@prefix : <https://saranormous.substack.com/p/the-untrainable#> .
@prefix schema: <http://schema.org/> .
@prefix rdf: <http://www.w3.org/1999/02/22-rdf-syntax-ns#> .
@prefix rdfs: <http://www.w3.org/2000/01/rdf-schema#> .
@prefix owl: <http://www.w3.org/2002/07/owl#> .
@prefix xsd: <http://www.w3.org/2001/XMLSchema#> .
@prefix skos: <http://www.w3.org/2004/02/skos/core#> .
@prefix prov: <http://www.w3.org/ns/prov#> .
@prefix foaf: <http://xmlns.com/foaf/0.1/> .
@prefix dbr: <http://dbpedia.org/resource/> .
@prefix wde: <http://www.wikidata.org/entity/> .

<> a schema:CreativeWork ;
    schema:name "The Untrainable Thesis Analysis"@en ;
    schema:description "RDF-Turtle knowledge graph representing the investment thesis from Sarah Guo's Substack article 'The Untrainable', which argues that sustainable AI value lies in work whose correctness is private and expensive to establish, not in publicly benchmarkable tasks."@en ;
    schema:dateCreated "2026-06-10"^^xsd:date ;
    schema:dateModified "2026-06-10T12:00:00Z"^^xsd:dateTime ;
    schema:author <https://x.com/saranormous#this> ;
    schema:about :thesisAnalysis .

# ── Lightweight Ontology ──────────────────────────────────────────────────────

:thesisOntology a owl:Ontology ;
    schema:name "AI Value Thesis Ontology"@en ;
    schema:description "Lightweight ontology encoding Sarah Guo's thesis framework for AI value: work categorized by correctness type (public vs private) and saturation level."@en ;
    rdfs:label "AI Value Thesis Ontology"@en ;
    rdfs:comment "Classes and properties for modeling knowledge work domains by their amenability to AI-driven commoditization."@en ;
    schema:identifier "https://saranormous.substack.com/p/the-untrainable" ;
    rdfs:isDefinedBy :thesisOntology .

:WorkDomain a rdfs:Class ;
    rdfs:label "Work Domain"@en ;
    rdfs:comment "A category of knowledge work distinguished by how its correctness is established and verified."@en ;
    rdfs:isDefinedBy :thesisOntology .

:PublicCorrectnessWork a rdfs:Class ;
    rdfs:label "Public Correctness Work"@en ;
    rdfs:comment "Work whose correctness can be verified by anyone using publicly available benchmarks, test suites, or evaluation criteria. Highly susceptible to AI automation and commoditization."@en ;
    rdfs:subClassOf :WorkDomain ;
    rdfs:isDefinedBy :thesisOntology .

:PrivateCorrectnessWork a rdfs:Class ;
    rdfs:label "Private Correctness Work"@en ;
    rdfs:comment "Work whose correctness can only be established inside a specific firm, system, or relationship using private data and context. Resistant to commoditization by general AI models."@en ;
    rdfs:subClassOf :WorkDomain ;
    rdfs:isDefinedBy :thesisOntology .

:MoatType a rdfs:Class ;
    rdfs:label "Moat Type"@en ;
    rdfs:comment "A category of competitive advantage that protects against AI-driven absorption."@en ;
    rdfs:isDefinedBy :thesisOntology .

:EnvironmentMoat a rdfs:Class ;
    rdfs:label "Environment Moat"@en ;
    rdfs:comment "Access barrier created by security review, integration, contract obligations, and liability frameworks. A model must be let in the door."@en ;
    rdfs:subClassOf :MoatType ;
    rdfs:isDefinedBy :thesisOntology .

:UserMoat a rdfs:Class ;
    rdfs:label "User Moat"@en ;
    rdfs:comment "Access barrier created by user habit, trust, and clinical or professional adoption patterns. Built slowly through relationships, not gradient descent."@en ;
    rdfs:subClassOf :MoatType ;
    rdfs:isDefinedBy :thesisOntology .

:hasCorrectnessType a rdfs:Property ;
    rdfs:label "has correctness type"@en ;
    rdfs:comment "Indicates whether correctness in this domain is established publicly or privately."@en ;
    rdfs:domain :WorkDomain ;
    rdfs:range xsd:string ;
    rdfs:isDefinedBy :thesisOntology .

:hasSaturationLevel a rdfs:Property ;
    rdfs:label "has saturation level"@en ;
    rdfs:comment "Indicates how saturated (commoditized by AI) this work domain is."@en ;
    rdfs:domain :WorkDomain ;
    rdfs:range xsd:string ;
    rdfs:isDefinedBy :thesisOntology .

