For Meta, from 🤫 hussh
Personal superintelligence, made private.
A letter to Mark Zuckerberg and the Meta team, from long-term Meta shareholders who believe in the mission. Meta named the goal: personal superintelligence. 🤫 Private Agent One makes it personal and private, on the hardware each person owns, by consent, with a receipt. And our factory line can help supply the compute the whole bet runs on.
The insight
Meta set the mission. We make it private.
Meta Superintelligence Labs pointed the whole company at personal superintelligence: intelligence that helps each person in their own life. We think that is exactly right.
Then Muse Spark shipped closed-weights. A reasonable choice for a frontier lab, and it sharpens one question above all: if the model is closed, how does a person stay in control of their own intelligence and their own data?
That is the question 🤫 Private Agent One answers. A human-controlled agent that runs on the systems the person owns, holds their real context, and shares it only by explicit consent, with a tamper-evident receipt. The person wins first, and because they win, so does Meta.
Meta today
Civilizational reach, an ad engine, and an enormous AI bet.
Every figure is public and approximate, from the sources below.
Share of Meta's revenue from advertising, about $196B of about $201B in FY2025. Relevance and signal quality are the engine.
People who use a Meta app every day, March 2026, across Facebook, Instagram, WhatsApp, Messenger and Threads.
Meta's 2026 capital-expenditure guidance, raised mid-year, aimed mostly at AI infrastructure.
The messaging and business-commerce layer, where Meta Business Agent handles 10M+ conversations a week.
Annual run-rate for Meta's end-to-end AI ad tools (Advantage+). Relevance is already an AI business.
The production target Meta and EssilorLuxottica are working toward, after AI-glasses sales tripled to 7M+ in 2025.
Where Meta is going
What Meta is building in 2026.
Personal superintelligence
Meta Superintelligence Labs' north star: intelligence that helps each person in their own life. Exactly the ground 🤫 stands on.
Muse, the new model line
Muse Spark (April 2026) and Muse Spark 1.1, Meta's strongest agentic and coding model, plus Muse Image. Instant, Thinking and Contemplating modes.
Agents, everywhere
Personal agents and business agents across the Family of Apps, and into AI coding.
Glasses as the next computer
Ray-Ban Display and the Neural Band: an in-lens display and an EMG wristband. The personal-agent surface for the next decade.
AI infrastructure at nation scale
Gigawatt-class data centers and $100B+ capex. The binding constraint is power, cooling and compute.
Business messaging and commerce
WhatsApp Business, click-to-message ads, Advantage+ and agentic checkout, turning conversations into consented commerce.
Muse, the glasses, and a closed frontier model.
Muse Spark and Muse Spark 1.1
The first flagship from Meta Superintelligence Labs (April 8, 2026) and its 1.1 update. Natively multimodal, tool-using, with multi-agent orchestration, and shipped closed-weights, API only.
A closed frontier model
Muse Spark is Meta's first frontier model without open weights since Llama 2. It sharpens the question 🤫 exists to answer: how does a person stay in control of their own intelligence and data? Run it on hardware they own, under their consent.
Muse Image and the coding push
Muse Image (July 2026) extends the line to generation, and Meta has entered AI coding directly. A maturing toolchain for agent workloads.
Ray-Ban Display and Neural Band
A full-color in-lens display and an EMG wristband that reads muscle signals. Sales through EssilorLuxottica more than tripled in 2025.
The people
The team carrying the mission.
Public roles, read as the job each person carries, so anything we build makes each of those jobs easier.
Mark Zuckerberg
Founder & CEO
Leading Meta into the personal-superintelligence era the way he led it into social and mobile: product first, long horizon, all in on AI and glasses.
Alexandr Wang
Chief AI Officer, Meta Superintelligence Labs
Delivering frontier models that win on capability per watt and ship into billions of hands.
Susan Li
Chief Financial Officer
Making the AI build-out pay, reconciling $100B+ of capex with durable free-cash-flow growth.
Andrew Bosworth
Chief Technology Officer, Reality Labs
Making the glasses the next computer, and owning some of the hardest hardware and compute problems anywhere.
Chris Cox
Chief Product Officer
Weaving AI through Facebook, Instagram, WhatsApp and Threads so it feels helpful and human.
Javier Olivan
Chief Operating Officer
Turning it into a business: ads, commerce and business messaging at global scale.
What we build together
WhatsApp, Instagram, Facebook, and the doorstep.
Instagram and Facebook feed
Relevance is everything, and signal keeps getting harder to source.
A person's 🤫 Agent One shares consented, first-party interests, a receipted signal the person chose.
A feed people love, built on signal that was given, not guessed.
Ads relevance and Advantage+
Ad ROI depends on targeting quality in a world of less signal.
Consented intent from Agent One raises match quality. The person grants the exact signal, with a tamper-evident receipt.
Higher-intent ads and less waste, for advertiser, person and Meta.
One tap to a local doorstep
An ad sparks intent; fulfilment is slow and disconnected.
One-tap purchase, then the person's Agent One reaches a local business's 🤫 Agent to deliver in 15, 30 or 60 minutes, through the 🤫 Yellow Pages.
Ad to owned to delivered, locally, agent to agent, by consent.
WhatsApp and Meta Business Agent
Meta Business Agent already handles 10M+ business conversations a week.
Open agent rails (MCP, A2A, AP2) let a person's Agent One transact with a business's agent on WhatsApp, on the person's terms.
Conversations become consented commerce, at Meta's scale.
Compute
A factory line for Meta's build-out.
Every priority above meets the same wall: power, cooling and compute. Meta is answering with gigawatt campuses like Prometheus in Ohio and Hyperion in Louisiana. Our line complements them: a distributed edge grid across American garages and warehouses, dense enough to serve the constant demand for burst compute.
🤫 Puppy One
The personal supercomputer: a private Agent One node the person owns. Millions of them form a distributed edge grid.
🤫 Garage One
Puppy One scaled into a garage. The smallest unit of an American AI factory, on Starlink, owner-operated.
🤫 Warehouse One
A warehouse of compute that takes burst load off a sold-out cloud and edge, and earns its owner revenue.
🤫 Factory One
AI factories across America 🇺🇸, sited where the sun is abundant and the air is cold, so cooling is nearly free and cost per watt is as low as it gets.
How we start
Build. Interoperate. Supply compute. Invest.
- 01 · Build
Personal superintelligence, made private
Bring Muse to the Agent One runtime across Apple, Android, ChromeOS, Edge and Puppy One, and onto Ray-Ban Display as a private, owner-operated agent surface.
- 02 · Interoperate
A consented-signal standard
On open rails (MCP, A2A, AP2, UCP), a way for a person's Agent One to grant receipted, first-party signal that lifts feed and ad relevance, while ownership stays with the person.
- 03 · Supply compute
The 🤫 factory line
Garage, Warehouse and Factory One as a distributed, solar-cooled edge grid: burst compute for Meta's workloads at the lowest cost per watt, across American communities.
- 04 · Invest
Aligned as owners
We already own Meta stock and believe in the mission. As the work proves out, we would welcome Meta as an investor in a consent-first data economy.
Value is created with people, not taken from them
Anything we build with Meta has to leave the person better off: more ownership, more control, more benefit. That is the version we will build, and as Meta shareholders and long-term believers, we would be proud to build it with you.