Birth of the Intelligence Substrate

I've been thinking about the mycorrhizal network — the intelligent strata of the biological world — as an analogue to the intelligence abstraction layer that's emerging.

I recently wrote about our team’s mission in my final days at Microsoft. I’d spent the prior two years working on “applied intelligence” — services powered by personal assistants and machine learning that understood your preferences and your active context. Our final mandate was to roll out a service connecting every app and service at Microsoft: an intelligence layer, planned and executed in anticipation of the next-gen assistant everyone assumed was around the corner — the Cortana 2.0 or Siri 2.0. It didn’t arrive. At least not for another decade.

I worked on quite a few vision projects in those years — prototypes that simulated the promise of what was to come. One combined sensors, home and vehicle connectivity, and access to your email and calendar to simulate the nuanced awareness of a human personal assistant. A change in the environment, like someone new entering the room, would recalibrate an active conversation and change tack if needed, the way a real person would. I wrote a paper on “intelligent service etiquette” covering the courtesies we’d expect. The service was largely theoretical, but the simulations exposed a real need for machine politeness.

My Sixth Sense project (2012) explored subconscious notification — think spidey-sense. Contextual awareness, deep personalization, and intelligence together make serendipity something you can design for.

Over the past couple of years, I’ve watched those vision projects become viable — and not only for large, well-funded teams. If you casually use Claude, ChatGPT, Gemini, or others, you’ve likely started wading into this paradigm without noticing: connecting your Google Drive, granting access to your hard drive, even your email. The boundaries are getting blurrier, and each new grant of access unlocks a more complete view of who you are, for better or worse.

I have real concerns about handing intimate access to the world’s largest corporations — which is partly why the experiment I’ll share later was built on my own terms. But first, some grounding.

What’s an abstraction layer?

An abstraction layer hides the complexity below so something new can be built on top. Life has been doing this since the beginning. You don’t have to think about your heart beating or remember to breathe — that’s a biological abstraction layer. A forest doesn’t manage its own nutrient exchange; the mycorrhizal network beneath it does, quietly routing resources and signals between trees. When you drive, you don’t need to understand how the engine works. When you send a text, you don’t need to understand data packets. Every layer takes something complex, makes it invisible, and removes the burden. Abstraction is the mechanism by which complexity compounds.

What’s emerging from this “AI era” is a new layer in the technology stack, and what’s being abstracted isn’t mechanical or informational. It’s effort. The cognitive overhead of operating an interface.

Since Douglas Engelbart’s 1968 “Mother of All Demos” we’ve largely been refining the same paradigm — better tools to control computers.

As an interaction designer, I’ve spent thirty years working on that overhead. The primary job of design is transformation — changing an existing situation into a preferred one, as Herbert Simon put it. When it works, the experience feels intuitive. But even the best designs never completely removed the burden; they just made it lighter. You still had to learn and hold the mental model.

Writing code is now ground zero for where rolls and functions are being abstracted. Andrej Karpathy, co-founder of OpenAI, wrote that the profession is being dramatically refactored as “the bits contributed by the programmer are increasingly sparse and between.” Boris Cherny, creator of Claude Code, put it more simply: “Once the plan is good, the code is good.” Translating ideas into a language machines understand — the central labor of engineering — is being absorbed by the layer. This isn’t AI hype; I’m watching it happen firsthand.

For my project Attention Management (2011), my illustration of contextual intelligence as an opportunity to increase relevance and reduce noise.

The intelligence substrate

At Microsoft, we called this layer the “Intelligence Substrate.” I like to think of it as a surface endowed with ability and sensibility: you express what you want, and it works out the rest. Earlier layers did this too, but mechanically — a compiler, a command line, the first voice assistants all ran on fixed translation, a known input to a predetermined output. This one behaves more like the layers biology builds. It doesn’t execute your intent; it interprets it, filling in what you didn’t say. Either way, the value is the same — freeing your attention.

For design, this might be the biggest shift since the graphical interface. If the interface no longer needs to teach a person how to operate the machine, design stops being about arranging complexity and becomes something closer to defining relationships. What should the layer handle without asking? When should it check in? How does it show its work when you want to see it, and disappear when you don’t? How does something novel feel familiar immediately? These are some of the questions I’m now exploring — and they’re more human than technical. They’re largely about earning trust.