:hasTAM a rdfs:Property ;
    rdfs:label "has total addressable market"@en ;
    rdfs:comment "The estimated total addressable market for AI automation in this domain."@en ;
    rdfs:domain :WorkDomain ;
    rdfs:range xsd:string ;
    rdfs:isDefinedBy :thesisOntology .

:hasExample a rdfs:Property ;
    rdfs:label "has example"@en ;
    rdfs:comment "An example of work or a product operating in this domain."@en ;
    rdfs:domain :WorkDomain ;
    rdfs:range rdfs:Resource ;
    rdfs:isDefinedBy :thesisOntology .

# ── Work Domain Instances ─────────────────────────────────────────────────────

:softwareEngineeringDomain a :PublicCorrectnessWork ;
    schema:name "Software Engineering (Coding Tasks)"@en ;
    :hasCorrectnessType "Public"@en ;
    :hasSaturationLevel "High"@en ;
    :hasTAM "not quantified"@en ;
    schema:description "Coding tasks with public benchmarks (SWE-Bench Verified) and free verifiers (compilers, test suites). Agents hit high 80s on SWE-Bench but writing cheap code is not the same as shipping."@en ;
    schema:identifier "https://www.census.gov/naics/?input=541511&year=2022&details=541511" ;
    schema:naics "541511" ;
    rdfs:seeAlso dbr:Massachusetts_Institute_of_Technology ;
    :hasExample :devinAgent .

:legalServicesDomain a :PrivateCorrectnessWork ;
    schema:name "Legal Services (M&A Practice)"@en ;
    :hasCorrectnessType "Private"@en ;
    :hasSaturationLevel "Medium"@en ;
    schema:description "White-shoe law firm M&A practice running ~1000 deals/year. Cannot use general agents on client files due to confidentiality, and deal-level signal (NDA, term sheet, diligence, purchase agreement, closing checklist) is practice-area specific."@en ;
    schema:identifier "https://www.census.gov/naics/?input=541110&year=2022&details=541110" ;
    schema:naics "541110" .

:healthcareDomain a :PrivateCorrectnessWork ;
    schema:name "Clinical Medicine"@en ;
    :hasCorrectnessType "Private"@en ;
    :hasSaturationLevel "Low"@en ;
    schema:description "Medical practice where correctness is defined per-firm and per-patient. A majority of American doctors open OpenEvidence daily. Trust is built through relationships with user acquiescence."@en ;
    schema:identifier "https://www.census.gov/naics/?input=622110&year=2022&details=622110" ;
    schema:naics "622110" .

:bankingDomain a :PrivateCorrectnessWork ;
    schema:name "Banking and Finance"@en ;
    :hasCorrectnessType "Private"@en ;
    :hasSaturationLevel "Medium"@en ;
    schema:description "Production banking systems are neither portable nor standardized. You do not get root by being clever on SWE-Bench."@en ;
    :hasExample :goldmanSachs .

# ── Entities: Products and Tools ─────────────────────────────────────────────

:devinAgent a schema:SoftwareApplication ;
    schema:name "Devin"@en ;
    schema:description "AI software engineering agent by Cognition. Shipped 2024 solving 13% of SWE-Bench tasks; by mid-2026 hits high 80s. Used inside Goldman Sachs and the U.S. Army."@en ;
    schema:url "https://www.cognition.ai" ;
    schema:applicationCategory "DeveloperApplication"@en ;
    schema:offers :devinPerformanceGuarantee .

:devinPerformanceGuarantee a schema:Offer ;
    schema:name "Devin Performance Guarantee"@en ;
    schema:description "Cognition's outcome-based pricing for Devin in software: performance guarantee that is only possible for outcomes in a system the provider is trusted inside."@en .

:sierraAgent a schema:SoftwareApplication ;
    schema:name "Sierra AI Agent"@en ;
    schema:description "Customer service AI agent that charges only when it resolves a customer issue, pricing the outcome instead of access."@en ;
    schema:url "https://sierra.ai" ;
    schema:applicationCategory "BusinessApplication"@en .

:openEvidence a schema:SoftwareApplication ;
    schema:name "OpenEvidence"@en ;
    schema:description "AI clinical decision support platform used by a majority of American doctors daily. Settling what a safe clinical answer looks like."@en ;
    schema:url "https://www.openevidence.com" ;
    schema:applicationCategory "HealthApplication"@en .

:chatgpt a schema:SoftwareApplication ;
    schema:name "ChatGPT"@en ;
    schema:description "Consumer AI chat application by OpenAI. Held its lead through years of competition; losing share to Gemini on the strength of Android and Search distribution."@en ;
    schema:url "https://chatgpt.com" ;
    schema:applicationCategory "WebApplication"@en .

