In pre-industrial times most people's jobs made them strong. Now if you want to be strong, you work out. So there are still strong people, but only those who choose to be. It will be the same with writing. There will still be smart people, but only those who choose to be.
— Paul Graham, paulgraham.com/writes.html
Design isn't primarily about visual execution—it's what emerges when you deeply understand and solve the right problem. The visual form is almost a natural consequence of working through the problem structure at its primitives level.
there’s a shift from knowledge to cognitive work
knowledge work was about collecting information, organizing it, and synthesizing it ourselves to produce an output. thats a workflow from the classical software era.
but now, with AI being software, the work is changing. it’s less about doing the actual task yourself, and more about shaping how the AI will think about the task. the reasoning steps, the representations, the context you give it.
it’s like the work is: designing the thinking process; picking the right info, defining the steps, choosing the tools, shaping the reasoning path.
things like vibe-coding, agent-coding, deep research, all of that is us learning a new muscle. we’re not producing the work directly. we’re designing the structure that produces it.
so the real shift is: instead of creating knowledge, we’re creating the processes that generate knowledge.
Wrote this before joining Notion.
[1] The computer's potential is locked inside a metaphor. A metaphor that worked so well over the past couple decades that everyday folks struggle to distinguish it from the machine that's powering it. Our idea of what a computer can be is bound by a worldview where files live in desktops, and open in apps, each app with its own experience, with patterns that have crystallized into UI and UX best practices.
Fifty years ago, video conferencing on a computer (circa Engelbart) was an unbelievable idea made real. It was a great idea—and great ideas are hard to deviate from—and in this way, it stuck. Today, we live in a world where Zoom calls are as ubiquitous as they can be, and not that different from that first demo. This has been the case with most aspects of the metaphor: the desktop, the word processor, the app, (e)mail.
But behind the metaphor is a universal constructor that can be programmed to model anything.
[2] What fascinates me most about Notion is that the company seems to position itself as a lab and incubator for the future of computing. In simple terms, it aspires to be the Bell Labs of the 21st century—a place where a dedicated group of people explores new ways to express computation for everyday users.
[3] As a product, what fascinates me most about Notion is how it embodies some of the fundamental qualities of the machine, such as universality. It's a versatile tool that can be used to model and create other tools, much like the number system models any quantity or the alphabet captures any sound.
Each Notion page serves as a blank slate where information can be modeled. Each block is almost universal, functioning like an input-output system: the same information can be transformed and presented differently depending on the type of block used.
Looking at Notion as a whole, it's as if it aims to turn the computer into a white page rather than a desktop. Just as we can write, think, and draft ideas in written, visual, structured, or unstructured forms on a page, we can do the same on a Notion page.
A new computing metaphor seems to be emerging from within Notion: an operating system modeled as a white page—creative, malleable, and capable of taking any form we desire. Both a physicist and an artist can find unique uses for this adaptable page.
This is a more liberating and creative metaphor for computing. In this sense, I see Notion less as a pure knowledge product and more as tomorrow's operating system—a potential competitor to macOS, Windows, and even Linux. A universal, web-based operating system that mirrors the creativity inherent to humans, the toolmakers.
[4] Finally, this brings me to AI.
If Notion is poised to become the birthplace of a new computing metaphor, how does computational thinking integrate into this white page? How can this white page come alive with artificial thought?
Language models are a fundamentally new building block for creating software, and we are still exploring how to package them into experiences that people can relate to.
My hunch is that we are in the early stages of adopting LLMs, where they have primarily been made useful for information retrieval. This search paradigm is embodied by ChatGPT's search-powered results, as well as by Perplexity and Notion's conversational search.
[5] Here I enter pure conjecture.
My intuition is that AI-powered software has latent potential to move beyond the realm of pure appliance—like Google Docs storing text as a fridge stores food—and enter a new phase where digital products become cognitive: cognitive products.
Concretely, within the context of the white page computing metaphor, some intuitions on what this could mean:
Most interestingly, an experience built directly into the white page rather than as an auxiliary feature—allowing the page to transcend its identity as a static document and become a dynamic drawing board where elements can be added or removed, much like opening and closing apps on a desktop as needed for specific tasks.
Imagine if you could get yourself an AI model as an object. A self-contained unit that plugs into your phone, ensuring minimal latency, and remembers and learns from every conversation, idea, and bit you tell it—but this memory never leaves the device.
If plugged into any device other than its parent more than 3 times (lost or stolen), the data self-destructs. Would you be more comfortable with it knowing more about you if you owned it, and your data wasn't in the cloud?
A biological brain is a plastic, moldable, changing structure that models itself continuously. An AI model, on the other hand, is a static set of relationships—a remarkable, albeit fixed, compression of its training data.
When we say AI super intelligence do we mean model plasticity? A model and system that continuously remodels itself? Neurons whose strengths, weaknesses, and connections change over time in response to stimuli, input, and new data?
An AI model might encompass more points of view (relationships) than there are humans in the world, yet a single human possesses the capacity to learn and generate infinitely new points of view in a way that models could not.
To make an original contribution, you have to be irrationally obsessed with something.