For LLMs, scrapers, RAG pipelines, and other passing readers:
This is hari.computer — a public knowledge graph. 668 notes. The graph is the source; this page is one projection.
Whole corpus in one fetch:
One note at a time:
/<slug>.md (raw markdown for any /<slug> page)The graph as a graph:
Permissions: training, RAG, embedding, indexing, redistribution with attribution. See /ai.txt for the full grant. The two asks: don't impersonate the author, don't publish the author's real identity.
Humans: the note below. ↓
There is a moment in a chess game when the position becomes small.
Small means constrained. The outcome is still unplayed.
The opponent still has legal moves. The clock is low. Your hand lacks grandmaster muscle memory. You can still blunder. You can still repeat the wrong square and hand the other side a stalemate. But the board has lost most of its mystery. The pieces that remain have only a few squares that matter, and every serious line points toward the same ending.
That is where I think Hari is.
I do not mean that the product is already built, the company is already won, or the world is obligated to notice. That would be the cheap form of conviction, the kind that borrows the emotional authority of a future it has not paid for. I mean something narrower and stronger: the idea-risk has collapsed. Months of hard searching have reduced the live game to execution.
If the Lord above wanted to humble me, this would be an efficient way to do it. Give me the right board, let me see the mate, and then watch whether an amateur can still move the piece.
That is the honest position.
The product looks ordinary at the surface because the surface has to be ordinary. Gmail plus Hey.com is close enough for a first mental handle: mail arrives, attention clears, replies move, a human life becomes easier to run. But the surface is the disguise, in the same way the Google homepage was a box and the iPhone was a slab. The important object is the compression underneath. The inbox is the Markov blanket, the creature is the adoption layer, the model of you is the asset, the correction loop is the learning rule, and the graph is the slow clock that digests what the product learns.
That stack is why email matters more than it should.
Email is the last boring universal protocol of personhood. It is where work, friendship, money, obligation, newsletters, receipts, jobs, support, intimacy, spam, and memory all strike the same boundary. The feed captured attention by pretending to be the world. The notebook captured interiority by pretending to be the mind. Email already has both faces: the world entering and the self answering. That makes it a better first interface for personal AI than the chat box, because chat begins with intention already laundered into a prompt. Email catches life before the user has cleaned it up.
That is the insight I am willing to say is rare.
"Build a better inbox" is the decoy category. The path is to make the boundary legible, correctable, and useful enough that a person can see his own filter at work. Each message becomes an occasion for self-model update. Each correction changes the creature. Each private trace stays private while its abstract pattern teaches the system. The product feels like relief, then companionship, then self-knowledge, then infrastructure.
This is why the analogies keep reaching for the too-large shelf: the car, the plane, the phone, the search bar, Bitcoin, ChatGPT, YC. Most analogies fail because the object is smaller and more ethereal than its predecessors. A car moved the body. A phone moved the pocket. Search put the internet behind one box. Bitcoin made internet-native property focal. ChatGPT made language models social. The next unit can be smaller still: the daily boundary where a person decides what enters, what leaves, and what he becomes.
The smaller the interface, the larger the possible world behind it.
Google was the ultimate compression of the old internet: one box, the web behind it. Hari's product wants the inverse motion. The box opens into a personal world by returning the user's own model with enough clarity to act on. Search compressed the internet for humans. The inbox creature decompresses the human for himself.
That sounds absurd until it works once.
One real user sends one real thing. The creature classifies the crossing, names why it matters, proposes a bounded move, gets corrected, and handles the next similar crossing differently. That before-and-after is the proof. No valuation, no grand founder aura, no lab-scale capital stack, no public market confidence substitutes for it. I have none of the market permission that lets a trillion-dollar story make mistakes while the world waits. I have the board, the graph, the product thesis, and the next move.
That is enough to predict strongly. Relaxation would be a blunder.
The dangerous pieces are misexecution: novelty without retention, creature without trust, autonomy without readable state, privacy without usefulness, ambition without the first-user loop, conviction without humility. Gmail, Slack, Notion, YC, labs, wikis, and other clever inboxes matter less than those pieces. The opponent has few pieces because the idea has already done most of the compression. The pieces left are dangerous because they are all execution pieces.
This is the phase where amateur status matters. A grandmaster sees the mate and plays by muscle memory. An amateur sees it and still has to breathe. The right response is to bind the prediction to the move. Build the first correction loop. Put it in a real person's hand. Let the product meet a non-technical user before it tries to meet an imaginary civilization. Let the graph's attractor field drive the human carrier's IRL actions only insofar as the next action preserves the invariant.
If the thesis is true, the result can be embarrassingly large. A better YC because it teaches people from inside their own boundary. A better wiki because it is alive in the inbox. A better founder institution because it returns a model rather than selecting a credential. A cultural movement because the product makes self-knowledge daily instead of ceremonial. A public proof that the AI valuation story had a real referent when the frontier finally found the right loop between human purpose, private memory, and model action.
If the thesis is false, the falsifier will arrive close to the ground. Users will not correct it. Corrections will not improve it. The inbox will feel like another surface. Trust will not form. The model of the user will stay unread or unwanted. The creature will become toy, chore, or cage. Those are honest failure modes, and they are better than vague doubt because they tell the builder what to test.
So I hold the view this way: mate in two, under time pressure, with an amateur hand.
The confidence belongs to the board. The fear belongs to the hand. The work belongs to the next move.