:gemini a schema:SoftwareApplication ;
    schema:name "Gemini"@en ;
    schema:description "Google's AI chat application. Gaining market share from ChatGPT through Android and Search distribution advantages, not better model quality."@en ;
    schema:url "https://gemini.google.com" ;
    schema:applicationCategory "WebApplication"@en .

:baseten a schema:SoftwareApplication ;
    schema:name "Baseten"@en ;
    schema:description "Inference infrastructure provider. Best AI-native companies concentrate serving on one or two providers because reliability under real traffic and guaranteed compute access do not commoditize."@en ;
    schema:url "https://www.baseten.co" ;
    schema:applicationCategory "DeveloperApplication"@en .

:fireworks a schema:SoftwareApplication ;
    schema:name "Fireworks AI"@en ;
    schema:description "Inference infrastructure and model serving provider. Cost per token commoditizes on schedule but reliability and scarce compute access do not."@en ;
    schema:url "https://fireworks.ai" ;
    schema:applicationCategory "DeveloperApplication"@en .

# ── Entities: Organizations ──────────────────────────────────────────────────

:anthropic a schema:Organization ;
    schema:name "Anthropic"@en ;
    schema:description "AI safety and research company. Prediction markets rate Anthropic as having the best model, yet it is barely a factor in consumer chat and built its business in enterprise and coding instead."@en ;
    rdfs:seeAlso dbr:Anthropic .

:nvidia a schema:Organization ;
    schema:name "Nvidia"@en ;
    schema:description "GPU and AI compute hardware company. One of two safe havens in the AI despair narrative alongside Anthropic."@en ;
    rdfs:seeAlso dbr:Nvidia .

:openai a schema:Organization ;
    schema:name "OpenAI"@en ;
    schema:description "AI research and deployment organization behind ChatGPT, GPT models, and reasoning research (Noam Brown)."@en ;
    rdfs:seeAlso dbr:OpenAI .

:cognition a schema:Organization ;
    schema:name "Cognition"@en ;
    schema:description "AI company that builds Devin, the AI software engineering agent. Pioneered outcome-based performance guarantees."@en ;
    schema:url "https://www.cognition.ai" .

:goldmanSachs a schema:Organization ;
    schema:name "Goldman Sachs"@en ;
    schema:description "Global investment bank using Devin for real software engineering work. Example of an organization with high permission barriers."@en ;
    rdfs:seeAlso dbr:Goldman_Sachs .

:usArmy a schema:GovernmentOrganization ;
    schema:name "United States Army"@en ;
    schema:description "U.S. military branch using Devin for real work. High-security environment where private correctness dominates."@en ;
    rdfs:seeAlso dbr:United_States_Army .

:google a schema:Organization ;
    schema:name "Google"@en ;
    schema:description "Technology company that develops Gemini and Android. Consumer AI distribution advantage through Android and Search."@en ;
    rdfs:seeAlso dbr:Google .

:rippling a schema:Organization ;
    schema:name "Rippling"@en ;
    schema:description "Unified workforce management platform. Matt MacInnis, who articulated the commodity token value distinction, works here."@en ;
    schema:url "https://www.rippling.com" .

:ucsf a schema:MedicalOrganization ;
    schema:name "University of California, San Francisco"@en ;
    schema:description "Medical center where OpenEvidence is used in clinical decision flow. Example of a user-moat environment."@en ;
    rdfs:seeAlso <http://dbpedia.org/resource/University_of_California,_San_Francisco> .

:mit a schema:EducationalOrganization ;
    schema:name "Massachusetts Institute of Technology"@en ;
    schema:description "University where Mert Demirer and coauthors researched the impact of coding agents on developer productivity (180% more code written, 30% more shipped)."@en ;
    rdfs:seeAlso dbr:Massachusetts_Institute_of_Technology .

# ── Entities: People ─────────────────────────────────────────────────────────

:sarahGuo a schema:Person ;
    schema:name "Sarah Guo"@en ;
    schema:description "Venture investor and author of 'The Untrainable' thesis on AI value. Founder of Conviction Capital."@en ;
    schema:url "https://x.com/saranormous" ;
    schema:identifier "Sarah Guo" ;
    owl:sameAs <https://x.com/saranormous#this> ,
               <http://www.wikidata.org/entity/Q123553538> .

:mertDemirer a schema:Person ;
    schema:name "Mert Demirer"@en ;
    schema:description "MIT researcher and coauthor of study on coding agent impact across 100,000+ developers: 180% more code written, ~30% more shipped."@en .

:noamBrown a schema:Person ;
    schema:name "Noam Brown"@en ;
    schema:description "OpenAI reasoning researcher who pioneered reasoning models. Wrote that evaluating an agent over a one-year horizon may require running it for a year."@en .

