Lens · Location
The map Pages placed by their links, regions found by community detection over the wikilinks alone. A region is named by what its pages keep citing.
Hover a page for its neighbourhood; click a region below to isolate it. The layout is computed at build time, so the same page sits in the same place tomorrow.
The AI Daily Brief — Why Agents Still Need Humans Every Method Reflection Legibility in the Lens, Not the Land Working Heuristics — Data, Matching, Product, Org, Tooling Data Contract Verification Requires Domain Knowledge Method Reflection Agent Autonomy Spectrum Agent Autonomy Spectrum
The range of how hands-off you are with an agent — from turn-based "assisted AI" through semi-synchronous harness work to mega-autonomy — with NLW arguing the productive sweet spot is settling in the middle. Agent-Maintained Wiki · auto-ingested
A set of linked pages that an AI agent writes and maintains about a body of material. Each page gathers what is known about one subject and links to related pages, so the agent has a map for finding knowledge rather than re-searching raw files. It is the pattern from Karpathy's LLM-wiki gist, and the pattern this wiki is built on. Anthropic — The Components of Agent Memory evaluates it; Blumenfeld — Scaling Karpathy's LLM Wiki (Neo4j) attacks its scaling. Agent Memory · auto-ingested
Knowledge an AI agent gains from experience (its work with users and on tasks) and keeps between sessions as readable context, such as a folder of notes. It is distinct from weight updates, in-context learning and compaction (Anthropic — The Components of Agent Memory). Agents as Employees vs Human-Agent Collaboration
Dan Shipper's taxonomy of the two ways teams actually work with agents. The discourse predicted the first; the second is "stranger and, in my experience, more important." AI-Accelerated Science · auto-ingested
How much AI could speed up research. The gain depends less on raw capability than on bottlenecks AI can't remove: physical work, AI's cost and possibly ideation. It also depends on how fast researchers and institutions adopt AI. (Anthropic — How Scientists Use Claude) AI and Employment — Replacement vs Expansion
The open debate over whether agentic AI is a net destroyer or net creator of expert jobs. NLW argues expansion; the episode preserves a real replacement camp it does not refute. AI Commoditizes Expertise AI Commoditizes Expertise
Dan Shipper's mechanism for why automation increases expert human work: LLMs cheapen yesterday's competence, abundance breeds sameness, and sameness creates demand for the one thing models can't supply — human difference. AI Use Tracks AI Capability · auto-ingested
Across tasks and fields, people use AI in proportion to how well AI can do the work. Tasks that are cognitive, quickly verifiable and suited to reinforcement learning get far more use than the time humans spend on them would predict. Physical, interpersonal and regulated tasks get little. (Anthropic — How Scientists Use Claude) AIDB — Nine Tips for Getting the Most Out of Codex · to-read
#to-read — The The AI Daily Brief episode (earlier the same week as The AI Daily Brief — Why Agents Still Need Humans) summarizing Jason Liu (OpenAI)'s nine tips for Codex. Not yet ingested. AIDB — The New Jobs AI Will Create · to-read
#to-read — The The AI Daily Brief episode where NLW introduced the "human premium": seven categories of value that don't transfer when you remove the human. Not yet ingested. AIDB — Why Agents Make Every Job a Startup · to-read
#to-read — The The AI Daily Brief episode where NLW introduced The Infinite Backlog. Not yet ingested. Alexander — A City Is Not a Tree
Christopher Alexander's 1965 essay (Architectural Forum 122:1–2). It argues that cities that grew on their own have an overlapping structure, a semilattice, while planned cities are trees, and that the mind instinctively flattens overlap into trees. Agent-fetched 2026-11-08 to ground the front door (Legibility in the Lens, Not the Land). Alexander — The Origins of Pattern Theory (OOPSLA 1996)
Christopher Alexander's keynote to software designers (San Jose, 1996; printed in IEEE Software 16(5), 1999). It covers what pattern languages were for, the deeper theory that replaced them (centers, wholeness, structure-preserving unfolding), and a challenge to programmers. Agent-fetched 2026-11-08 to ground the front door. Allowlist over Blocklist
For sensitive data, name what you keep and drop everything else (allowlist), rather than enumerating what to exclude (blocklist). An allowlist fails closed; a blocklist leaks by omission the moment a new sensitive field appears that nobody remembered to ban. Anthropic
