What is enshittification?

Cory Doctorow's name for the predictable four-stage decay of every platform business. Stage one: be good to users to acquire them. Stage two: be good to business customers to extract value from users. Stage three: be good to shareholders by extracting value from business customers. Stage four: die. The mechanism is structural — switching costs grow at each stage, the platform tilts the playing field a little further toward extraction, and the asymmetric power of the platform over both sides of the market makes the decay nearly inevitable.

Why is enshittification different from ordinary corporate decline?

Ordinary corporate decline happens when a company loses to better competitors. Enshittification happens even when a company faces no competition — in fact, especially then. The mechanism is that platform businesses sit between two or more groups (users and advertisers, drivers and riders, sellers and buyers) and progressively redistribute the surplus among them, taking a larger share each cycle. Switching costs make exit expensive for everyone, regulation lags the practice, and what Doctorow calls 'twiddling' — fine-grained algorithmic adjustment — lets the platform extract right up to the point where users would leave, and then a little further.

What is adversarial interoperability?

Doctorow's preferred policy lever — the right to repair, modify, reverse-engineer, and interoperate with services and products without the original vendor's permission. Adversarial interoperability raises the cost of enshittification by letting users and competitors route around the lock-in mechanisms platforms depend on. It is the legal-engineering counter-move to the platform's structural advantage, and it has substantial historical precedent: the entire history of the open web ran on adversarial interoperability before it was named.

How does the Techno-Memetic Commons answer enshittification?

By licensing reciprocity into the substrate itself. The TMC licence triggers a copyleft obligation after 45 days of any closed internal use of the commons, and requires a publicly visible Stewardship Mark for users who depend on the commons commercially. This makes the closed-source enclosure that enshittification depends on structurally illegal under the commons license, while permitting the legitimate commercial use that funds the commons. The federated unicorn architecture goes further — distributing the platform layer across 10,000 federated proprietors instead of concentrating it in one shareholder-owned firm. Anti-rivalrous substrate plus distributed ownership is what enshittification cannot uniformly eat.

Enshittification.

The four-stage decay of every platform. Once you see the pattern, you cannot un-see it — and you have probably watched it happen to every product you used to like.

Codex · Western Canon · ≈8 min read · Doctorow, 2022 · ADS Word of the Year, 2023

TL;DR

First they are good to users. Then they are good to business customers at the expense of users. Then they are good to shareholders at the expense of business customers. Then they die. Cory Doctorow named the pattern in 2022; the American Dialect Society made it 2023's Word of the Year. The decay is structural, not malicious, and the only way to route around it is to refuse the substrate it requires — switching costs, asymmetric power, closed-source enclosure. That is the substrate the Techno-Memetic Commons and the federated unicorn are deliberately designed to refuse.

The essay heard around the internet

January 2023. The science-fiction novelist and activist Cory Doctorow publishes a post on his blog with a title designed to be quoted: Tiktok's enshittification. The argument is half a page long, the new word does most of the work, and the timing is exactly right — every literate person has just lived through Twitter's collapse under Elon Musk and is watching Reddit start the same arc. The post goes everywhere. By year-end the American Dialect Society has crowned enshittification Word of the Year. By 2025 it is in the Oxford English Dictionary.

The word stuck because the pattern is exact. Doctorow gave the diagnostic its single most quotable formulation:

Here is how platforms die: first, they are good to their users; then they abuse their users to make things better for their business customers; finally, they abuse those business customers to claw back all the value for themselves. Then, they die. — Cory Doctorow, 2023

Every well-quoted account of why your favourite product got worse — Facebook, Google Search, Amazon, Twitter, Instagram, Reddit, Uber, Airbnb, DoorDash, every dating app — is a particular instance of this general curve. Once you have the word, you find you have been seeing the pattern for years without a name for it.