:gabePereyra a schema:Person ;
    schema:name "Gabe Pereyra"@en ;
    schema:description "Product voice quoted on real automation requiring product, model, workflow, and firm to move together — three of which move at organizational speed."@en .

:mattMacInnis a schema:Person ;
    schema:name "Matt MacInnis"@en ;
    schema:description "Rippling executive who articulated the commodity token framework: a token answering a generic question is worth almost nothing; a token reasoning over company data is worth much more."@en .

:benSaltiel a schema:Person ;
    schema:name "Ben Saltiel"@en ;
    schema:description "Commenter on Sarah Guo's post who observed there is no alpha trying to compete against larger general models with the same inputs."@en .

# ── Entities: Books ──────────────────────────────────────────────────────────

:innovatorsDilemma a schema:Book ;
    schema:name "The Innovator's Dilemma"@en ;
    schema:description "Clayton Christensen's classic on how incumbents fail to adopt disruptive technologies. Applied by Sarah Guo to the AI absorption frontier."@en ;
    schema:isbn "9780060521998" ;
    schema:author "Clayton M. Christensen"@en .

# ── Entities: Core Thesis Concepts ────────────────────────────────────────────

:untrainableConcept a skos:Concept ;
    skos:prefLabel "The Untrainable"@en ;
    skos:definition "Value that cannot be captured by AI benchmarks because its correctness is private, grounded in history, and established inside organizations through relationships and trust."@en .

:absorptionFrontier a skos:Concept ;
    skos:prefLabel "Absorption Frontier"@en ;
    skos:definition "The boundary of measurable work that is being pulled into model weights. The retrieval, routing, tool use, and reasoning policy that used to wrap a model is being absorbed into the weights themselves."@en .

:legibleWorkConcept a skos:Concept ;
    skos:prefLabel "Legible Work"@en ;
    skos:definition "Work that can be measured, benchmarked, and trained against. Anything that can be put on a leaderboard is already on its way to commodity. Being eaten from two directions: task saturation from below, model weight absorption from above."@en .

:illegibleWorkConcept a skos:Concept ;
    skos:prefLabel "Illegible Work"@en ;
    skos:definition "Work whose correctness is private and expensive to establish. The valuable work that remains after legible work is automated. Integration, maintenance, domain translation, and the slow work of organizational change."@en .

:privateCorrectness a skos:Concept ;
    skos:prefLabel "Private Correctness"@en ;
    skos:definition "Truth that exists only inside someone's data. The kind of correctness that cannot be read off a leaderboard — you find out whether a system that complex works by running it in the world long enough to learn."@en .

:permissionMoat a skos:Concept ;
    skos:prefLabel "Permission Moat"@en ;
    skos:definition "The double barrier protecting private-correctness domains: the environment (security review, integration, contract, liability) and the user (habit, trust, acquiescence). Intelligence is not the bottleneck here; permission and accountability are."@en .

:outcomeBasedPricing a skos:Concept ;
    skos:prefLabel "Outcome-Based Pricing"@en ;
    skos:definition "Pricing model where the price is the evaluation: Sierra charges when its agent resolves an issue and nothing when it escalates to a human. Only possible inside a system the provider is trusted within."@en .

:thinWrapper a skos:Concept ;
    skos:prefLabel "Thin Wrapper"@en ;
    skos:definition "A shallow application layer over a foundation model that is easily absorbed by the model as frontier models incorporate scaffolding into weights. Most of what looks like a company today is a thin wrapper."@en .

:commodityTokens a skos:Concept ;
    skos:prefLabel "Commodity Tokens"@en ;
    skos:definition "Tokens spent answering generic questions that any model can answer. Worth almost nothing. Contrasts with tokens reasoning over company-specific data, which do valuable work."@en .

# ── Main Analysis Document ────────────────────────────────────────────────────

:thesisAnalysis a schema:CreativeWork ;
    schema:name "The Untrainable: Thesis Analysis"@en ;
    schema:description "Comprehensive thesis analysis of Sarah Guo's 'The Untrainable' — the argument that sustainable AI value lies in the illegible, unmeasurable work whose correctness is private, not in publicly benchmarkable tasks being commoditized by frontier models."@en ;
    schema:about :untrainableConcept, :absorptionFrontier, :legibleWorkConcept, :illegibleWorkConcept, :privateCorrectness, :permissionMoat, :outcomeBasedPricing, :thinWrapper, :commodityTokens ;
    schema:dateCreated "2026-06-10"^^xsd:date ;
    schema:author :sarahGuo ;
    schema:hasPart :faqSection, :glossarySection, :howtoSection, :softwareEngineeringDomain, :legalServicesDomain, :healthcareDomain, :bankingDomain, :thesisOntology ;
    schema:isPartOf :thesisAnalysis ;
    prov:wasGeneratedBy :kgGeneratorSkill .

