Bond
Apple
Download
←News
AnnouncementAug 18, 2026

Superpowers, Not Superintelligence

Mark Zuckerberg is right about the important parts in “The Future is for Everyone”. Right that the defining question is who gets access to superintelligence and what we point it at. Right that concentrating it in a few institutions would be a catastrophe. Right that there’s no objective answer to what a good life looks like, and people should decide that for themselves.

He’s also right about the mechanism. Power checks power. The safest arrangement is one where nobody accumulates enough of it to stop being checked. He applies the balance-of-power test to companies, to countries, to labs, to AI. He applies it to model access and to governance.

He never once applies it to data acquisition.

The incentives we set now are the ones we live with for the next 20 years. We wrote a contract like this before for the internet, pioneered by companies like Meta. Free products in exchange for data collection, surveillance, and ads. It looked like a good deal.

In the AI era, the power sits in the input layer. The input has to come from you and be owned by you.

I should declare my own interest. I’m building in this new direction. This essay is about a new path forward. Not why you shouldn’t trust anyone. What becomes possible when we put humans first ahead of machines.

Look at Meta’s Actions

Zuckerberg promises an agent that understands you, your goals, everything you care about. Look at how it learns. It monitors his sleep. It watches him train. It flags what it thinks he’ll find interesting. It plans his daughter’s recipes and orders the ingredients.

The agent acts in every sentence. He is the object of all of them. Now notice who’s missing. You.

This isn’t an accident of the examples. A year ago, in July 2025, he wrote what the input would be. Glasses that “see what we see, hear what we hear,” worn all day, becoming our primary computing device. A camera and a microphone on your face, understanding your context ambiently because it never stops recording it.

That is a design decision, made early and stated plainly. The intelligence has to watch you, because nobody ever tells it anything. It needs nothing but your presence. That is the asymmetry his own essay warns about, moved one level down. If concentrated power is the danger, then an agent who collects data silently and whose inputs you don’t choose concentrates power too. Not in a government. Not in a lab. Whoever owns the input layer.

Glasses that watch you all day feed the same profile to the same company. Distributing an open weights model while centralizing the capture isn’t a balance of power. It’s the same arrangement with better optics.

Zuckerberg promises a fully private mode where even Meta can’t see your information. His actions highlight this is likely just a mode, not the default. On December 16, 2025, Meta’s updated privacy policy took effect that signaled the opposite. Your private conversations with Meta AI now personalize the content you see and the ads you’re shown. ~$200 billion of advertising revenue is a powerful incentive after all. His actions speak louder than his words.

A Meta notice titled “What you should know” explaining that on December 16, 2025, interactions with Meta AIs will start being used to personalize experiences and ads.
Meta’s private policy updates in December 2025

Tools, Not Machines

Personal superintelligence. Read the noun. Superintelligence puts the intelligence in the machine. The machine is the focus and the one that’s smart. Not you. People want superpowers, not superintelligence. A superpower puts the power in the person. You become the one who can do something you couldn’t before.

That sounds like semantics. It isn’t. It decides where the information comes from and who owns it.

If the machine is the focus, its goal is to learn about everyone and act on its own. And the only way to learn about you without your participation is to observe you. Follow you around. Make inferences. Assemble a profile from the residue of a life. Then use that data to make itself better.

If you’re the focus, the machine has to be told. It asks. You answer. You’re the author and the owner of your own data. It’s the instrument. You get better.

Two architectures. One requires your participation to function and serves you. The other requires only your presence and serves itself.

Every significant tool we’ve ever called a superpower is the first kind.

  • With a phone, we can communicate across the world in seconds.
  • With cameras, we can preserve our experiences and knowledge.
  • With search engines, we can reach information from anywhere on earth.
  • With MRIs and microscopes, we can see inside our own bodies.
  • With telescopes, we can see past the limits of our eyes.
  • With GPS, we can extend our sense of where we are.

Every one of them allows you to control the input and own the output. An agent that captures data ambiently breaks that. It becomes a machine that you serve. Not a tool that serves you.

Facebook serves as a cautionary tale of this dichotomy. It was originally built to be a tool to connect people. It evolved into a machine that creates addiction, monopolizes time, and maximizes ad revenue. On August 7, 2026, a judge had to order Meta to stop pushing notifications at children at midnight. The tool became a machine that now controls you.