The mechanism — why it happens even without malice

Doctorow's deeper claim is that the decay is structural. It is not principally caused by bad leadership, greed, or a particular CEO's character — though those certainly accelerate it. It is caused by the structural position of the platform between two or more groups that have come to depend on it. Three forces compound:

Switching costs. Every successful platform raises the cost of leaving for everyone connected to it. Users have built up profiles, contacts, photos, reputations. Business customers have built up audiences, listings, supply chains, reviews. The longer the platform has run, the higher these costs are — and the more value the platform can extract before users actually leave.
Twiddling. Doctorow's term for the platform's ability to make tiny, continuous, opaque adjustments to who sees what, who is charged what, who is ranked where. Algorithmic feed ordering, search-result ranking, dynamic pricing, ad load — every dial can be tuned a percentage point at a time, with no visible commitment to any settings. The platform extracts right up to the point where users would visibly notice, and then a little further once it learns where that line is.
Asymmetric power. The platform sees everything; users and business customers see only their own data. The platform can run hundreds of A/B tests a week; the user has one experience and is not sure if it is degraded compared to what someone else is being served. The information asymmetry is the structural inheritance of two-sided markets.

Put these together and the gradient is one-way. Within the platform's normal operating logic, extracting more from locked-in users is the rational move. The shareholders will reward it. The competitors cannot match it because the competitors would have to assemble the same lock-in first. The regulators will be three to five years behind. The users will leave eventually, but they will leave slowly, and by then the platform will have extracted the value it was built to extract.

The decay is not a bug. It is the substrate revealing what it was always going to do.

The four stages, in slow motion

Stage	Who is the customer	What it feels like	What it actually is
1. Acquisition	Users	"This is amazing. It is free / cheap / better than what I had."	VC subsidy is buying your attention, your contacts, your routines. The unit economics do not yet work; they are not supposed to yet.
2. Tilt	Business customers (advertisers, sellers, drivers)	"Hmm, more ads. The feed is weirder. Search is worse. But everyone is still here."	The platform is now monetising the locked-in user base by selling reach, ranking, or access to business customers — and quietly tilting the experience to maximise that revenue.
3. Squeeze	Shareholders	"This product is unrecognisable. Why is everything broken?"	Now the business customers are locked in too. The platform extracts harder from both sides — higher commissions, lower organic reach, more friction, more upsells.
4. Death	No one in particular	"Remember when this was good?"	The platform has extracted everything possible. Quality is so degraded that the social cost of leaving is finally lower than the cost of staying. The migration happens. The carcass continues to extract from the holdouts for years.

Stage four is not always a literal corporate death. Sometimes it is a long zombie phase, the platform extracting decreasing value from a shrinking base. Sometimes it is a fire-sale acquisition. Sometimes it is a regulatory intervention that breaks the lock-in. The dynamic, however, is the same: the platform built on extracting value from connected groups eventually runs out of value to extract, and the connected groups disperse to the next platform — which begins the cycle again.

What makes a platform vulnerable

Not all software businesses enshittify on the same schedule. Doctorow's analysis surfaces the conditions that accelerate the decay, and the ones that resist it:

Network effects + switching costs are accelerants. The bigger the network you would have to abandon, the more the platform can extract.
Algorithmic mediation is an accelerant. The more the user's experience is shaped by opaque ranking, the more space the platform has to twiddle without visible commitment.
Closed-source enclosure is an accelerant. If the platform is the only one who knows how the ranking works, no competitor can offer the same experience without it.
Anti-interoperability law (DMCA §1201, CFAA, terms-of-service-as-felony interpretations) is an accelerant. If reverse-engineering your way out is illegal, the lock-in becomes a legal moat as well as a technical one.
VC-shareholder capital structure is an accelerant. Capital that needs 30× returns on a 7-year horizon will reward the squeeze stage.

The mirror image — what resists enshittification — is where the operational response lives:

Open protocols (SMTP, RSS, ActivityPub, IRC, BitTorrent) are structurally resistant because no single party controls the substrate. The user's contacts move with them; the publisher's audience moves with them; the lock-in is shallow.
Federation (Mastodon, Matrix, the Fediverse, peer-to-peer rails) splits the platform across many independently governed nodes, each too small to extract at scale without losing its node-mates.
Patient capital with realistic return horizons removes the squeeze gradient. Cooperatives, member-owned firms, mission-locked corporations.
Commons licensing with reciprocity triggers makes the closed-source enclosure that enshittification depends on legally hostile to the commons. See Techno-Memetic Commons.
Adversarial interoperability rights — the legal capacity to reverse-engineer, scrape, port out, and connect across without vendor permission — raises the cost of every other accelerant.