AI lab; maker of the Claude model family and the Claude Code harness, and the model base for agents like Manus. Led by CEO Dario Amodei. Anthropic — How Scientists Use Claude · auto-ingested
Anthropic Economics report on 207,049 privacy-preserved research sessions on Claude, Claude Code and Claude Science. Scientists use AI where it works best: mostly for data analysis and writing, rarely for the ideation they rank most important. Scenarios put research acceleration at about 2× with today's AI and up to about 15× if all non-physical work is automated. _Subscribed-feed capture (Anthropic Research), auto-fetched and auto-ingested 2026-11-08; awaiting human review._ Anthropic — Investigating Vault Particles with Claude · auto-ingested
A result from Anthropic's in-house biology lab: the first 3D structure of a vault particle in prokaryotes, with Claude used at every step — literature and sequence search, DNA construct design, purification-method optimization, and cryo-EM processing. The lab-notebook detail matters more to this wiki than the biology: it is the wiki's first worked case of AI-accelerated research, including where Claude was wrong. _Subscribed-feed capture (Anthropic Research), auto-fetched and auto-ingested 2026-11-08; awaiting human review._ Anthropic — Progress on Provable Inference for Large Language Models · auto-ingested
Anthropic's Frontier Red Team releases FlashProve, an open-source prototype that attaches a cryptographic proof (a zkSNARK) to a model's output, showing the response came from a committed set of weights without revealing them. Claude built much of it; Lean proofs back the security claims. It is the wiki's first source on verifying what an AI did without trusting or re-running it, which makes it a counterpoint to Verification Requires Domain Knowledge. _Subscribed-feed capture (Anthropic Research), auto-fetched and auto-ingested 2026-11-08; awaiting human review._ Anthropic — Teaching AI to Build Scientific Models · auto-ingested
A guest post on Anthropic's research blog by three AI for Science grantees. They argue that AI can now build, constrain and partly interpret scientific models, but a single best-fitting model gives "an answer, but… not understanding." What AI should deliver is a distribution of plausible models plus an account of which alternatives failed. _Subscribed-feed capture (Anthropic Research), auto-fetched and auto-ingested 2026-11-08; awaiting human review._ Anthropic — The Components of Agent Memory Anthropic — The Components of Agent Memory · auto-ingested
Anthropic Engineering post on how Claude decides what to remember between sessions, and how offline "dreaming" corrects, restructures and prunes that memory. It includes an evaluation of agent-maintained wikis, the same pattern this wiki runs on. _Subscribed-feed capture (Anthropic Research), auto-fetched and auto-ingested 2026-11-08; awaiting human review._ Ashby — An Introduction to Cybernetics
W. Ross Ashby's 1956 textbook, the standard introduction to cybernetics after Wiener. Read here for Chapter 1 (What is new) and Chapter 11 (Requisite variety), which show regulation to be a communication problem: a regulator can control no more than its channel can carry. Agent-fetched 2026-11-09 during the human-requested daily pass, to ground Communication and Control Are Intertwined. The full text is in raw/; only these two chapters were read. Blumenfeld — Scaling Karpathy's LLM Wiki (Neo4j)
A Neo4j engineer's argument that the LLM-wiki pattern fails at scale on read, search and navigation, not on writing, and that the fix is to index the markdown as a graph (documents, sections, containment, reading order, links) so "a hop is a pointer rather than a scan." Agent-fetched 2026-11-08 at the human's request after a discussion of the article. Build It as an Extension
A new capability is adopted fastest when it reads as an extension of something the user already knows — place it where their mental model already puts the data, and reuse the nearest proven mechanism, so it isn't "a new thing to learn." Build With, Not Against
A stance for working across team boundaries: when two teams' work overlaps, the answer is almost never to out-build the other side. Draw the seam, concede where they're better, follow what's best for the customer, and make your durable layer their extension — not a land-grab. Bush — As We May Think · to-read
Vannevar Bush's 1945 Atlantic essay proposing the Memex: a personal, curated store of documents joined by "associative trails" the owner builds and follows. Choose the Platform by the Hardest Recurring Part
