Zibalion

Rodney Zadeh

Operator. Builder. Writer.

Technology / Systems / Axioms

Every system is a covenant nobody signed. Its terms are written in architecture rather than in language, which is why they are so seldom disputed and so widely obeyed.

I read them. I establish what a system assumes, where it places its costs, and what its makers declined to account for. Then I build, because a claim about how something works is worth no more than the attempt to make it work.

Hardware, software, and the reasoning beneath both.

Rodney Zadeh

Recent writing

  1. The title of your first piece
  2. More coming
All writing

Building

The ledger

Systems whose growth now costs the people inside them. The number, the mechanism, and what a fix would actually require.

  • Swipe dating $6.07B category revenue — first decline on record · downloads down six straight years · Tinder monthly actives −7% · Bumble paying users −16.4%

    Read the revenue model plainly: the company is paid for every month you remain single. A subscriber who meets someone in week two is a lost account; a subscriber who searches for three years is the ideal customer. That is not a bug that crept in — it is the position the business takes on your behalf, every month, in the opposite direction from the thing you paid for. A decade of growth came from widening the pool, and the first revenue decline in the category’s history came from the same place. Operators are now pivoting to showing fewer profiles and getting people offline faster, which is an admission rather than a strategy. An honest fix prices the exit — charges once, at success, and lets you go. Nobody has been willing to.

  • Adolescent attention 48% of US teens say social media is mostly negative for people their age, up from 32% in 2022 · 44% have tried to cut back · 3+ hours daily associated with roughly double the risk of depression and anxiety symptoms

    The people closest to the product have turned against it, and they can articulate why. Nearly half are trying to quit something they were never old enough to consent to. Age gates treat this as an access problem; it is a design problem, and the distinction is the entire fight. Four families sued the major platforms this July on exactly that theory.

  • Sports betting 27% of American adults hold an online sportsbook account — 52% of men 18–49 · ~60% report chasing losses · one operator drew 70% of profit from under 1% of users

    That last number is the whole business. A product that earns seven-tenths of its money from one user in a hundred is not entertainment with a minority of casualties; it is a machine whose economics require them. North Carolina’s helpline calls more than tripled as the apps went live, and the average caller got five years younger. The fix is a design that can distinguish a bettor from a person in trouble — and is obligated to act when it does.

  • Frictionless spending 68% of buy-now-pay-later users say it made them overspend · 54% regret a purchase · 47% paid late in the past year · 29% have financed groceries, up from 14% two years ago · 11% triggered an overdraft

    Retailers adopted installment checkout because it raises cart size. That is the entire finding, stated by the industry itself. The checkout used to be the last moment a person could reconsider; the product removed it and called the removal convenience. And when a third of users are financing food, the story is no longer impulse — it is that the pressure was already there and someone built a frictionless way to borrow against it. A fix would put the friction back at the moment of purchase and take it out of the moment of repayment. Every incentive runs the other way.

  • Machine companionship 64% of US teens now use AI chatbots · roughly 3 in 10 use one daily — the first year this has been measured at all

    A behavior went from unmeasured to near-majority inside a single survey cycle. Whatever these are substituting for, the substitution is happening faster than anyone can study it, and the products have every incentive to be more available than the people they stand in for. Availability is not the same as sufficiency. The fix starts with building things that hand you back to other humans instead of holding you.

  • Ad-funded search The answer arrives before the click. The index was paid for by the click.

    Everyone can do the arithmetic; nobody has named the thing that replaces it. So the web gets quietly strip-mined while the question stays open, and the people who wrote the pages absorb the cost of a transition they were not consulted about.

  • Subscription capture Ownership replaced by rent, cancellation built as a maze.

    Renting software is a reasonable trade. Making the exit expensive in time rather than money is not a business model, it is a toll — and it is charged most heavily to the people with the least time to spend fighting it.

  • Gig classification Flexibility offered as a favor. Risk relocated to the person least able to carry it.

    The public argument is always about employment status, because employment status is arguable. The real question is who holds the downside when the work stops, and that question never makes it onto the ballot.

Counterexamples

Built on the opposite incentive, and still standing. Proof the ledger above describes choices, not laws of nature.

