We should be able to ask AI for help without sharing everything with the company behind it.
We might be working on an unfinished research idea, reading a private document, or thinking through a decision we're not ready to discuss. An assistant could help. Having to send that material to someone else's servers gives us a reason to hesitate.
Our everyday computers could do more of this work themselves. An AI could help with our files and code while keeping them on the device. We'd still have access to more powerful models online, and we'd decide when to use them and what to send.
Useful enough for the work we actually do
For everyday work, an AI needs to get a few basic things right. A code change should pass its tests. An answer about a document should point us to the right passage. And it should all happen quickly enough to be worth using.
Smaller models and better software can help us get there. An update could let a computer handle a task it used to struggle with. We might get a useful upgrade without buying a new machine.
This is why efficiency matters. It can make useful AI available on computers more of us can afford.
How much extra would we pay?
Suppose we're already replacing a computer. We'd normally spend $1,800, or $2,000 for a version that also runs the AI we need. In that case, we're paying an extra $200 for AI. If we bought a separate $2,000 machine just for AI, the full price would be an extra expense.
We still need to count electricity, setup time, and any slowdown to our other work. Savings depend on what we stop paying for. Keeping the same online AI subscription means that part of the bill stays the same.
A $200 upgrade is easier to consider than another whole computer. That's one reason personal AI could reach us through the laptops and desktops we'd buy anyway.
NVIDIA, AMD, and Apple have different strengths
We won't all want the same thing from an AI computer. Each of these companies has a different way to make personal AI useful to us:
- NVIDIA: It builds both AI models and the hardware they run on, so it can design the two together and make better use of its chips. Its mature software tools also help it get new models running from the day they're released. For us, that can mean earlier access to new models and less work getting them to run well.
- AMD: Its path is to keep useful AI affordable. We can already see that in AMD-based PCs selling for around $2,000, within the budget many of us would set for a work computer. That gives personal AI a way into everyday machines at a price more people can afford.
- Apple: Its hardware keeps improving, including its chips and how quickly they can access memory. Those gains can help Macs handle more AI work directly. For those of us already buying a Mac, personal AI can become another thing the same computer does well.
The real test is whether we keep using these tools day after day. Speed matters, but so does having software that works without constant fixes. And adding AI to a laptop doesn't make its entire price new AI spending. We may have been buying a laptop anyway.
We should decide what our AI shares
An assistant should be able to start with the files on our computers. If it needs a more powerful model online, it should show us what it wants to send and ask first.
We should be able to take saved conversations and notes to another app, switch models, cut off access to a folder, or delete something we no longer want it to remember. Keeping things on our own devices helps only if the software actually protects our files and respects those choices.
As AI comes to know more about our work and our lives, we should have the final say over what it sees, what it remembers, and what it shares.