Zuckerberg Promises an AI That Will Know Everything About You. Why Does Meta Want That?
Can a personal AI belong to the person using it, when the system is built and controlled by the corporation that owns Facebook, Instagram and WhatsApp?
The reader learns to read a technology company's manifesto as a business document: to separate a declared direction from a working system, to see whose interest each proposal serves, and to test a promise not against the author's intent but against what people already use.
What to watch for
Key takeaways
On 10 August 2026 Mark Zuckerberg published a position piece titled «The Future Belongs to Everyone». At the same time Meta released the open-weights model Muse Glimmer. Business Insider responded with a headline about Zuckerberg's assumption of AI-for-everyone running into a big problem.
Meta is distrusted for concrete reasons: the storage and study of data, fear of job loss, privacy. On top of that comes resistance to data-centre construction: power, water, noise, taxes, the load on local infrastructure. The one-billion-dollar fund is offered as an answer to exactly that.
Zuckerberg's thesis: everyone should have an agent that knows their goals, health, work, family and interests. But ChatGPT already knows a great deal about its user. The question is not whether such an agent will appear but what would have to happen for it to know everything — and who ends up owning that everything.
Nineteen years in business argue against the thesis that small companies will get stronger. Fifteen years ago opening a good coffee shop, winning at search optimisation or launching paid search was noticeably easier. Technology has not brought that ease back — it raised the bar for everyone at once.
Jensen Huang and Mira Murati have both spoken for openness. The open share will grow, for the simple reason of price. But a local model needs hardware of your own, and a genuinely good product will still cost money: free here usually means someone else is paying.
The agent should serve the person's goals rather than impose the company's values. But a person's goals are not set by them alone: the country they live in imposes limits they did not choose. Asking a system to serve a person's goals without naming who formulates them leaves the question open rather than answering it.
Meta claims the position of a company building personal AI for billions of ordinary people, unlike labs working for companies and governments. But Gemini has a billion app users, ChatGPT is around the same, and Anthropic runs large free initiatives. Meta's models live inside WhatsApp, Instagram and Facebook.
The list of what the agent should do ends with picking recipes, ordering groceries and smart glasses. It is August 2026: Codex and Claude Code build entire infrastructure solutions, systems are starting to do work that used to require a doctor and to save companies serious money. The promise describes a wave that has already passed.
A mode is promised in which even Meta will not be able to read a user's data. No architecture, no launch dates, no required devices have been published. This is a declared direction, not a working system — and the difference matters, because there is nothing to check.
The fund for people living near data centres is one billion dollars against infrastructure spending on the order of a hundred and fifty billion. At California's minimum wage of eighteen and a half dollars an hour, a hundred dollars of support spread over ten million people is five hours' work. At the scale of billions of people the sum stops meaning anything.
Hand over intermediate model versions before and during training, lend engineers to protect critical infrastructure, do not impose long reviews if they leave you behind China. The strongest models are not Meta's — and the proposal to share them is addressed first of all to someone else.
Zuckerberg proposes allowing models to train on other models' outputs. Shortly before, Anthropic filed a submission with the US Senate saying Chinese companies had trained on Claude and distilled it through masses of accounts. Elon Musk's own X flatly forbids scraping under penalty, and API access costs real money.
This is not a philosophical text, nor a billion-dollar act of charity. Meta owns Facebook, Instagram and WhatsApp — one of the largest audiences in the world. Building a personal assistant straight into those apps is full integration and serious corporate power, not AI that belongs to people.
What this episode is about
On 10 August 2026 Mark Zuckerberg published a roughly six-and-a-half-thousand-word position piece titled «The Future Belongs to Everyone». The same day Meta released a new open-weights model, Muse Glimmer. Business Insider headlined the piece this way: Zuckerberg's assumption about artificial intelligence for everyone runs into a big problem.
The episode goes through the manifesto point by point. First, the personal superintelligence: every person should have an agent that knows their goals, health, work, family and interests. The host notes that the thing described already exists — ChatGPT knows a great deal about its user — so the question is not whether such an agent will appear, but what would have to happen for it to know everything.
Second, small companies will get stronger. The objection comes from experience: nineteen years in business suggest the opposite. Small companies get not more opportunity but more competition. Fifteen years ago opening a good coffee shop, winning at search optimisation or launching paid search was noticeably easier than it is now — and technology has not brought that ease back.
Third, open weights. Both Jensen Huang and Mira Murati support the idea. The host does not dispute that the open share will grow — it is cheaper — but doubts that openness is what produces a genuinely strong system, and points out that a local model needs hardware of your own, while a genuinely good product will still cost money.
Fourth, the user's personal values: the agent should serve the person's goals rather than impose the company's. That raises the question of who defines a person's goals when they live in a country with its own restrictions — and, separately, how it squares with what the head of Meta has said in public.
Fifth, the distribution of power: several competing models are safer than one centralised superintelligence. Meta positions itself against labs that, in its telling, build AI for companies and governments. But Gemini has a billion users, ChatGPT is around the same, and Meta's models live inside WhatsApp, Instagram and Facebook.
A separate block covers what the agent is meant to do: analyse sleep, help with career and finances, pick recipes, order groceries, see the world through Meta's smart glasses. The host answers directly: that is a 2023 class of task. In August 2026 Codex and Claude Code build entire infrastructure solutions, and systems are starting to do work that used to require a doctor.
Then the private mode, in which even Meta supposedly will not be able to read a user's data: no architecture, no dates, no devices published — a declared direction, not a working system. A one-billion-dollar fund against infrastructure spending on the order of a hundred and fifty billion is seven tenths of one per cent; at the scale of billions of people the sum stops meaning anything.
The last block is the proposals to government: hand over intermediate model versions before and during training, lend engineers to protect critical infrastructure, do not impose long reviews if they leave American companies behind China, allow models to train on other models' outputs. The last one reads especially interestingly next to Anthropic's complaint to the US Senate about Chinese companies distilling Claude.
The conclusion is about interest. Meta's is concrete and commercial: to build a personal assistant into the apps billions of people already use.
The manifesto is neither charity nor philosophy. Meta owns one of the world's largest audiences, and a personal assistant built into its apps is not «AI that belongs to people» but corporate power carried into every user's kitchen. Meanwhile the promises describe yesterday's class of task: picking recipes and ordering groceries — in August 2026, when Codex and Claude Code build whole infrastructures.
Episode transcript
The episode is in Russian; below is an English reading guide to the transcript (the full EN transcript is a machine translation). Voice matching applied to 66 segments: 66 identified, 0 mixed, 0 marked with ✓, and 0 unresolved.
Loading…