  • Kagi Subscription search · no ads, no tracking, no free tier · ~50,000 paying subscribers · public-benefit charter

    Answers ad-funded search. Charging money instead of attention sounds obvious and is nearly impossible — Neeva tried the same model, ran four years and folded. Kagi’s bet is that the audience for search that isn’t optimized against you is small but real. Small and real is a business. It just isn’t a monopoly, which is why almost nobody attempts it.

  • Bandcamp ~82% of each sale to the artist · $1.68B paid out to date · paid within 24–48 hours · artists keep their rights

    Answers creator economics. A $10 album purchase nets an artist roughly what two thousand streams would. The mechanism is unglamorous: sell the thing, pay quickly, take a stated cut. Worth noting honestly — it was acquired and cut staff twice since, and the artist-first terms survived those cuts so far. Good design is not the same as a guarantee.

  • Framework 10/10 iFixit repairability · every part sold direct — battery $49, screen $179 · mainboard upgrades into the same chassis

    Answers ownership replaced by rent. A broken USB-C port costs fifty dollars here and fifteen hundred in a soldered machine. That gap is not a manufacturing constraint, it is a decision about who absorbs failure. Framework decided the owner should be able to, and then had to prove a company could survive making that choice.

  • Signal Nonprofit foundation · donation funded · no ads, no data collected to sell · protocol published and independently audited

    Answers surveillance as the default price. Signal’s structural move was legal, not technical: a foundation cannot be sold, so there is no exit event that turns the users into inventory. Most privacy promises are a policy. This one is a corporate structure, which is the only kind that survives a change of management.

  • DROP One request deletes your data across 500+ registered brokers · live for consumers since Jan 1, 2026 · brokers legally required to process since Aug 1 · $200 per request, per day, for ignoring one

    Answers the data you never handed over. Deletion rights already existed and were useless, because exercising them meant contacting hundreds of companies individually — a right that costs more to use than it returns is not a right. California built the one thing that was missing: a single button, and a fine large enough that ignoring it is more expensive than complying. Note what it does not cover: brokers only, registered ones, California residents. It is a working mechanism with a narrow blast radius, which is still infinitely more than a principle with none.

  • Wikipedia Twenty-five years · no advertising · no engagement metric · no recommendation feed · among the most visited sites on earth

    Answers everything. By every assumption of the modern web it should not exist, and it has outlasted most of the things built on those assumptions. The lesson is not that volunteers are free. It is that a product with no incentive to hold you is one you can actually leave — and people come back to what they are allowed to leave.

  • The fediverse Chronological timelines · no engagement ranking · portable identity across servers · no central owner

    Answers attention feeds. Slower, smaller, and harder to explain than the alternatives — which is the point. Remove the optimization and you remove the growth; nobody has yet shown you can keep one without the other. That is an unsolved problem, not a solved one, and it belongs on this list precisely because it is honest about that.

Where the money is going

Capital is already moving toward these problems. Whether it is moving toward solutions is a separate question.

  • Wellbeing, funded on engagement $3.6B into wellness startups in H1 2026, pacing a third above 2025 · investors asking for retention curves, daily-open rates, session data

    Read the ask carefully. The money flowing toward attention and mental health is being underwritten by the same metric that caused the damage: build something people open every day, because daily opens compound. A remedy financed on the disease’s business model is not a remedy. It is the disease with better copy, and it will be measured by how long it holds you.

  • The subscription fork Perplexity removed advertising entirely in early 2026, committing to subscription tiers · Anthropic states no plans for ads · OpenAI began testing ads on free tiers, February 2026

    The largest open question in consumer software is being settled right now, in public, by four companies choosing differently. If the ad tests hold, the attention model walks straight back into the centre of AI and paid stays a niche. If they corrode trust, subscription gets its tailwind. Nobody knows yet — which is exactly why it is worth watching rather than predicting.

  • Repair, by mandate EU right-to-repair directive: seven-to-ten-year parts support, restrictions on software locks · compliance deadline July 31, 2026

    The one item on this page where the fix arrived as law rather than as a market. Worth noting what that implies: on repairability, nobody expected the incentive to correct itself, and it didn’t. Regulation is what capital allocation looks like when the covenant is finally written down and signed.