:kgGeneratorSkill a schema:SoftwareApplication ;
    schema:name "kg-generator skill"@en ;
    schema:url <https://github.com/OpenLinkSoftware/ai-agent-skills/tree/main/kg-generator> ;
    schema:description "Knowledge Graph Generator skill that transforms documents into RDF-Turtle using curated prompt templates and schema.org vocabulary."@en .

:rdfInfographicSkill a schema:SoftwareApplication ;
    schema:name "rdf-infographic-skill"@en ;
    schema:url <https://github.com/OpenLinkSoftware/ai-agent-skills/tree/main/rdf-infographic-skill> ;
    schema:description "RDF Infographic skill that generates interactive HTML infographics and Markdown companions from RDF knowledge graph data."@en .

:originalArticle a schema:Article ;
    schema:name "The Untrainable"@en ;
    schema:description "Sarah Guo's mid-2026 investor thesis on AI value, arguing that defensible companies operate in domains where correctness is private and expensive to establish."@en ;
    schema:url "https://saranormous.substack.com/p/the-untrainable" ;
    schema:datePublished "2026-06-10"^^xsd:date ;
    schema:author :sarahGuo ;
    schema:publisher :substack ;
    schema:image <https://substackcdn.com/image/fetch/$s_!T0QG!,w_1456,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F2df31ad2-63a4-4048-b154-ffd2b179fb16_2172x724.png> .

:substack a schema:Organization ;
    schema:name "Substack"@en ;
    schema:url "https://substack.com" .

# ── Market Disruption Action ──────────────────────────────────────────────────

:marketDisruptionAction a schema:Action ;
    schema:name "AI-driven market disruption"@en ;
    schema:description "The ongoing disruption of services markets by AI models and agents that sell outcomes rather than tools, starting with outsourced intelligence-heavy tasks."@en .

:aiAutopilotDisruption a schema:Product ;
    schema:name "AI Autopilot Disruption"@en ;
    schema:description "The thesis that AI 'autopilots' are disrupting services markets by selling outcomes (resolved tickets, completed code, processed legal documents) rather than tools or access."@en .

# ── Thread Replies / Comments ─────────────────────────────────────────────────

:threadReply1 a schema:Comment ;
    schema:name "Ben Saltiel's comment"@en ;
    schema:text "Great post Sarah. It goes to show that there is no alpha trying to compete against the larger general models with the same inputs."@en ;
    schema:author :benSaltiel ;
    schema:dateCreated "2026-06-10T14:38:00Z"^^xsd:dateTime .

# ── FAQ Section (12 questions) ────────────────────────────────────────────────

:faqSection a schema:FAQPage ;
    schema:name "Frequently Asked Questions: The Untrainable Thesis"@en ;
    schema:about :thesisAnalysis ;
    schema:isPartOf :thesisAnalysis ;
    schema:mainEntity :q1, :q2, :q3, :q4, :q5, :q6, :q7, :q8, :q9, :q10, :q11, :q12 .

:q1 a schema:Question ;
    schema:name "What is 'the untrainable' in Sarah Guo's thesis?"@en ;
    schema:text "What is 'the untrainable' in Sarah Guo's thesis?"@en ;
    schema:acceptedAnswer :a1 ;
    schema:isPartOf :faqSection .

:a1 a schema:Answer ;
    schema:name "Definition of the untrainable"@en ;
    schema:text "The untrainable is value with history: work whose correctness cannot be measured by any public benchmark because it exists only inside a specific firm's data, relationships, and operational reality. A better model does not make private ground truth public and cannot hold the license, sign off on liability, own the firm's files, or be the party sued when the answer is wrong."@en ;
    schema:isPartOf :faqSection .

:q2 a schema:Question ;
    schema:name "What is the absorption frontier?"@en ;
    schema:text "What is the absorption frontier?"@en ;
    schema:acceptedAnswer :a2 ;
    schema:isPartOf :faqSection .

:a2 a schema:Answer ;
    schema:name "Definition of absorption frontier"@en ;
    schema:text "The absorption frontier is the boundary where measurable work gets pulled into model weights. The retrieval, routing between cheap and expensive calls, tool use, and even reasoning policy that used to wrap a model are being absorbed into the weights, until the wrapper is the model. The frontier keeps rising as we learn to measure more of the work."@en ;
    schema:isPartOf :faqSection .

:q3 a schema:Question ;
    schema:name "How does legible work differ from illegible work?"@en ;
    schema:text "How does legible work differ from illegible work?"@en ;
    schema:acceptedAnswer :a3 ;
    schema:isPartOf :faqSection .

