One of my favorite pieces of software is Apple photo search. If you've got an iPhone, try it: great searches to try are "animal selfie", "bird", "ice cream", or "cake".

What's particularly amazing about these searches is that the machine learning is performed on-device. In fact, Apple provides developer tools for on-device machine learning in any app across its platforms. There's no cloud processing and the privacy issues related to that. The power to identify which photos are a selfie with your cat lies in the palm of your hand.

Not everyone can afford a top-end iPhone, but these represent the leading edge of the technology; in the near future, every phone will be able to perform rapid machine learning tasks.

Another thing my phone does is connect to 5G networks. 5G has a theoretical maximum bandwidth speed of 10gbps, which is faster than the kind of home cable internet you might get from a company like Comcast. In practice, the networks don't quite work that way, but we can expect them to improve over time. 5G networks will allow us to have incredibly fast internet virtually anywhere.

Again: not every phone supports 5G. But every phone will. (And, inevitably, 6G is around the corner.)

Finally, my phone has roughly the same amount of storage as my computer, and every bit as fast. Not everyone has 256GB of storage on their phone - but, once again, everyone will.

On the internet, we mostly deal with clients and servers. The services we use are powered by data centers so vast that they sometimes have their own power stations. Technology startup founders have to consider the cost of virtualized infrastructure as a key part of their plans: how many servers will they need, what kinds of databases, and so on.

Meanwhile, the client side is fairly thin. We provide small web interfaces and APIs that connect from our server infrastructure to our devices, as if our devices are weak and not to be trusted.

The result is a privacy nightmare: all our data is stored in the same few places, and we usually just have to trust that nobody will peek. (It's fair to assume that somebody is peeking.) It also represents a single point of failure: if just one Amazon datacenter in Virginia encounters a problem, it can seem like half the internet has gone down. Finally, the capabilities of a service are limited by the throughput of low-powered virtualized servers.

But the world has changed. We're addicted to these tiny devices that happen to have huge amounts of storage, sophisticated processors, and incredibly fast, always-on connectivity. I think it's only a matter of time before someone - potentially Apple, potentially someone exponentially smaller than Apple - uses this to create an entirely new kind of peer to peer application infrastructure.

If I'm in the next room to you and I send you a Facebook message, the data finds its way to Facebook's datacenter and back to you. It's an incredibly wasteful process. What if the message just went straight to you over peer to peer wifi (or whatever connection method was most convenient)? And what if there was a developer kit that made it easy for any engineer to really easily build an application over this opportunistic infrastructure without worrying about the details?

Lately I've been obsessed with this idea. The capabilities of our technology have radically changed, but our business models and architectural paradigms haven't caught up. There's an exciting opportunity here - not just to be disruptive, but to create a more private, more immediate, and more dynamically functional internet.