Pick the platform that best handles the hardest, most-recurring part of the workflow — not the part that's cheapest to prototype. The demo is a one-time cost; the hard recurring step is paid forever. Christopher Alexander
Architect and theorist (1936–2022). Author of A Pattern Language, The Timeless Way of Building and The Nature of Order. His pattern languages became the root of software design patterns. Claude Code
Anthropic's agentic coding/work harness, cited alongside Codex as a shared workspace for human-agent collaboration. The episode also references "Claude Cowork" / Coworker as a related collaborative surface. Clichés Carry Less Information
Norbert Wiener's information-theoretic aesthetics: predictable, conventional language literally contains less information than innovative, idiosyncratic expression — so originality is, in a precise sense, higher-information. A ~1950 antecedent of today's debate about AI Slop. Codex
OpenAI's agentic coding/work harness — one of the tools anchoring the semi-synchronous middle of the Agent Autonomy Spectrum. A full-featured Codex now lives inside the ChatGPT app, including mobile. Communication and Control Are Intertwined
Control is exercised through communication, and a controller can control no more than its channel can carry. Regulation, read in cybernetic terms, is a communication problem (Ashby — An Introduction to Cybernetics). Control-Flow Integrity
A software-security property/technique that prevents attackers from hijacking a program's execution by enforcing that every control transfer follows a path of a control-flow graph (CFG) fixed ahead of time. Transfers the CFG doesn't sanction are blocked at runtime. Control-Flow Integrity — Abadi et al.
The foundational paper that named and formalized Control-Flow Integrity. Journal version (ACM TISSEC, 2009) of work first presented at ACM CCS 2005 (and ICFEM 2005). Auto-grounded 2026-06-20 to back the Control-Flow Integrity concept page (fetched from the author's faculty page). Critical Thinking Is Domain-Bound
Critical thinking is not a portable skill that, once learned, can be used on anything. The processes of thinking are intertwined with the content of thought: you can only think critically about what you know enough about, and only when something cues you that this is the moment to do it (Willingham — Critical Thinking, Why Is It So Hard to Teach). Dan Shipper
CEO of Every and author of the essay "After Automation", the primary text behind The AI Daily Brief — Why Agents Still Need Humans. Dan Shipper — After Automation · to-read
#to-read — The Every essay by Dan Shipper that supplies roughly half of The AI Daily Brief — Why Agents Still Need Humans. Not yet ingested: we have only NLW's read-aloud excerpts, not the primary text. Daniel T. Willingham
Cognitive psychologist at the University of Virginia who translates cognitive science for teachers. He wrote the long-running "Ask the Cognitive Scientist" column in American Educator. His research focuses on the role of consciousness in learning (Willingham — Critical Thinking, Why Is It So Hard to Teach). Dario Amodei
CEO of Anthropic; cited as a lead voice in the replacement camp of AI and Employment — Replacement vs Expansion. Data Contract
The agreed interface between a data producer and its consumers — schema, units, semantics, and the derivations performed at the boundary. The governing rule: derive once, at the contract, so every consumer inherits a correct value instead of re-deriving (and re-bugging) it. Data Gravity
Large data is expensive to move, so it exerts a "gravity" that should pull compute toward it: move the small thing (the query, the spec) to the data — not the data to the app. Databricks — What is Medallion Architecture
Databricks' canonical description of the medallion (multi-hop) lakehouse pattern. Agent-fetched 2026-06-26 to ground the Medallion Architecture entity. Distributions over Best Fits
Many parameter settings, and many model structures, explain the same data equally well. A single best fit hides all the others. The useful output is the range of models that fit, what they share, where they disagree, and which alternatives failed. That record, not the winning model, is where the insight is (Anthropic — Teaching AI to Build Scientific Models). Do It Once, at the Right Place Do It Once, at the Right Place
A meta-principle behind several of the wiki's engineering heuristics: re-doing work — re-deriving, re-computing, re-confirming, re-churning — is the waste to design out. Do the work once, at the place (or time) where the result is correct for everyone downstream, and let them inherit it instead of redoing it. Dreaming (Offline Memory Consolidation) · auto-ingested
