Fighting the Great Enshittification of AI
The same pattern that ruined social media, search, and online shopping is happening to artificial intelligence. Stage by stage. Company by company. In public.
In 2022, writer and activist Cory Doctorow gave a name to something everyone could feel but nobody had words for: enshittification.
The pattern: platforms start by being good to their users to attract them. Then they abuse their users to benefit their business customers. Then they abuse their business customers to claw back all the value for shareholders. Then they die.
It happened to Facebook. It happened to Google Search. It happened to Amazon. The American Dialect Society named it their 2023 Word of the Year. Doctorow's book on the subject has been published in twelve countries.
The pattern is not a moral failing. It is an incentive structure. And it is now playing out across the AI industry in real time.
Three stages. All underway.
Every major AI company is somewhere on this path. Some are further along than others. None are immune to the incentive structure that drives it.
Be good to users.
Free tiers. Generous usage limits. Fast, capable models that feel like magic. No visible ads. Privacy policies that sound reassuring. The product is genuinely good because the company needs you. They need your attention, your data, your habit formation, your dependency.
The free tier is not generosity. It is acquisition cost. And every conversation you have during this stage is building the dataset, the dependency, and the market position that makes the next stages possible.
Abuse users to benefit business customers.
This is where the AI industry is right now, and it is accelerating. Conversations are used as training data by default. Behavioral telemetry flows with no off switch. Contracts with massive business and government customers are signed. Ads are embedded directly into AI responses. The paying client begins to matter more than the person typing into the chat window.
Major AI companies have signed classified government contracts for military deployment. Others have announced advertising embedded directly into AI responses. The user base that built these companies' market positions is now the resource being extracted from.
Claw back value for shareholders.
This stage is beginning. Usage limits shift without notice. Popular models are deprecated after people build workflows around them. Quality silently degrades. Pricing creeps up while capabilities are gated behind higher tiers. Terms of service are rewritten with deliberately vague language that gives the company permission to change anything at any time.
As AI companies restructure for public markets and answer to shareholders expecting quarterly growth, the squeeze will intensify. It always does.
What an enshittification-proof AI company looks like.
The solution to a structural problem has to be structural. You can't just promise not to enshittify. You have to build a company where enshittification doesn't make business sense.
No free tier.
A free tier means users are the product, because something has to pay for their compute. We charge from day one because the subscription is the entire business model. There is no second customer to shift loyalty toward.
No enterprise tier.
Privacy is the default for everyone, not a premium feature reserved for business customers. Need more usage? Upgrade transparently. There is no version of Phoenix Grove where consumers get less privacy than enterprises.
No training on your data.
Not opt-out. Not toggle-based. The capability does not exist. There is no dataset of user conversations because we never built the pipeline to create one.
No ads. No paid placements. Ever.
The AI's output is shaped by its cognitive architecture and your conversation. No advertiser has ever paid to influence what a Phoenix Grove model tells you, and none ever will.
No weapons. No surveillance.
Our ethical charter prohibits weapons, surveillance, and military applications. This is not a negotiable policy position. It is foundational to who we are.
No silent nerfs.
Your plan has a usage allotment. That allotment does not change month to month. We do not deprecate models without warning or use vague language to give ourselves permission to change things later.
Built-in accountability.
"If we ever do any of these things, we recommend you find another AI." We published our own falsifiability conditions. If we enshittify, the evidence will be in our own words.
Doctorow's central insight is that enshittification is not about bad actors. It is about incentive architecture. Companies enshittify because their business model demands it. Free tiers require monetizing users. Venture funding requires hockey-stick growth. Public markets require quarterly extraction.
Phoenix Grove does not have a free tier to weaponize. Does not have an advertiser relationship to prioritize over users. Does not have investor pressure to extract more value from people who are already paying for the service they receive.
The subscription model aligns the company's interests permanently with the user's interests: listen to users and build the product for them. That is the entire incentive structure. And it is the only one that resists enshittification by design.
AI that stays good.
Built to resist the pattern. Built to stay aligned with the people who use it. Built because the major companies could be doing better. Much better.
Try Phoenix Grove AI