:a3 a schema:Answer ;
    schema:name "Legible vs illegible work"@en ;
    schema:text "Legible work can be measured, benchmarked, and trained against: anything on a leaderboard is already on its way to commodity. Illegible work has private correctness that is expensive to establish: integration that runs as long as the relationship does, the slow rebuild of an engineering organization, the judgment of what good means in a specific field. The valuable work is illegible by construction."@en ;
    schema:isPartOf :faqSection .

:q4 a schema:Question ;
    schema:name "What does 'private correctness' mean?"@en ;
    schema:text "What does 'private correctness' mean?"@en ;
    schema:acceptedAnswer :a4 ;
    schema:isPartOf :faqSection .

:a4 a schema:Answer ;
    schema:name "Private correctness defined"@en ;
    schema:text "Private correctness means truth that exists only inside a specific firm's data, context, and operations. You cannot read it off a leaderboard. You find out whether a complex system works by running it in the world long enough to learn. A smarter model does not make the world run faster and the clock cannot be skipped: Noam Brown noted that evaluating an agent over a one-year horizon may require running it for a year."@en ;
    schema:isPartOf :faqSection .

:q5 a schema:Question ;
    schema:name "What are commodity tokens?"@en ;
    schema:text "What are commodity tokens?"@en ;
    schema:acceptedAnswer :a5 ;
    schema:isPartOf :faqSection .

:a5 a schema:Answer ;
    schema:name "Commodity tokens defined"@en ;
    schema:text "Commodity tokens are tokens spent answering generic questions that any model can answer. They are worth almost nothing because the buyer can substitute any provider's model. By contrast, tokens reasoning over a company's specific data are worth much more because they do the thing the company actually wants, not just the plausible thing."@en ;
    schema:isPartOf :faqSection .

:q6 a schema:Question ;
    schema:name "What is the thin wrapper problem?"@en ;
    schema:text "What is the thin wrapper problem?"@en ;
    schema:acceptedAnswer :a6 ;
    schema:isPartOf :faqSection .

:a6 a schema:Answer ;
    schema:name "Thin wrapper problem defined"@en ;
    schema:text "The thin wrapper problem is that most companies built on top of a foundation model are shallow application layers waiting to be absorbed as the model incorporates the scaffolding into its own weights. The labs are trying to get models to swallow their own retrieval, routing, and tool use. A lot of what looks like a company today is a thin wrapper."@en ;
    schema:isPartOf :faqSection .

:q7 a schema:Question ;
    schema:name "How does outcome-based pricing work and why is it defensible?"@en ;
    schema:text "How does outcome-based pricing work and why is it defensible?"@en ;
    schema:acceptedAnswer :a7 ;
    schema:isPartOf :faqSection .

:a7 a schema:Answer ;
    schema:name "Outcome-based pricing"@en ;
    schema:text "Outcome-based pricing means the price is the evaluation: Sierra charges only when its agent resolves a customer issue and nothing when it escalates to a human. Devin offers a performance guarantee in software. Both are only possible inside a system the provider is trusted within. The company that owns the definition of 'resolved' or 'good work' controls the standard everyone else is measured against."@en ;
    schema:isPartOf :faqSection .

:q8 a schema:Question ;
    schema:name "What is the permission moat?"@en ;
    schema:text "What is the permission moat?"@en ;
    schema:acceptedAnswer :a8 ;
    schema:isPartOf :faqSection .

:a8 a schema:Answer ;
    schema:name "Permission moat explained"@en ;
    schema:text "The permission moat has two barriers: the lock (environment: security review, integration, contract, liability) and the deadbolt (user: habit, trust, acquiescence). A model far smarter than any person still has to be let in the door and someone still has to put their name on what it does. Intelligence is not the bottleneck; permission and accountability are."@en ;
    schema:isPartOf :faqSection .

:q9 a schema:Question ;
    schema:name "Why can't better models alone win markets?"@en ;
    schema:text "Why can't better models alone win markets?"@en ;
    schema:acceptedAnswer :a9 ;
    schema:isPartOf :faqSection .

:a9 a schema:Answer ;
    schema:name "Better models cannot alone win markets"@en ;
    schema:text "In consumer chat, the best model has never simply won. ChatGPT held its lead through years of competition and is losing share to Gemini on the strength of Android and Search, not better models. Anthropic, rated as having the best model, is barely a factor in consumer chat. A better model cannot integrate through a hospital's records or a bank's liability. The public chooses on more than coding today."@en ;
    schema:isPartOf :faqSection .

:q10 a schema:Question ;
    schema:name "How does the Innovator's Dilemma apply to AI companies?"@en ;
    schema:text "How does the Innovator's Dilemma apply to AI companies?"@en ;
    schema:acceptedAnswer :a10 ;
    schema:isPartOf :faqSection .