An agent uses idle time between tasks to reflect over its logs and transcripts. It corrects wrong notes, builds and prunes representations, promotes key notes into the always-loaded layer, and fills gaps a single session left behind. The model is sleep replay in the hippocampus, and the work builds on Letta/UC Berkeley's "sleep-time compute" (Anthropic — The Components of Agent Memory). Ethan Mollick
Wharton professor and author of the One Useful Thing newsletter; a leading writer on AI's practical impact on work and organizations. Author of Making AI Work — Leadership, Lab, and Crowd. Every
An AI-native company — "part publication, part product company, part consultancy" — used throughout The AI Daily Brief — Why Agents Still Need Humans as the lead case study for how agent-saturated work actually plays out. Led by CEO Dan Shipper. Falsification Is Asymmetric
One counterexample can break an "always" claim, but no finite number of successes proves a "never." So: a single probe disproves "always," not "never" — and before asserting a categorical negative, sample across the spectrum rather than generalizing from one case. Front Door Pattern Language
The generative sequence for the wiki's front door, a static SvelteKit app built from wiki/. Six patterns applied in order, each fitting the context the last one left (Pattern Language). Written 2026-11-08 from Legibility in the Lens, Not the Land before any code, so the thin slice can be generated from it and the language judged by what it produces. Human-in-the-Loop by Exception Human-in-the-Loop by Exception
How to design the seam between automation and people: automate the structured bulk, reserve humans for true exceptions and confirmation, pay each human decision only once, and make every confirmation teach the system. Humans are the scarce, high-judgment resource — spend them only where judgment is actually required. Individual vs Organizational AI Gains
Ethan Mollick's productivity paradox: AI demonstrably boosts individual performance and adoption is high, yet organizational performance and macro wage/hours data barely move — because individual gains don't translate to organizational gains by default. James C. Scott · to-read
Political scientist and anthropologist (1936–2024). Author of Seeing Like a State (1998), the origin of the Legibility concept. Jean E. Fox Tree · to-read
One-line definition: psycholinguist (UC Santa Cruz) studying how spontaneous conversational devices — filled pauses, discourse markers, "uh"/"um" — shape comprehension and coordination in dialogue. Karpathy — LLM Wiki Gist
Andrej Karpathy's ~2,000-word "idea file" (April 2026) for a personal knowledge base an LLM builds and maintains: immutable raw/, an LLM-owned wiki/, and a schema file that disciplines the agent. This wiki is an instance of it, and CLAUDE.md descends from it almost line by line (Agent-Maintained Wiki). Agent-fetched 2026-11-08 at the human's request; it had been a #to-read lead for one day. Kimball — Declaring the Grain
Ralph Kimball's foundational Design Tip on grain — the canonical source behind One Fact Per Grain. Agent-fetched 2026-06-26. Lands vs Lives
Two different questions about the same data: where it lands (the one physical base table the bytes are actually written to) versus where it lives (the layer that gives them business meaning). In a view-only medallion these can be different objects, and confusing them sends you to the wrong place. LATCH
Richard Saul Wurman's claim that information can be organized in only five ways: Location, Alphabet, Time, Category, Hierarchy. "Information may be infinite, however... The organization of information is finite" (Wurman — Five Hat Racks and LATCH; secondary source). Leadership, Lab, and Crowd
Ethan Mollick's framework for turning AI's individual productivity gains into organizational ones — by coordinating three roles. The premise: there is no playbook, so advantage goes to whoever learns fastest. Legibility
Making a complex reality readable to a central eye by simplifying it. That often destroys the illegible, local workings that made it function. "The deep failure… is the mistaken assumption that thriving, successful and functional realities must necessarily be legible" (Rao — A Big Little Idea Called Legibility, after James C. Scott). Legibility in the Lens, Not the Land Legibility in the Lens, Not the Land
The founding principle for the wiki's front door. Organize the views freely; never reorganize the knowledge to suit a view. Wurman, Rao and Alexander look contradictory on legibility, and they reconcile once you separate the lens (how the reader sees) from the land (what is there). Living Structure
