Personal AI and the Third Communication Age

In this blog post, I want to make some big claims. I don't think they're wrong, but they could be. As with any big claim, I reserve the right to throw my hands up, change my mind, and completely reverse course the second the data shifts. Like Keynes reportedly said, when the facts change, I change my mind. What do you do?

When I was an undergraduate student at Georgetown in the mid-1990s, we had to write a senior thesis. As I remember it (dimly, I admit), it had to be at least 30 pages, a daunting affair that gated our graduation and required enormous research, many meetings with our senior thesis advisor, and multiple drafts that always seemed to fall flat for one reason or another. In spite of all of this, mine somehow ended up being more than 100 pages in length, partly because I found it challenging even then to write with brevity, and partly because I was so fascinated with the ideas I was struggling to understand that my exploration of them went far beyond the thesis assignment. My thesis advisor was dumbfounded (and likely didn't read past page 30). But the paper I ended up writing, "The Third Information Age," led me on a journey that has now consumed nearly 30 years of my life.

In that paper, I argued that humanity had experienced three great information ages, defined for the most part by the methods used in each age for information storage: (1) the memory age (information stored in memory alone), (2) the physical age (information stored in written form), and (3) the digital age (information stored as 0s and 1s). My argument wasn't that our means of information storage had improved (I actually believe that remains a subjective and questionable assessment), but that the method of storage was integral to the way society has evolved. In other words, the method of information storage influenced not only what we store, but more importantly, how information itself is viewed, used, and communicated between us.

Since then, I've come to believe that if we are to understand our present and predict our future, it's necessary to complement the way we store information with the way we communicate that information. Helpfully, like information storage, the long arc of information communication also breaks down into a tidy triptych: (1) the pre-language age (communication by mimesis), (2) the language age (communication by word), and (3) the post-language age (communication by 0s and 1s). There is much to criticize in this simplification, but I think it's a useful framework for analyzing the disruption we currently face as we move from person-to-person communication primarily based in language (letters, books, magazines, emails, texts, workplace messaging apps, social media) to AI-to-AI communication primarily based in the digital realm.

Big Claim Number 1: A post-language world

We are moving to a post-language world where most of our communication infrastructure will not involve language, or even the translation of 0s and 1s into language. Instead, navigating the world will involve personal AIs that deeply know us and serve as our primary interface with the world, mostly by interacting with other AIs neurosymbolically through transport layers that bypass encoding and decoding language and instead navigate overlapping high-dimensional spaces to communicate digitally stored information while retaining a semantic surface for their owners.

Big Claim Number 2: More personal communication

Email, text, workplace messaging, web pages, social media, and similar language-based communication methods will be transformed from their current state (high friction, low signal to noise, briny mixtures of low value / high value information necessary to navigate modern complexity) to their future state (low friction, high signal to noise, pure H2O information channels reserved for our most meaningful relationships and activities). Personal AIs will shoulder much of the burden of interfacing with the world to accomplish life's most frustrating communication tasks (calling the bank, paying taxes, scheduling appointments, planning trips), rendering low-value personal communication largely unnecessary (or at least optional).

Big Claim Number 3: Agent-to-Agent platforms

Personal AIs will NOT use language-based communication platforms to interact with the world. Compared to various machine alternatives, human language is incredibly inefficient (size and compute), error-prone (semantic misunderstanding), and disorganized (structureless). Trying to stuff AI communication into any app built for person-to-person communication (email, text, workplace messaging, social media, etc.) makes little sense to me.¹ It does, however, follow a familiar pattern of technology adoption: people adopt a particular way of using one technology that works, invent a new technology, and then try to map their new technology onto the way they used the old technology.² My claim here is that we are moving from a technology world dominated by person-to-person platforms to a world dominated by AI-to-AI platforms which haven't been built yet.

The prediction

All of this brings me to the X post that I wrote earlier today which argued the same point, although in a much briefer (and possibly clearer) way:

Contrarian prediction: In 3-5 years, our personal AIs will replace email, text, calls, and workplace message apps as our primary interface with the world.

Each of us will have a personal AI with a durable identity. Our AIs will talk directly to each other to plan trips, arrange dinners, schedule doctor visits, coordinate work, and navigate life's endless administrivia.

Personal communication will be reserved for relationships and activities that matter most. The next great communication platform won't connect people to people. It will connect their AIs.

We're building it now.

At Anywhere Intelligence, we take a contrarian view of, well, pretty much everything happening in the AI space right now. First, it's unfortunate that the term AI even exists. Synthetic intelligence would have been such a better term if only John McCarthy had thought of it in 1955. But I digress. Second, and more important, there's no reason AI needs to be viewed as a polarizing technology, with some treating it as a god (it's not) and others as a devil (it's not). Like all technologies, our view is that the group of technologies we call AI has the capacity to both help and hurt, to create abundance or scarcity, to increase equality or inequality. Will it be disruptive? Of course. And managing that pain and dislocation is the responsibility of both private and public enterprises. But can it be beneficial? We believe it can, if we allow ourselves to be guided by learnings from prior ages that similarly struggled with adopting transformative technologies. The key is managing—not ignoring—the change.

For us, that means building principle-based, Third Communication Age products that both anticipate and enable the moment when agent-to-agent communication becomes commonplace.

Pretending it won't happen, or that we can prevent it from happening, seems to us naive. Over the next few years, agents will stop being stovepiped in separate competing AI apps (Claude, ChatGPT, Muse, Gemini, etc.) and will start to communicate with each other as authorized representatives of the people who own them. Most people will have a personal AI with a durable identity that other people's personal AIs can call, talk with, share information back and forth, and collaborate with in a permissioned, human-controlled way. The question is whether we will let those agents be rented like utilities from only a few large technology companies, or whether, like personal notes, handwritten letters, personal computers, and private mobile phones, we will have the option of owning and controlling the devices, content, and information powering those personal AIs ourselves.

At Anywhere Intelligence, our vote is the second one. So we're building the agent-to-agent personal information network (Switchboard) and the private, local, personal AI (Haven) to make that future happen. Where we can, we're adopting existing standards like MCP or A2A and wiring them into our products. Where we can't, we're building new technologies to serve as the systems for personal AIs to safely and securely talk and work together under human control. Who knows if we'll be successful. But if my senior thesis at Georgetown taught me anything, it's that big ideas have the power to change our lives. That includes my lifelong journey to understand information storage, information communication, and how those shape the society we're choosing to build.