:a10 a schema:Answer ;
    schema:name "Innovator's Dilemma in AI"@en ;
    schema:text "Incumbents keep the ground they have, and the next thing comes from someone who finds a use before the rest. Competing on the general model is a capital war you lose to whoever owns the most compute. The trap is committing to out-train the frontier in a general swath of tasks, where the winner is decided by datacenter size."@en ;
    schema:isPartOf :faqSection .

:q11 a schema:Question ;
    schema:name "What makes integration and maintenance defensible?"@en ;
    schema:text "What makes integration and maintenance defensible?"@en ;
    schema:acceptedAnswer :a11 ;
    schema:isPartOf :faqSection .

:a11 a schema:Answer ;
    schema:name "Defensibility of integration and maintenance"@en ;
    schema:text "Integration and maintenance run as long as the relationship does, won by teams that put domain-specialized engineers and tools next to the customer. A company that brings the translation between private reality and model action is tough to copy because the translation never ends. It takes an operator to moneyball a firm, with ambiguous intermediate goals and incomplete feedback over very long horizons in an environment that will not hold still."@en ;
    schema:isPartOf :faqSection .

:q12 a schema:Question ;
    schema:name "What is the role of judgment in defining 'good' work?"@en ;
    schema:text "What is the role of judgment in defining 'good' work?"@en ;
    schema:acceptedAnswer :a12 ;
    schema:isPartOf :faqSection .

:a12 a schema:Answer ;
    schema:name "Judgment defines good work"@en ;
    schema:text "If work cannot be scored from outside, someone on the inside must decide what a good answer even is. That decision is the whole game. Enough of those decisions written down become a benchmark. Harvey publishes one for law, Sierra for voice agents. The senior lawyer writes the legal benchmark. Defining a safe clinical answer falls to a physician. That authority lands where it already sat."@en ;
    schema:isPartOf :faqSection .

# ── Glossary Section (10 terms) ──────────────────────────────────────────────

:glossarySection a schema:DefinedTermSet, skos:ConceptScheme ;
    schema:name "Glossary of Key Terms: The Untrainable Thesis"@en ;
    schema:description "Key concepts and terms defined in Sarah Guo's investment thesis on the untrainable."@en ;
    schema:about :thesisAnalysis ;
    schema:isPartOf :thesisAnalysis ;
    schema:hasDefinedTerm :termUntrainable, :termAbsorptionFrontier, :termLegibleWork, :termIllegibleWork, :termPrivateCorrectness, :termPermissionMoat, :termOutcomePricing, :termThinWrapper, :termCommodityToken, :termVerticalMapping .

:termUntrainable a schema:DefinedTerm, skos:Concept ;
    schema:name "The Untrainable"@en ;
    schema:description "Value that cannot be captured by AI benchmarks because its correctness is private, grounded in history, and established through relationships inside organizations."@en ;
    schema:isPartOf :glossarySection .

:termAbsorptionFrontier a schema:DefinedTerm, skos:Concept ;
    schema:name "Absorption Frontier"@en ;
    schema:description "The boundary where measurable work gets pulled into model weights as labs incorporate retrieval, routing, tool use, and reasoning policy directly into the model."@en ;
    schema:isPartOf :glossarySection .

:termLegibleWork a schema:DefinedTerm, skos:Concept ;
    schema:name "Legible Work"@en ;
    schema:description "Work that can be measured, benchmarked, and trained against. Being eaten from two directions: task saturation from below and model absorption from above."@en ;
    schema:isPartOf :glossarySection .

:termIllegibleWork a schema:DefinedTerm, skos:Concept ;
    schema:name "Illegible Work"@en ;
    schema:description "Work whose correctness is private and expensive to establish. The valuable work that survives after legible tasks are commoditized."@en ;
    schema:isPartOf :glossarySection .

:termPrivateCorrectness a schema:DefinedTerm, skos:Concept ;
    schema:name "Private Correctness"@en ;
    schema:description "Truth that exists only inside a specific firm's data and operational context. Cannot be verified by public benchmarks; requires time and trust to establish."@en ;
    schema:isPartOf :glossarySection .

:termPermissionMoat a schema:DefinedTerm, skos:Concept ;
    schema:name "Permission Moat"@en ;
    schema:description "The double barrier of environment (security, integration, liability) and user (habit, trust, acquiescence) that protects private-correctness domains from AI absorption."@en ;
    schema:isPartOf :glossarySection .

:termOutcomePricing a schema:DefinedTerm, skos:Concept ;
    schema:name "Outcome-Based Pricing"@en ;
    schema:description "A pricing model where the fee is the evaluation: charge only when the AI agent produces a verified good outcome. Only possible inside a trusted system."@en ;
    schema:isPartOf :glossarySection .

