Founder Notes
This optional formatting bolds the leading part of each word to give your eye a focus point; some readers find it helps them stay locked in.
For most of recorded history, depth was scarce and breadth was cheap. A person could glance at many subjects through common sense and ordinary observation, but going deep on any one subject meant rare access: a library a day's ride away, an apprenticeship under a master, a decade in a laboratory. The scarce thing was the depth itself. The valuable person was the one who had walked the ten thousand hours.
AI has inverted that constraint. Depth is now the cheap resource. Any reasonable model in 2026 can carry a user to graduate-level treatment of almost any recognizable field in minutes, with live citations, cross-referencing, and worked examples. The specialist's archive is now a dial, not a mountain. What used to be a career's worth of accumulation is a well-formed prompt.
The part that has not been automated is knowing WHAT to ask about. Knowing that a subject called operations research exists. Knowing that industrial designers and mechanical engineers solve overlapping but different problems. Knowing which domain to route a question to before the question is even formed. The scarce cognitive asset in the AI era is the map. The depth any map points at is now a commodity.
This is a structural inversion, not a mood. It does not mean depth is worthless. It means the bottleneck on good decisions has moved. The old bottleneck was finding and absorbing depth. The new bottleneck is routing to the right depth at the right time.
A well-indexed mind can traverse many domains quickly because it knows where to look. A poorly-indexed mind stays in the domain it already knows because the alternatives are invisible to it. Before AI, the second mind could compensate through deep expertise in its one domain and do reasonably well. After AI, the first mind scales, and the second mind stalls.
The librarian has always been a strange kind of expert. A research librarian does not necessarily know the content of every book on every shelf. The librarian knows the call number system, the classification logic, the cross-references, and the shape of the collection. That knowledge is structural. It is orthogonal to subject-matter depth. And it is precisely the skill that turns a million disconnected facts into a navigable system.
In an AI-saturated environment, the librarian mind generalizes. The person who can say "this is a scheduling problem, which is operations research, which has well-developed methods going back to Dantzig's simplex work, which a modern model will recognize immediately" routes the question in seconds. The specialist who only knows, say, marketing, tries to solve a scheduling problem with marketing tools. Both have access to the same AI. Only one gets the right answer out.
Routing beats answering. The answer is always one query away. The routing is a lifetime habit.
Traditional academic taxonomies organize knowledge by subject matter. Biology studies living things. Economics studies markets. Civil engineering studies static structures. This made sense when the cost of moving between subjects was high, because the subjects were genuinely isolated. You had to pick one.
When depth is a commodity and routing is the skill, subject-matter taxonomy becomes the wrong organizing principle. A better principle is function. What does this domain do, at the level of the verb? Organized by verb, knowledge clusters differently, and the clusters are more useful for routing.
A working functional taxonomy needs roughly twelve meta-functions: diagnose, predict, construct, transport, allocate, persuade, protect, regulate, cultivate, record, transform, coordinate. Every professional domain maps to a primary function and one or two secondary functions. A physician primarily diagnoses. A civil engineer primarily constructs. A logistician primarily transports. A financial analyst primarily allocates. A lawyer primarily regulates and persuades. A farmer primarily cultivates. A historian primarily records. A chef primarily transforms. A project manager primarily coordinates.
The consequence of a functional taxonomy is that domains performing the same function become partially interchangeable as reference points. A diagnostic technique in one field often suggests a diagnostic technique in another. The physician's differential diagnosis, the mechanic's fault tree, the software engineer's bisect search, the auditor's anomaly detection are structurally the same procedure running on different data. Once you see the function, you can borrow across the clusters.
This is not new. Biologists borrowed error-correcting codes from information theory. Economists borrowed equilibrium concepts from physics. Historians borrow statistical methods from demography. What is new is that a mind equipped with a functional index can do this borrowing as a default operating mode rather than as a research program.
Creativity, under this framing, is an idea molded by the knowledge of the holder while constrained by the reality of the time. The holder's knowledge is no longer bounded by the depth of their own study. It is bounded by the breadth of their index, because an indexed mind can query the depth it needs. The constraint of the time now includes AI as a live operand. So creativity shifts from what a person has mastered to what a person can point at.
Education optimized for depth is a thousand-year-old institution. It was designed for an environment where the scarce asset was depth. Most of its architecture, from the lecture hall to the semester to the terminal degree, exists to pack specific content into a specific mind over a specific duration.
Education optimized for breadth looks different. It teaches the shape of the map rather than the contents of any one region. It spends a week on the twelve meta-functions. It spends a semester on the relationship between academic disciplines, industries, and roles. It gives students practice routing real problems to the right cluster before it teaches them any cluster in depth. Meta-literacy becomes the curriculum.
The practical implication is not that specialist training disappears. Surgeons still need ten thousand hours. Pilots still need simulator time. Nuclear engineers still need their reactor physics. The implication is that the generalist routing layer, currently taught accidentally or not at all, becomes the ground floor of the educational stack. Depth gets built on top, when needed, for specific applications.
The generalist router beats the specialist answerer for any task whose frame is not obvious. Most real work has non-obvious frames. "What should this company do next" is not a marketing question or a product question or a finance question. It is a routing question. A generalist who can say "this is a distribution problem first, then a pricing problem, then a capital allocation problem, in that order" produces more value than any single specialist working alone.
In an AI-assisted team, the valuable unit becomes a generalist paired with AI as their on-call specialist. The generalist sets the frame. The AI supplies depth on demand. The output is a route plus a depth-on-demand answer, ten times faster than either alone. Specialists still exist for the deep work that AI cannot perform reliably, but their role shrinks toward the hard tails.
If routing is the skill, a personal knowledge system should be built around the index, not the content. The mental index is the asset. Notes, vaults, and personal wikis have traditionally optimized for storage: more capture, more cross-links, more coverage. A system built for the AI era optimizes for structure: the shape of the entries, the verbs they use, the way they connect domains to industries to roles.
The shape of a working personal system: twelve meta-function nodes, about a hundred domain nodes, about eighty academic-discipline nodes, about eighty industry nodes, about two hundred role nodes. Every node carries a four-section body: functions, duties, deliverables, tasks. Every node cross-links across the other four types. The graph view of such a system is not a collection of disconnected topics. It is a routing table.
The personal brain then does two things. It holds the routing table. And it sends queries through AI when depth is needed. The routing table is trained and maintained by the person. The depth is rented by the query.
The system brain is not a storage vault. It is a routing table.
The person most valuable in the AI era is the one who knows where every kind of question lives, and can route questions to the right cluster before forming the answer. That person does not need to know everything. They need to know the shape of everything. And they need a working index in front of them so the routing happens in seconds rather than hours.
Depth was the scarce asset for a thousand years. Breadth is the scarce asset now. The map is the new expertise.