I’ve been working out some of these questions both in my work, which centers around AI innovation, and in my free time, where I’m thinking a little further out. The creative process feels different, and I’ve been trying to wrap my head around what it means. The old cost structure of building — where every idea had to justify weeks of work before you could feel whether it was right — has collapsed. My process is now: imagine what’s possible, build it, try it on, and toss it if it doesn’t fit. Like sketching, except the sketches work. I’ve built and discarded more ideas in the past year than I’ve build in my entire career. When trying something costs almost nothing, taste gets to lead and intention gets tested against the real thing.

The hardest and most interesting problem has been designing form factors that don’t exist yet. There’s no pattern library for a website that’s proactive. No convention for how inferred intentions should be handled. Every choice is a negotiation between new and familiar — push too far into the new and it’s alienating; hold too close to the familiar and you’ve just rebuilt the old thing with extra steps. Finding that line, over and over, has been the real work.

Building on the substrate

My site is now a playground for building on top of an intelligence layer. Every action changes the path and potential of what's next. Over time, the site learns enough to bring what's relevant to you — but only if you want that.

Here’s the experiment I mentioned. My personal site used to run on a content management system — templates, admin panels, a small bureaucracy of forms I operated every time I wanted to share something. Most of what I did there was maintenance, and maintenance is the part of any system nobody wants to own. The new site has no management interface at all. When I finish a piece of writing, I tell my helper it exists, and it takes care of the rest. My helper lives in a Mac Mini sitting on my desk, and sends me updates every morning.

Imagine harder is my mantra for my helper who lives in this little box. My Pico Mac Nano and Ganesh keep it company, although I draw the line at anthropomorphized AI.

What it does now is an order of magnitude more involved than anything I used to do. It finds relationships to other content on the site and explores the themes and ideas that connect them. It considers which of my known visitors might enjoy the piece — people who have chosen to be known, on terms they set — and may decide to let them know. It suggests ideas for sharing the content externally, acting as a kind of social media manager. And it prepares the content to be crawled by other agents, on my terms rather than theirs. The layer holds the complexity. I hold the intent.

The movie Inception comes to mind as each new layer of previously unreachable capability opens up. In the film, every dream nested inside another gets weirder and more dangerous, and the deeper you go, the harder it is to find your way back. Is this an untethered departure from reality? Or is it unlocking fantastic new realms that stretch what’s possible? Both, probably.

Christopher Nolan’s hand drawn map of dreams-within-dreams.

Where it’s headed

I don’t think this layer of intelligence stays in the cloud, or in the hands of the companies that got there first. Models are already running on devices and at the edge, and the more capable they get, the less reason there is to send your most valuable resources to corporations. The surface itself becomes intelligent — a form that helps you fill itself in, a site that knows why you came. The interface stops being a place you go and becomes something that comes to you, carrying what you need and leaving the rest behind. Most of today’s AI investment is still directed at legacy problem spaces. The interesting work is building natively on the new layer, and deciding, deliberately, who it serves.

Since my time at Microsoft, I’ve been obsessed with digital sovereignty — reclaiming rightful ownership of our experiences, our creations, our lives, and our world. I struggle with the current state of technology, and I also see this moment as the birth of something new: a connective layer activating pathways between things that couldn’t talk to each other before, the way the network under a forest floor does. What grows on it depends on who’s tending it.

Resources & references

My coding primer — I believe Cloudflare is the best stack for building AI-native experiences, and this teaches you everything you need to know.
Steve Yegge — both enjoyable to read and always two steps ahead.
Blaise Agüera y Arcas, What Is Intelligence — inspired some of the thinking about the biology abstraction continuum.
James Bridle, Ways of Being — before Blaise’s book, this book opened my eyes to a more connected continuum (thanks Jenny!)
Itamar Friedman, Software 3.0 — another take on the intelligence substrate
Christopher Alexander, A Pattern Language — my introduction to abstraction and design for human thriving
Stewart Brand, Maintenance: Of Everything — a bit of an anti-pattern, but central to this thesis (I also love Stuart Brand :)
Center for Humane Technology — principles to keep us in the drivers seat