Christopher Alexander's later theory. Wholeness is made of centers, and centers are recursively made of other centers. Structure comes alive only through structure-preserving transformations: steps that each keep the whole intact while adding one differentiation (Alexander — The Origins of Pattern Theory (OOPSLA 1996)). Making AI Work: Leadership, Lab, and Crowd
Ethan Mollick's One Useful Thing essay (2025-05-22) arguing that AI's individual productivity gains don't become organizational gains without deliberate organizational innovation — coordinated across three roles: Leadership, Lab, and Crowd. Manus
An autonomous AI agent "based on Claude" (Anthropic), used by Ethan Mollick as a reference point for what agentic systems can do. Match Process Weight to Stage and Stakes
Dial the weight of process (planning, documentation, review gates, ceremony) to a project's stage and stakes: light discovery loops early, ceremony added only where a specific pain appears, heavy process only when coordination or correctness risk exceeds the ritual's cost. Grounded in part by Rao on Legibility; the stage × stakes dial itself is still a hunch. McCrory — Data Gravity in the Clouds
The 2010 blog post that coined "data gravity." Agent-fetched 2026-06-26 to ground Data Gravity, which had been resting on the human's heuristic alone. Medallion Architecture
A layered data-design pattern that refines data through successive named tiers — conventionally bronze (raw landing) → silver (cleaned/conformed) → gold (business-level, modeled) — so each layer has a clear contract with the next. Metacognitive Strategies
Small, learnable chunks of knowledge about how to steer your own thinking, such as "look for a problem's deep structure," "consider both sides," or "don't accept the first reasonable-sounding solution." They make good thinking more likely, but they tell you what to do, not how. The how comes from domain knowledge and practice (Willingham — Critical Thinking, Why Is It So Hard to Teach). Method Reflection Method Reflection
The wiki looking at itself. Overview maps what the wiki knows; this page tracks how it knows, and where its sources argue that method is wrong, incomplete or working. It is a second home page: start here when you want to improve the wiki rather than read it. Maintenance rule: on every INGEST, LINT or QUERY, the agent asks whether the material bears on how this wiki captures, stores or uses knowledge. If it does, it updates a tension below (or opens a new one) and logs it. Tensions are never silently resolved. They close only with a recorded verdict. Navigation vs Search
Two ways to retrieve from a body of linked notes. Search matches a phrase you already have against every page and returns a ranked list. Navigation follows the structure the notes already carry (links, backlinks, sections, containment) one hop at a time. They answer different questions, and a wiki needs both (Blumenfeld — Scaling Karpathy's LLM Wiki (Neo4j), Anthropic — The Components of Agent Memory). NLW (Nathaniel Whittemore)
Host of The AI Daily Brief and author of this episode's framing. Also founder of Superintelligent (an AI-maturity-assessment company, plugged as a sponsor in the episode). Norbert Wiener
American mathematician (1894–1964) who founded cybernetics — "control and communication in the animal and the machine." In the wiki via a secondary commentary on his The Human Use of Human Beings (1950). One Fact Per Grain
A dimensional-modeling discipline: every fact or dimension table is defined by a single grain (the level its key identifies), and you never widen it with attributes that don't share that key. A corollary: a table can only answer a question if it actually carries the rows that question needs. One Useful Thing
Ethan Mollick's newsletter on the practical use of AI (oneusefulthing.org). A recurring source channel for this wiki — individual essays are ingested as their own source pages. OpenAI
AI lab; maker of ChatGPT and the Codex agentic harness. OpenClaw
The early-2026 paradigm (and tool) for maximally autonomous personal agents: run an agent 24/7 on a dedicated Mac mini/studio, interface with it via Telegram, and let it drive itself with "heartbeats." Now widely seen as overshooting the useful level of autonomy. Pattern Language
A connected set of design patterns meant to be used in sequence by the people who live with the result. It is supposed to generate a coherent whole, not just store good ideas (Alexander — The Origins of Pattern Theory (OOPSLA 1996)). Personal vs Team Agents
Every's reversal on how to deploy agents in an organization: from personal agents (one per employee, a digital replica of them) to shared team / "unit" agents that reliably perform parts of many people's jobs. Pipeline Legibility
