We seem prepared to believe almost anything about the future of AI except that ordinary people might have control over it.
Personal AI has introduced systems with unprecedented access to our lives. As we begin to rely on them to make decisions and take action on our behalf, who controls that intelligence will become incredibly important.
Now more than ever, we must build for a future owned by the individual.
Owned is a better product.
In the internet era, there were fundamental reasons to centralize.
Marketplace businesses like Amazon, Uber, or DoorDash only worked because supply and demand were aggregated in one place. Facebook or Instagram were useful because everyone you knew was already there. Products like Google were prohibitively large, petabyte-scale technical problems. In all of these cases, owning your own version would defeat the whole point.
Personal AI is different. Its value comes from how completely it becomes yours rather than how many other people use that same system.
The most useful personal agent will have complete context over your entire life. Everything from your email, health records, and finances to the conversations you have and the devices in your home. If you don’t fully trust the companies you’re giving this to, the bottleneck is no longer the model's intelligence but the access you're willing to give it.
Owning the machine your agent runs on removes that tradeoff. You can benefit from the value of your data without the worry of it sent to a third-party to be warehoused and analyzed.
Dedicated hardware also drastically increases how much work intelligence can do for you. With no provider charging for every additional token, we can run models that continually process new information as it comes in and take action proactively. If done right, AI will quickly become better at providing value based on its awareness of what’s happening in your life than you could ever be at prompting it.
Over time, your agent should grow to think differently, respond differently, and work differently from mine because our lives and what we want from them are different. This should be true at every level, from the context the agent reasons over to the harness, even down to the weights themselves.
Intelligence is power.
The question then becomes: where do the benefits of a system like this accrue?
The more capable and essential AI becomes to our lives, the more powerful whoever owns it becomes. If we surrender control of intelligence to a handful of labs and big tech companies, we give them unilateral authority to determine how intelligence should understand and act in the world.
Who can be trusted with this much?
A lab that claims it knows what is best for humanity does not entitle it to impose that judgment centrally. The last generation of tech companies made similar moralizing promises (“don’t be evil”) and folded when incentives became misaligned.
Ultimately, when an agent acts on my behalf, I should have the final say over how it behaves. That authority must remain mine even when the company that built it disagrees with me.
It is deeply dystopian to imagine a system constructed from the most sensitive information in my life being owned by a multi-trillion-dollar company and rented back to me by the token.
The more of ourselves we entrust to these systems, the less acceptable it becomes for ownership to sit anywhere else.
People deserve choice.
We have become dangerously disconnected from how people actually feel about the technology we build.
People protest Flock cameras, campaign to keep Palantir out of their public institutions, rail against labs training on their private data, yet the industry insists that “consumers don’t care about privacy”. We promise them abundance while building a future in which they own less and depend on us more. Why should we expect them to be excited about that?
We have spent years making surveillance convenient and ownership difficult. But we have barely begun to offer people another choice.
Well-intentioned alternatives like DuckDuckGo or self-hosting tried to push back and grant privacy, but meant accepting a significantly worse product or doing more of the work yourself. You cannot ask consumers to choose a principle over a better, easier, and often free experience.
That tradeoff is no longer true. We now have open models that can run on a high-end laptop and deliver Sonnet-5-caliber intelligence entirely locally. Within the next three to six months, we will have today’s Opus-level intelligence running privately on consumer hardware. For 90% of day-to-day tasks, local models are now good enough.
We are watching the capability of enormous, centrally hosted systems arrive on machines ordinary people can own. For the first time, choosing ownership does not mean choosing less.
Our Focus
The terms on which we bring personal AI into our lives are still being written. As these systems become more capable, more intimate, and harder to live without, those terms will become harder to change.
Independent systems like local AI have largely been built for hobbyists and tinkerers. Our job is to make that experience useful and simple enough for ownership to become the default.
I want to live in a world where the most powerful technology humanity has ever built belongs to the people whose lives it changes. We must be unapologetic about fighting to build it.
