines, and they are not secret or unreasonable on their face. The broad categories are the ones you would guess: strong language, violence, adult content, shocking material, harmful or dangerous acts, drugs, firearms, controversial issues, sensitive events.

Read them as a creator and the practical translation is quick.

A documentary about addiction is a problem. War is a problem. A frank piece about a health diagnosis is a problem. Grief, depending how it is handled. Anything about a recent tragedy. Politics, obviously. True crime, which is an enormous and genuinely popular category, sits permanently under a cloud. And swearing, which for a lot of people is simply how they talk.

None of that is banned. It is all still up there. It just earns a fraction of what the same creator’s lighter work earns, or nothing.

What that does over a decade

Here is the part I find most interesting, and it is not the direct effect.

A creator makes a video about something difficult. It gets limited. They make another one, it gets limited. The third time, they are aware while filming that this one is probably going to get the icon.

And then, eventually, they stop pitching those ideas to themselves.

Not as a decision. Nobody sits down and writes “stop making serious work” in a notebook. It is slower than that. The idea arrives, some part of your brain does the arithmetic on whether it will earn, and the idea quietly does not get written down.

I know this happens because a smaller version of it happened to me with photography. I shoot long exposures, patient stuff, the kind of frame that takes an hour and sometimes a second trip. It never performed. And over about two years I drifted toward shorter, punchier work that travelled better and meant less to me, without ever once deciding to.

Nobody offered me anything. There was no moment of selling out. There was just a reward signal, applied consistently, for long enough.

Why this is harder to argue with than censorship

If a platform removed these videos, there would be an obvious fight and everyone would know how to have it. Free expression, platform responsibility, the usual positions.

But nothing is removed. The video is right there, fully public, watchable by anyone. What changed is whether making it is economically survivable for the person who made it.

That is a much quieter intervention and it is far more effective, because it operates before the work exists rather than after. You cannot protest a video that never got made, and the creator who did not make it frequently does not realize they did not make it.

The aggregate effect, across a million channels and a decade, is that an entire register of work simply thins out. Not banned. Just not funded, which over time amounts to something similar.

The honest counterargument

Advertisers have every right to choose where their money goes.

A soap company does not want its spot running against a video about a mass shooting, and it is not obviously wrong about that. The brand is not obliged to fund anything. Nobody has a right to an advertiser’s money.

I actually agree with all of that, and I want to be clear that I do, because the version of this argument that treats advertisers as villains is both unfair and analytically lazy.

The problem is not that advertisers have preferences. It is that advertising became the only funding mechanism, so advertiser preferences became the only filter. If there were five ways for a creator to get paid and one of them was ad supported, the yellow icon would be an inconvenience. Because it is effectively the only way at scale, it is a gate.

What this looks like from the inside

You can watch it in how creators talk now.

The pauses before certain words. The careful phrasing around anything medical. “Unalive” and similar euphemisms, which started as a way to get past an automated filter and have now become actual vocabulary for an entire generation.

The self censorship is so normalized that most of the people doing it would not describe it as self censorship. It is just how you talk in a video, in the same way you dress differently for a job interview.

That is the aesthetic. Advertiser friendly is not a restriction any more. It is a style, with a pacing and a vocabulary and a set of topics, and most viewers now experience it as what online video simply sounds like.

What a different funding source changes

Here is the thing I have become convinced of: this is not a moderation problem or a policy problem. It is a funding problem, and it resolves differently depending on who is paying.

If the money comes from advertisers, the filter is whatever advertisers will tolerate next to their logo. That filter will always exclude difficulty, because difficulty is a risk to a brand and essentially never an asset.

If the money comes from the audience, the filter is whatever that audience will pay for. Which is a different set, and importantly it includes the hard stuff. People pay for documentaries. People subscribe to journalism about war. The entire history of patronage and subscription suggests that audiences will fund work advertisers would not touch.

This is not a claim that subscription models are morally superior. It is a claim about what gets filtered, which is a structural question with a structural answer.

The disclosure

I built a platform, which means I have an obvious interest in you finding this persuasive.

It is called Kindred, there is no advertising product on it at all, and creators earn from a shared pool of subscription revenue. Which means nothing on it can be demonetized for being quiet, dark, sweary or unbrandable, because there is no brand to be safe for.

It is small. The pool is small. I am not going to pretend a new platform solves an industry wide incentive problem.

But the specific thing I would like a creator reading this to take away is not “come to Kindred.” It is the question underneath: what have you not made in the last three years because of a calculation you ran without noticing you were running it?

Most people who answer that honestly find something.

Adrian Hallberg is a film photographer and the founder of Kindred Content. He lives full time in an RV, mostly in the western US.