Design a pipeline so the next reader can tell what's actually happening from what they can see — name objects for the direction data flows, trust only the lineage you can read, and build ingestion to the source's real export contract. The opposite is a pipeline whose names and assumptions quietly lie. Placement and Judgment Over Raw Execution
The wiki now spans two domains that look unrelated — AI and work (how agents change knowledge work) and engineering craft (how to build data products). Yet they keep landing on the same three structural moves. The reason they rhyme: both are responses to execution getting cheap. When doing the work — compute, AI output, raw labor — is abundant, the scarce, value-determining inputs are where you place the work and where you spend human judgment. Execution is the commodity; placement and judgment are the craft. Provable Inference · auto-ingested
Attaching a short, cheap-to-check proof to an AI model's output that shows it was produced by a specific, committed set of weights, without revealing the weights and without the checker re-running the model (Anthropic — Progress on Provable Inference for Large Language Models). Anthropic now calls the broader goal inference verification. Rao — A Big Little Idea Called Legibility
Venkatesh Rao's Ribbonfarm introduction to James C. Scott's Seeing Like a State. A recurring failure: an authority can't read a complex reality, calls it irrational, and imposes a simplified order that kills what worked. Agent-fetched 2026-11-08 to ground the front door and Rao's #to-read lead. Richard Saul Wurman · to-read
Architect and graphic designer who coined "information architect" (1976) and founded TED (1984). Author of Information Anxiety (1989) and Information Architects (1996). Sanderson — The Rise of Data Contracts
The article widely credited with defining the modern data-contract movement. Agent-fetched 2026-06-26 to ground Data Contract. Scope Hygiene for Tickets and Tooling
The craft of keeping work-tracking and shared tooling clean: scope tickets so they don't block each other or duplicate, manage cross-ticket dependencies instead of refactoring through them, keep relationship links pointed the right way, and place shared tools and project names where the people who need them will find them. Secret Cyborgs
Ethan Mollick's term for workers who use AI productively but hide it — the gap between ~20% official tool adoption and 40%+ who privately admit large AI-driven gains. Shrink the Candidate Space
When the join key is unreliable, the leverage is in reducing how many pairs you even consider — not in making a smarter matcher decide each pair. Cut the candidate space first; tune the comparison second. Slop
Dan Shipper's definition: slop is visible sameness repeated ad nauseam — the default output when many people use the same model, trained on the same corpus, "without thinking too hard." Surface vs Deep Structure
A problem's surface structure is its scenario (bands, gardens, caves). Its deep structure is the principle that solves it (least common multiple, "leave a trail"). People encode and retrieve by surface, so a known solution often fails to transfer to a problem that looks different but is the same underneath (Willingham — Critical Thinking, Why Is It So Hard to Teach). System of Record
The authoritative source for a given fact — the one a conflict defers to. The governing caution: a convenient source isn't automatically a system of record. Verify authority before you promote a source to source-of-truth (SOT). The AI Daily Brief
A daily podcast and video on the most important news and discussions in AI, hosted by NLW (aidailybrief.ai). A recurring source channel for this wiki — individual episodes are ingested as their own source pages. The AI Daily Brief — Why Agents Still Need Humans The AI Daily Brief — Why Agents Still Need Humans
A "long read / big think" episode of The AI Daily Brief (host NLW) arguing that as agents automate more work, the volume of expert human work expands rather than collapses — built largely around Dan Shipper's essay "After Automation" and Every's operating experience. The Borges Map
A knowledge store that grows as large as the sources it summarizes "effectively abstracted nothing." It is a map at 1:1 scale: complete, and useless. Named for Borges's one-paragraph story "On Exactitude in Science," whose cartographers drew a map the size of the empire. The Bottleneck Shift The Bottleneck Shift
When AI makes execution cheap, the scarce input shifts from doing the work to deciding what work to do. Articulated by Ethan Mollick and independently by Dan Shipper — the strongest cross-source convergence in the wiki so far. The Human Sandwich The Human Sandwich