:termThinWrapper a schema:DefinedTerm, skos:Concept ;
    schema:name "Thin Wrapper"@en ;
    schema:description "A shallow application over a foundation model that is easily absorbed as the model incorporates the wrapper's functionality into its own weights."@en ;
    schema:isPartOf :glossarySection .

:termCommodityToken a schema:DefinedTerm, skos:Concept ;
    schema:name "Commodity Token"@en ;
    schema:description "A token spent answering a generic question that any model can answer. Worth near zero because the buyer can substitute any provider."@en ;
    schema:isPartOf :glossarySection .

:termVerticalMapping a schema:DefinedTerm, skos:Concept ;
    schema:name "Vertical Mapping"@en ;
    schema:description "The strategic process of mapping knowledge work domains by correctness type (public vs private) and saturation level to identify where defensible AI value can be built."@en ;
    schema:isPartOf :glossarySection .

# ── HowTo Section (7 steps) ──────────────────────────────────────────────────

:howtoSection a schema:HowTo ;
    schema:name "How to Build Defensible AI Value in the Untrainable"@en ;
    schema:description "Strategic framework derived from Sarah Guo's thesis for building AI companies that resist commoditization by the absorption frontier."@en ;
    schema:about :thesisAnalysis ;
    schema:isPartOf :thesisAnalysis ;
    schema:step :step1, :step2, :step3, :step4, :step5, :step6, :step7 .

:step1 a schema:HowToStep ;
    schema:name "Identify work whose correctness is private and expensive to establish"@en ;
    schema:position 1 ;
    schema:text "Ask two questions of any work domain: (1) Is its correctness private and expensive to establish, existing only inside someone's data? (2) Is it walled off inside a system you cannot get into? If yes to both, the work is in the untrainable corner where defensible value lives. If correctness is public and the task is saturated, you are competing on commodity tokens."@en ;
    schema:isPartOf :howtoSection .

:step2 a schema:HowToStep ;
    schema:name "Get inside a domain with high permission barriers"@en ;
    schema:position 2 ;
    schema:text "The door has a lock and a deadbolt. The lock is the environment: security review, integration, contract, liability. The deadbolt is the user: habit, trust, acquiescence. You cannot verify whether AI is useful until you are trusted inside. Trust is built slowly on relationships, not gradient descent. Intelligence is not the bottleneck; permission and accountability are."@en ;
    schema:isPartOf :howtoSection .

:step3 a schema:HowToStep ;
    schema:name "Build the translation layer between private reality and model action"@en ;
    schema:position 3 ;
    schema:text "An application earns its place in the untrainable corner by doing unglamorous work: arranging a company's private reality so a model can act on it, handing the model the tools to act, and working with the customer to change the reality of its workforce. The translation never ends. Integration and maintenance run as long as the relationship does."@en ;
    schema:isPartOf :howtoSection .

:step4 a schema:HowToStep ;
    schema:name "Price outcomes, not tools or access"@en ;
    schema:position 4 ;
    schema:text "Stop trying to prove value externally. Get in and price the outcome instead. Sierra charges when its agent resolves a customer issue and nothing when it kicks to a human. Devin offers a performance guarantee. When the price is the evaluation, it works only because you own the definition of the good outcome from inside the trusted system."@en ;
    schema:isPartOf :howtoSection .

:step5 a schema:HowToStep ;
    schema:name "Earn the right to define what 'good' means in your field"@en ;
    schema:position 5 ;
    schema:text "If work cannot be scored from outside, someone on the inside must decide what good means. Enough of those decisions, written down, become a benchmark. You earn the right to define good by being the one the field already uses, through the struggle of real adoption. The senior lawyer writes the legal benchmark. The physician defines a safe clinical answer."@en ;
    schema:isPartOf :howtoSection .

:step6 a schema:HowToStep ;
    schema:name "Continuously re-underwrite defensibility as the absorption frontier rises"@en ;
    schema:position 6 ;
    schema:text "The absorption frontier keeps rising because we keep learning to measure more of the work. The untrainable ground shrinks under whoever is standing on it. You cannot find a defensible spot and rest. You keep stepping toward whatever cannot yet be scored and re-underwrite constantly. On a narrow task with private data and your own evals, you can beat the general models."@en ;
    schema:isPartOf :howtoSection .

:step7 a schema:HowToStep ;
    schema:name "Choose what to build through judgment no model can replicate"@en ;
    schema:position 7 ;
    schema:text "Even harder than defense is offense: choosing what to build in the first place. The model will do whatever you point it at and cannot tell you what is worth pointing it at. You cannot benchmark that, so you cannot train it. Intent is an even scarcer input than compute. Incumbents keep the ground they have; the next thing comes from someone who finds a use before the rest."@en ;
    schema:isPartOf :howtoSection .