Adversarial interoperability — the actual policy ask

Doctorow's most consistent operational ask is adversarial interoperability: the legally protected right to repair, modify, reverse-engineer, and interoperate with products and services without the original vendor's permission. The point is not to romanticise hackers. The point is that adversarial interoperability is what historically kept the platform layer competitive in the pre-DMCA era — and what its erosion has structurally enabled the contemporary platform's lock-in.

The history is on Doctorow's side. Email survived because no one owned SMTP. The web survived because no one owned HTTP. PC software survived because the IBM PC bios was reverse- engineered into a thousand clones. Music survived the major labels because file-sharing made enforcement infeasible long enough for new business models to emerge. Every single one of these would be illegal under contemporary platform-friendly legal interpretations, and almost none of them would be attempted by a venture-backed startup in 2026.

Restoring adversarial interoperability is therefore not a fringe libertarian ask. It is the most actionable contemporary policy lever for resisting platform-stage enshittification — and it has serious advocates across the political spectrum, from the EFF to the Open Markets Institute to the EU competition authorities. The legal surface is moving. The technical infrastructure is being built. The cultural permission is the variable.

The Indic counter-frame — anti-rivalrous substrate

The deepest move past enshittification is structural — to build the platform on a substrate that the four-stage decay cannot uniformly traverse. The Moloch & Ṛta essay makes this argument at the level of substrate: rivalrous goods invite Moloch by construction; anti-rivalrous goods structurally do not. A platform built on a rivalrous substrate (proprietary ranking, closed user data, vendor-controlled APIs) gives enshittification the gradient it needs. A platform built on an anti-rivalrous substrate (open protocols, portable user identity, federated governance) starves the gradient.

The federated unicorn is the operational expression of this for the venture layer. Where the Valley built ₹10,000 Cr concentrated unicorns on rivalrous substrates and watched them enshittify on schedule, the federated unicorn distributes the same arithmetic across 10,000 federated proprietors — none of whom can unilaterally twiddle the substrate, none of whom can extract beyond their node's tolerance, all of whom carry obligation to the larger commons. The decay gradient does not have the same shape. The platform does not have a single CEO whose career incentives reward stage three. The capital structure does not demand it.

The Techno-Memetic Commons licence is the legal-engineering counterpart. The 45-day internal-use trigger turns silent enclosure into a license violation. The Stewardship Mark requires public accountability from commercial users. Together they make the enshittification playbook structurally hostile to building on TMC-licensed substrates, while permitting the legitimate commercial use that funds the commons.

Quick answers

Is enshittification just a synonym for "platform got worse"?

It is more specific. Enshittification names a particular four-stage curve with a particular structural mechanism (switching costs + twiddling + asymmetric power between connected groups). A platform that gets worse because of bad design or honest competitive pressure is not enshittifying — it is just losing. Enshittification is what happens when the platform is winning and that is exactly what makes it worse.

Can a platform avoid enshittification entirely?

Probably not entirely, but it can resist meaningfully. Open protocols, federated governance, member ownership, commons licensing, and patient capital each remove one of the accelerants. The realistic ambition is not platform purity; it is platform substrate that bends rather than snaps under the same gradient that would otherwise produce the four-stage curve.

Does this apply to AI platforms?

The dynamics are exactly the same, possibly faster. AI platforms attract users with subsidised access (current model), then start monetising the locked-in user base through API pricing and feature gating (now visible), then start extracting harder from API-dependent business customers (next phase), then either die or transition to a different business. The TMC licence is structurally designed for this substrate; the federated unicorn architecture is the capital response. See AI is the Audit.

Where should I read Doctorow?

Start with the original 2023 enshittification post and the follow-up Tiktok, Amazon, Facebook piece. Chokepoint Capitalism (with Rebecca Giblin) is the long-form theory. His daily newsletter Pluralistic is the live-fire commentary. The Internet Con is the policy book.