Ambient data collected from surveillance is not the path forward. And it won’t unlock AI’s full potential.

The original Thefacebook homepage from 2004, welcoming Harvard students to an online directory that connects people through social networks at colleges.
Facebook was originally built to be a tool

AI is Limited by Input

AI is humanity’s next breakthrough technology. Its superpower is still being defined. For consumers today, it primarily serves as a better search engine, writing assistant, and learning tool.

AI is now limited by input, not capability. Frontier models can pass the bar exam and can’t tell you whether you should take the job. They know almost nothing about you. What input will unlock AI’s full potential?

We believe it will be stated inputs. Things in your mind you haven’t shared before. Not ambient data from your actions. The reason is simple. Behavior records what you did. It doesn’t tell you why.

I love the movie Moneyball. I’ve watched it five times. Billy Beane takes over a baseball team and uses data science to turn them into winners. He was contrarian and right, and went on to shape the evolution of the entire sport. His approach and impact appeal to my ambitious nature. The acting doesn’t hurt either. Brad Pitt and Jonah Hill are great. But I’m not a baseball fan. I wouldn’t watch any other baseball movie.

Now compare this to ambient data. Clicks, location, watch time, purchase history. Those already exist. Several companies own copies. None of it has made a single system understand a single person. “Personalized” recommendations come back in the form of long lists or endless feeds instead of single choices. I keep seeing baseball movies.

In the AI era, gathering stated input becomes simpler and addresses the key shortcomings of behavioral data.

Behavior is a revealed preference. The standard objection is that revealed preferences beat stated ones. Revealed preferences tell you what someone chose. Acquisition of this data is simple and the record is precise. However, understanding why something was done is underdetermined. Behavior cannot tell you why. It’s equifinal — one outcome with many possible paths. So the next recommendation becomes an option based on what others with similar paths made, rather than a single choice for you with a clear reason. The end result? Recommendations and ads with 1% click-through rates are viewed as a success and technology wastes your time.

Stated preferences address this error. They’re direct about the reasoning but unreliable in their acquisition. They come with friction, biases, and performative behavior. AI partially addresses these issues with testimony. It makes it easier to create, addresses conflicts in information, prompts questions to resolve ambiguities and biases, and analyzes the information on a deeper semantic level.

A table comparing behavioral and stated information: behavioral data is reliable to acquire (logs, transactions) but underdetermined — impossible to understand the “why” — while stated data is unreliable to acquire (free-form, biased, incomplete) but states the reasoning directly, subject to fabrication/exaggeration.

Stated input will turn AI into a superpower. Fortunately, there’s one body of material that already has it. Your memory.

Memory is the Input

Your memories are already powerful. Your first kiss. Your first day of college. The song that defined your childhood. I bet you can see or hear them right now. You might even remember the smell, the taste in your mouth, the feeling in your stomach. Memories transcend time and space. They let you go back, move somewhere else, sit with people who are gone.

More importantly, the memories you have are capital. The sociologist James Côté named this identity capital. Our accumulated resources and experiences that function as our currency in social and economic life. These memories are what you spend to get hired, to build trust, to convince someone to love you. Everyone holds this asset. Nobody has ever successfully built the tools to collect them.

Previously, it’s been too much work. The graveyard is full of evidence. Journaling apps. Life-logging cameras. A decade of products that asked people to sit down and record their lives. They all lost to systems that asked for nothing but consumption. Convenience has beaten intention every time it’s been tried.

With AI, the equation is finally changing. The friction of capturing the information from your mind into a digital interface has dramatically reduced. AI can now gather context from voice, from multimedia, from text. You create. It asks.

Capturing memories requires proactive creation. This is something only social networks have accomplished. This is now at risk as these networks optimize for consumption (e.g. short-form videos) in order to maximize ad revenue. This also rewards performative behavior for those that do create. Chatbots like ChatGPT or Claude fare no better. They’re single-player tools built for general knowledge retrieval and work. Without an external audience, there isn’t enough motivation to proactively create.

This is what we’re solving at Bond. Audience is what makes people create, and audience is what makes them perform. However, performed memories are worthless. They reintroduce exactly the unreliability that made stated preferences suspect in the first place. If we solve motivation with a feed, we lose the thing we came for.

