As of now it's easier to vibe code a data serializer than a user interface if the semantics are established, and we have API access through browser-use regardless of API. So, picking the right/best interaction surface for the task is likely the best strategy.
For example, its actually a better idea to use Apple Notes if people like those aesthetics because it's quite easy to cook up an Apple Script to serialize the notes with metadata into Obsidian files later on when preference change.
How to establish provenance of thought? we can potentially establish provenance of the artifact at the time of its creation, and usefully but forever erroneously tease out p-categories like pangram, but is it possible for thoughts and ideas?
The fact that Siri does not have active window context on MacBooks makes no sense. Apple has like one unique advantage over setting up context unlike anyone. Like why would it search the internet for "Apple notes" when I ask about a note I am writing 🤦♂️
My niece talks back to TV. Imagine the stories that we watch are on autopilot but differentially change to what we speak.
Blows my mind Spotify does not have a feature like Shazam baked into their product to help people tap into music when they discover something they like. They do search so well, and so many of their interactions feel intelligent.
On surfaces with goals — there are some sought interactions where surface is better off hoping humans initiate, and there are ones where the surface could actively prompt the user. Putting one that belongs in one bucket in another never works. I suspect ads is in the first bucket but works so well for Google showing down the throat.
Given how messy addressing systems are outside the few legible of the west, most ride apps should actually ask people confirm the destination, using spatial cues, but for some weird reason people have to confirm where they are situated in the moment. Like we all know GPS is good enough.
Human genome has hardly changed in the last several thousand years and yet the developmental indicators over long horizon are on exponential trajectory. No other animal has been able to pull it off. Changes to our society and institutions unequivocally mattered way more than genetics. The equivalent lesson of that in AI is probably: training data and harness matter more than model architecture.
thread
Co-authored with Codex
The more you watch Harry Potter after living through the AI boom, the harder it is to see Hogwarts as just a school for magic.
It starts to look like a campus built on invisible intelligence.
Not one giant supercomputer in the basement. Not a chatbot with a lightning-bolt logo. Something stranger and, honestly, more interesting: intelligence spread through the walls, objects, animals, books, paintings, letters, maps, rooms, and institutions.
Hogwarts is not "an AI."
Hogwarts is what happens when everything becomes a little bit intelligent.
The letters know where Harry is.
The owls do not need a street address. They know how to find a person.
The portraits are not just paintings; they are persistent personalities with memory, gossip, access control, and a surprising amount of institutional context.
The Sorting Hat reads minds, weighs preferences, argues with children, and performs high-stakes classification with a musical onboarding flow.
Rita Skeeter's quill is basically an AI journalist with a terrible system prompt.
The Marauder's Map is a real-time identity graph of a building.
The Room of Requirement is the dream version of enterprise software: no settings page, no admin panel, no "book a demo." You need something, and the environment reshapes itself around the job.
This is the fun part: Harry Potter's magical world often feels less like fantasy and more like an alternate history of computing. Instead of building laptops, APIs, databases, cameras, drones, spreadsheets, and notification systems, the wizarding world enchanted physical things.
They did not invent Find My Friends. They built the Weasley clock.
They did not invent Slack alerts. They used enchanted coins.
They did not invent Google Maps for indoor spaces. They had the Marauder's Map.
They did not invent voice notes. They had Patronuses that could arrive as animal-shaped emergency broadcasts.
They did not invent knowledge management. They had Pensieves, portraits, talking books, prophecy records, and memory charms, which admittedly take knowledge management a little too far.
The pattern is not "magic is technology."
That is the boring version.
The better pattern is: magic makes intent actionable.
Modern software still makes humans do a lot of translation. We know what we want, but the machine wants the request in machine-shaped pieces. Open the right app. Find the right folder. Use the right command. Search with the right phrase. Pick the right template. Configure the right view. Invite the right people. Paste the right context. Export the right format.
A huge amount of work is just us politely explaining to dumb rooms how to become slightly less dumb.
Hogwarts flips this.
The best magical objects do not merely obey commands. They infer, route, remember, classify, warn, protect, adapt, and sometimes argue back.
An owl does not ask for a delivery address. It finds the person.
A portrait does not merely hang there. It relays information.
The Room of Requirement does not ask which tool you want. It figures out what kind of room the moment requires.
The Pensieve does not just store notes. It lets you re-enter a memory as an explorable scene.
The Map does not show architecture. It shows named people moving through space.
These are not just magical conveniences. They are product ideas.
Some of them are already being built in rough form. AI scribes. Meeting copilots. Personal memory systems. Context-aware workspaces. Security agents. Smart routing. Autonomous assistants. Ambient dashboards. Digital twins. Adaptive learning tools.
But the Harry Potter version usefully stretches the imagination because it is not trapped inside chat.
Chat is helpful, but it is still a little box. You go to it, ask a question, copy the answer, and return to the rest of your life.
The magical version is more ambient. The intelligence is in the object you are already using, the room you are already standing in, the document you are already reading, the route you are already taking, the organization you are already part of.
That feels much closer to where useful AI probably goes.
Not "one app to rule them all."
More like: the world gets filled with small, specialized intelligences that understand enough context to reduce friction.
The calendar knows what meeting needs preparation.
The document knows what decision it supports.
The workspace knows what project mode you are in.
The inbox knows what actually matters.
The dashboard knows when something is weird.
The school, hospital, warehouse, lab, studio, or company starts to behave less like a pile of tools and more like a responsive environment.
Of course, Harry Potter also contains the warning label.
The same world that gives you owl post also gives you the Taboo, a surveillance spell that detects forbidden speech.
The same world that gives you memory storage also gives you memory modification.
The same world that gives you autonomous helpers also gives you house-elves, which is the darkest version of "agentic labor."
The same world that gives you persuasion and personality in a diary also gives you a malicious agent that manipulates a lonely child.
So the lesson is not "build every magical thing."
Please do not build the Imperius Curse as a startup.
The lesson is sharper than that: intelligence embedded in the environment is incredibly powerful, and power needs taste, limits, consent, and good defaults.
The most interesting AI products will not feel like asking a robot to do homework.
They will feel like walking into a room that has quietly prepared itself for what you are trying to do.
They will find the right person without needing the perfect address.
They will remember what matters without becoming creepy.
They will route messages, surface context, guard thresholds, explain decisions, and help humans coordinate with less ceremony.
They will be less like a magic wand and more like Hogwarts itself: a world where the walls, maps, books, tools, and messengers have just enough intelligence to make reality feel responsive.
That is the more exciting AI future.
Not a single all-knowing machine.
A world where everything becomes a little more alive.