The work pattern for complex agent collaboration, coined by an Every employee: humans are the bread on either side of the AI's work — a human sets the frame, the AI collapses the task, and a human judges and extends the result. The Human Use of Human Beings — Wiener (excerpt)
A short third-party commentary on Norbert Wiener's The Human Use of Human Beings (1950), plus one Wiener quotation — clipped from a local reading app. Its subject is information and originality, and it explicitly notes that "today's LLMs corroborate" Wiener's point. The Infinite Backlog
NLW's term for the condition agents create: because an agent never tires and only stops when you stop feeding it work, work no longer has a natural end — and the undone remainder feels like a personal failure to delegate. Token Shortage
The condition NLW describes for 2026: latent demand for AI exceeds available supply because of compute constraints — "the total amount of AI that would be consumed, if it could, is higher than the amount of AI that is available." Tree vs Semilattice
Two ways a whole can be made of parts. In a tree, any two units are nested or disjoint. In a semilattice, units overlap, and each overlap is itself a unit. Living things are semilattices, but the mind can only hold a tree at once (Alexander — A City Is Not a Tree). Venkatesh Rao
Writer and consultant (Ribbonfarm 2007–2024; Contraptions newsletter), known for idiosyncratic conceptual frameworks, often built as 2×2 matrices with sharply named quadrants. Verification Requires Domain Knowledge Verification Requires Domain Knowledge · auto-ingested
You can only catch an AI's mistakes in areas you already know. So the work you delegate because you lack the expertise is exactly the work whose errors you are least able to see — and those errors cost the most (Anthropic — Investigating Vault Particles with Claude). Vibe Coding
Building software (and increasingly, whole projects) by describing intent in natural language and letting AI generate and iterate — "speaking things into existence." Mentioned in both ingested sources. What's Scarce When Execution Is Cheap
Filed 2026-06-02. Three sources in this wiki — written in three different decades and disciplines — independently describe the same structure: when the cost of producing the default, conventional output collapses, value migrates to the scarce, high-information, high-judgment contribution that (for now) only humans reliably supply. Willingham — Critical Thinking, Why Is It So Hard to Teach
Cognitive psychologist Daniel T. Willingham's review of decades of research on why "critical thinking" resists being taught as a general skill: thinking is intertwined with domain knowledge. Arts Education Policy Review 109(4), 21–32 (2008), reprinted from American Educator (AFT), Summer 2007. Human-added to raw/ 2026-11-08. Opens the wiki's thinking & learning domain — and is the first source to bear directly on the wiki's own method (Method Reflection). Working Heuristics — Data, Matching, Product, Org, Tooling Working Heuristics — Data, Matching, Product, Org, Tooling
A list of ~30 hard-won rules of thumb the human distilled from their own data-engineering / product / org work, grouped into five themes. Supplied inline on 2026-06-26 and persisted to raw/. This opens the wiki's engineering-craft domain — distinct from the AI-and-work macro themes that dominate the rest of the wiki, though several heuristics rhyme with them (see _Connections_). Wurman — Five Hat Racks and LATCH
A secondary source. An InfoVis:Wiki (TU Wien) entry quoting Richard Saul Wurman's Information Anxiety (1989) and Information Architects (1996). Wurman's books aren't freely available online, so the primary text stays a #to-read lead. Agent-fetched 2026-11-08 to ground the front door. The AI Daily Brief — Why Agents Still Need Humans · Every · Agent Autonomy Spectrum 31 pages · tightly interlinked · conductance 0.24 Method Reflection · Legibility in the Lens, Not the Land · Tree vs Semilattice 29 pages · tightly interlinked · conductance 0.22 Working Heuristics — Data, Matching, Product, Org, Tooling · Data Contract · Pipeline Legibility 23 pages · tightly interlinked · conductance 0.25 Verification Requires Domain Knowledge · Method Reflection · AI Commoditizes Expertise 21 pages · moderately interlinked · conductance 0.39 Set aside as catalogue pages that link to everything: Index , Overview , Inbox .
Communities: Louvain over link counts, seeded, single-threaded; a region needs at least 3 pages. Cohesion from conductance: ≤0.35 tight, ≥0.60 loose.