So we keep the audience and remove the ranking. Nothing is amplified or scored against anyone else’s. Performance has no payoff. Being read by three people who asked is a different incentive from being ranked against everyone who didn’t. More importantly, the record cross-examines itself. Contradict something you said eleven months ago and you get asked about it. This is a cost a diary never imposes and behavioral data can’t impose at all. You’re also the primary consumer of your own record. A dishonest entry degrades your own recommendations and experience first. On a social network, lying is rewarded. Here it costs you.

The Moneyball paragraph you read above is part of my Bond memory. Here is what Bond did with it.

Three iPhone screens showing Bond flagging a memory conflict: a backstory about watching Moneyball, a card asking “You watched Moneyball but previously said you don’t like baseball. Why?”, and the resolved memory “Enjoys movies with contrarian plots and characters with ambition”.
Bond flagging a conflict with my Moneyball memory (real account)

Most importantly, no one captures memories because they haven’t been valuable. I have 40,000 photos on my iPhone but I get no value from them. It’s just a filing cabinet with better lighting.

That changes now. With AI, your memory becomes a superpower.

Memory is Your Next Superpower

With AI, you will have superhuman memory. This is the future we’re building at Bond. Here’s what your memories will allow you to do:

See patterns and insights. You just watched one conflict get resolved. The value isn’t in one. It’s in connecting 6,020 memories, watching the patterns you can’t see from inside your own life become legible. Seeing the bigger picture previously required work. Now it’s available at your fingertips.

Two iPhone screens showing a Bond Memory Recap: a monthly summary with a photo, topic tags like Bosnia, World Cup, and Toronto, and prose sections describing the month’s patterns and what they say about the user.
Preview of memory recaps in Bond (real account)

Make discoveries. Ask any agent on earth “what movie should I watch next” and you get a list of options based on a category of person. Ask a system holding 6,020 of your memories and it can apply your preferences semantically to find a singular choice. One you haven’t previously considered with a clear explanation as to why. You can’t do that from inside your own life. Nobody can. It requires a record you were honest in and enough distance to read and apply it.

An iPhone screen showing Bond recommending the movie War Dogs as a single personalized pick, with an explanation grounded in the user’s memories.
Recommendations in Bond (real account)

Visualize your mind. Humans connect and recall memories through associations. People, places, things, and experiences organized by their various relationships. Up until now, doing so in a digital interface required significant work through writing, tagging, linking, or other similar actions. They’re also one-dimensional. Now, with AI, associations can happen on their own and across numerous dimensions. Many successful products you use have built a one-dimensional graph. Your friend graph, your interest graph, your skills graph. It’s now possible to build a multi-dimensional graph. The most valuable graph. Your memory graph.

The Bond memory graph: thousands of memory dots clustered into Social, Places, Experience, Interests, and Things around a user’s profile, with photo thumbnails surfacing individual memories.
Preview of Bond memory graph (real account)

Stop starting over. Every new relationship or interaction begins from zero. You try to compress a decade into forty minutes and hope you picked the right forty. You compress at the expense of depth and differentiation. Now, you have an authored record that travels. You can make your life accessible to others with total control over who sees what. Your memories become portable.

An iPhone screen showing a full memory summary in Bond with an export menu offering Gemini, ChatGPT, and Claude.
Memory exports in Bond allowing you to carry memories into ChatGPT, Gemini, and Claude (real account)

Delegate for real. An agent acting on your behalf is only as good as its model of your standards. Behavioral data gives it your revealed preferences, which are mostly your worst impulses at your least deliberate moments. An authored record is the kernel of how you make decisions, what you actually decided, and why, at your best moments. That’s the difference between an agent that books what you’d click and one that books what you’d choose.

Judge AI’s work. This is the part nobody talks about. When a model tells you something about yourself and it’s wrong, an inferred system gives you nowhere to go. You can’t see the input, you can’t correct it, you can’t tell whether it’s a bad inference or a bad model. An authored record has a source. You can read it. You can fix it. You can watch the output change. Auditability and observability are necessary to make the output trustworthy enough to act on.

