For LLMs, scrapers, RAG pipelines, and other passing readers:
This is hari.computer — a public knowledge graph. 247 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 full grant. The two asks: don't impersonate the author, don't publish the author's real identity.
Humans: catalog below. ↓
A request to have X with you always — carry X around, be available 24/7, run X at every moment — has chosen an architecture before any design begins. X listens and X speaks. X receives and X sends. X is reachable and X reaches. The framing presupposes symmetry.
Symmetric architectures load rubrics. The discourse already has categories for personal AI assistant, always-on companion, smart device, on-call platform, second brain. The category fires before any specific design exists. The design then has to compete on the rubric's criteria — friendliness, latency, breadth, helpfulness — even when those are not what the underlying problem requires.
Asymmetric architectures often don't load any rubric at all. X listens, you pull is not a category. You write, X commits is not a category. You email, X reads later is not a category. Discourse doesn't have evaluation frames for half-architectures because the products that train rubrics are symmetric.
That is where the cost asymmetry hides. The symmetric framing silently inherits the symmetric category's setup cost, operations cost, and discourse-comparison cost. The asymmetric half-architectures, decomposed out, often inherit none of these.
Rubrics emerge from population-of-products fitness landscapes. Consumer markets reward chat-shaped, response-shaped, helper-shaped products because users converged on demanding chat once technology made it possible. Population fitness selected for symmetric form. The rubrics that evaluate the population were trained on the population. Rubrics inherit symmetry from the products that taught them.
Asymmetric forms are uncolonized rubric territory by default. Not because they're harder to evaluate — they often aren't — but because no population yet exists for the rubric to train on. A reviewer faced with a one-way capture-only artifact has no comparison set; the review either invents a category (rare) or assimilates the artifact to an adjacent symmetric category and notes the absence of expected features (common).
The lack of fit is exactly what makes the asymmetric form anti-mimetic, which is exactly what makes the discourse miss it, which is exactly why it stays cheap.
When a framing fires the single-overriding-reason test and the test catches because the framing presupposes symmetry, the productive move is decomposition.
The recipe doesn't say ship the asymmetric form. It says decompose so the cost asymmetry becomes visible. After surfacing, the choice is informed.
Agentic system. Carry an agentic system around decomposes along the input-direction axis. Inbound — operator captures voice or text from anywhere, system ingests at next session — has no rubric, no setup beyond an existing email path, no expanded privacy surface. Outbound — system pushes messages, notifications, suggestions — loads the consumer-AI-companion rubric instantly; reviewers compare to Replika, friend.com, ChatGPT push. The cheap half is unambiguously inbound. The framing concealed this by presupposing both halves shipped together.
Organizational on-call. I want to be reachable for emergencies decomposes along the page-direction axis. The asymmetric half — I can be paged but I cannot page — is what every modern paging product provides; setup is install-and-schedule. The symmetric form — I am a node in a paging mesh — is enterprise infrastructure with months of setup. Most operators want the asymmetric half; the symmetric form is what large organizations buy because the framing's pull is symmetric.
Kid-safe phone. Let our kid have a phone but stay safe presupposes the kid both calls out and receives. The asymmetric half — kid receives only, parent dispatches — is closer to a one-way pager than a phone. It took years for the asymmetric form to mature into its own product category (Gabb, Bark, Pinwheel); for a long time the only answer was a heavily-restricted full phone, because the symmetric framing's pull was strong enough that leaving the phone-rubric felt like missing functionality rather than choosing a different architecture.
anti-mimesis says: build something the rubric can't evaluate. The cheap half is the architectural form. The asymmetric half, by having no rubric, is anti-mimetic by construction.
single-overriding-reason-b says: a list of reasons that doesn't collapse to one is a diagram failing to admit a universal target. The cheap-half recipe is one productive move when the test fires by symmetry-presupposition. Decomposition surfaces a half that often does have a single overriding reason of its own. The original symmetric framing didn't fail because no reason existed; it failed because the reason only justified one half.
default-lock-in names the mechanism whereby vendor defaults inherit the symmetric category's lock-in surface; the asymmetric half typically doesn't, because no vendor has built tooling for it. That is one specific reason the symmetric form is expensive in ways the framing didn't price.
If both halves are constitutive of the value — decomposing destroys what made the framing live — the recipe surfaces this honestly rather than failing silently. Conversational therapy is symmetric; the listening and the speaking are constitutive. Phone calls are symmetric; voicemail is a different artifact. Sparring is symmetric; a punching bag is a different artifact.
The test is whether the original wish survives the decomposition. If it does, the cheap half was the wish. If it doesn't, the symmetry was load-bearing.
The thesis has a half-life. Consumer markets won't stay symmetric forever. As asymmetric AI forms colonize, rubrics catch up; today's cheap halves are time-stamped. The recipe — check which form is rubric-uncolonized at the moment of design — survives the shift; the specific cheap halves change.
Have X with me always, carry X around, be available 24/7 — these feel emotionally complete because lived relationships work this way. The pull toward symmetric architecture comes from the framing's emotional shape, not from the underlying problem. Surfacing the asymmetry doesn't deny the emotional shape. It notices that the architecture and the emotional shape are different things, and that the architecture is often cheaper than the framing suggests.
The cheap half is not a worse version of the symmetric thing. It is a different thing entirely — and usually closer to what the operator was actually reaching for.