An iPhone screen showing the “Memories used” sheet in Bond, listing the specific memories a movie recommendation drew on.
Clear explainability for your recommendations in Bond (real account)

Share knowledge without work. Accessing verified human experiences today is expensive. There has to be a need. Then you have to call a friend, meet with them, or hope you find the right content online. There is also a stigma around certain topics. Money, health, failure. Now you have a new way to share knowledge and foster deeper bonds. Send your profile, connect with a friend, and see what they ask. Let them access your experiences and knowledge in seconds to kickstart a deeper conversation in real life.

An iPhone screen showing a Bond external profile answering a friend’s question about San Francisco recommendations, drawing on 54 of the profile owner’s memories.
External profiles on Bond (real account)

Preserve memories forever. Every generation before this one left behind objects and secondhand accounts. Photographs are evidence that something happened. They aren’t the account of it. Your grandkids could have the thing itself. Your words, your voice, answering questions you never anticipated. Culture can be preserved. Progress can be accelerated. Your memories can live forever.

That’s the superpower. Not a better search box. Memory that stops decaying, becomes legible, and starts working for you.

Memory is Currency

Your memories will also become a source of financial capital, not just identity capital. The contract today is a user gets access to a free product and in return gives up their data and sees ads. We propose a new monetization model that gives the user the power.

Your memories will be yours. They’ll be encrypted. They’ll serve you, not advertisers. To the extent you want to share them with friends, you can. To the extent you want to monetize them, you can do so yourself and elect to answer questions from external parties. Both are clear value exchanges.

To monetize is not to hand over a file. A buyer asks questions and receives answers, the same way a friend does on your profile. The source stays encrypted and stays yours. What you sell is a response, an expression of your memories that is revocable and priced per interaction. That distinction is the whole thing. Sell the corpus once and you have sold it forever.

We know convenience beats intention. This applies to Bond as well. This is why the default has to be clearly established. Nothing is licensable, everything is opt-in only, and the price is visible at the moment of the decision. If any of those three are missing, the contract of the AI era will resemble the internet era.

Memories are already becoming valuable. Reddit licenses user-generated data to the AI labs for model training, generating over $100 million in revenue per year. The users who shared these experiences and memories see none of this value. This is not right. As the data licensing and data labeling industries grow, we believe users should have control over what gets shared and have the majority of the value go to them, not the intermediaries.

A Search Engine Journal article headlined “Reddit CEO Says LLMs ‘Would Not Exist’ Without Reddit Data”, with highlights on Reddit content as “modern oil”, new data licensing deals, and the community rejecting AI-written content.
Search Engine Journal article from May 2026

Beyond data licensing, there are new industries being born around expressing your memories.

The research already points this way. Park et al. (Stanford, 2024) showed that agents grounded in two-hour interviews replicate real people’s attitudes far more accurately than agents built from demographics alone. Simile has commercialized exactly that. Train on real qualitative interviews, produce synthetic respondents that answer like the originals. The interview is the asset. The person who sat for it owns none of what it becomes.

A news article headlined “Simile raises $200M Series B at $2B valuation to survey AI twins of consumers”, with an illustration of isometric miniature cities.
Dealroom article from July 2026

We envision an alternate approach. Where users create their own profiles, curate what matters to them in their daily lives, and then monetize their own data as they see fit. The buyers already exist. Research agencies and enterprises pay up to five figures per study to recruit hard-to-reach respondents, and the respondent sees a sliver of it. Nobody has built the rail that lets the respondent hold the asset and price the access. Importantly, the payment attaches to a record you already made, never to the act of making one. Pay per memory and you get manufactured memories. A ~$50 billion research industry in the US alone, waiting to be transformed by users who capture memories, not just behavior.

Ads represent the default monetization model of the internet era. They put the power with intermediaries first, then advertisers, then people last. Our proposal flips this for the AI era. It ensures that the power and economic value will be held by the people first.

A New Contract

We must learn from our past mistakes to ensure a more aligned future. A new contract is required to ensure humanity benefits from AI and unlocks its full potential.

Give everyone control over their input. Don’t surveil them. Let AI be a tool.

Provide a safe medium to allow everyone to capture their memories. First for themselves, then for others. Use AI to bring these memories to life.

Give everyone full ownership and control over their data. Turn memories into everyone’s most valuable asset.

If we do this, then memory will become everyone’s superpower. This is what people want after all. Superpowers, not superintelligence.

- Dino August 18, 2026

AboutNewsTermsPrivacy

Dots Connected Inc.