Supply Chain of Intelligence™(SCoI)
    Market MapLegal

    Vertical Market Map · Legal

    Legal AI, mapped across 10 layers & 50 sublayers.

    AI for law firms, in-house counsel, and contracting.

    As of AI-first companies · incl. early-stage · May 2026·Re-reviewed monthly·methodology
    ThesisValue is bifurcating to L5 Domain Execution at the top (Harvey, Legora) and L1 Data underneath (outcome data + proprietary corpora). At the early stage, L5a is already crowded, differentiation must come from a vertical data moat, not the workflow. The fundable open layers are L3a Compliance & Export Controls, L1d / L8e litigation-outcome data, and the AI-native firm model.
    Filter by stage:
    Solid = primary layer · faded = secondaryOpen whitespaceFeature (absorbed)Horizontal (owned outside vertical)
    · A ·
    · B ·
    · C ·
    · D ·
    · E ·
    L8
    Memory
    L8A
    Session & Short-Term Memory
    Framework / infra feature
    L8B
    User & Entity Profiles
    Absorbed into L5 platforms
    L8C
    Aggregated Network Learning
    ⌁ Network-effect whitespace
    L8D⌁ OPEN
    Institutional Knowledge
    L8E⌁ OPEN
    Learned World Models
    L7
    Surface
    L7A
    Conversational
    Commoditized by ChatGPT et al.
    L7B
    Visual Interfaces & Media
    Horizontal media tools
    L7C
    Embedded & Embodied AI
    L7D
    Transaction Surface
    ⌁ Transaction surface, held by DocuSign/Ironclad; AI-native version unbuilt
    L7E
    Async & Ambient Surfaces
    ⌁ Emerging ambient surface
    L6
    Orchestration
    L6A
    Agent Loops
    L6B
    Human-in-the-Loop
    L6C
    Role Routing & Task Decomposition
    L6D
    Context & State Management
    Built into L5 platforms
    L6E
    Runtime Assurance & Learning Loops
    Built into L5 platforms
    L5
    Execution
    L5A
    Domain Execution & Tool Use
    L5B
    Decision Frameworks & Reasoning Scaffolds
    L5C
    Retrieval-Augmented Workflows
    L5D
    Operating Playbooks
    ⌁ Firm-specific playbooks, under-built
    L5E
    Interaction Skills & Actuation
    Feature inside L5a/L7c products
    L4
    Access
    L4A
    API & Integration Layer
    Horizontal integration infra
    L4B
    Agent Interface Protocols
    Cross-industry protocol (emerging)
    L4C
    Access Governance & Agent Commerce
    Platform feature
    L4D
    Real-Time Interaction Infrastructure
    Horizontal infra
    L4E
    Agent Identity & Provenance
    ⌁ Agent identity / provenance, open
    L3
    Gates
    L3A⌁ OPEN
    Compliance & Export Controls
    L3B
    Quality Gates
    L3C
    Safety, Security & Provenance
    L3D
    Editorial Gates
    Platform feature
    L3E
    Distribution Gates
    Platform feature
    L2
    Models
    L2A
    Foundation & Multimodal Models
    Foundation labs (OpenAI / Anthropic / Google)
    L2B
    Specialized & Fine-Tuned Models
    L2C
    Embedding & Retrieval
    Horizontal infra
    L2D
    Model Routing & Composition
    Horizontal / infra
    L2E
    Reasoning & World Models
    Foundation labs
    L1
    Data
    L1A
    Public & Open Data
    L1B
    Proprietary Data
    L1C
    Behavioral & Sensor Data
    ⌁ Legal behavioral data, under-exploited
    L1D⌁ OPEN
    Outcome Data
    L1E
    Synthetic & Simulation Data
    ⌁ Synthetic legal data, emerging
    L0
    Infra
    Vertical-agnostic, shared GPUs / silicon / data centers / cloud. Not legal-specific.
    L-1
    Resources
    Vertical-agnostic, energy, materials, fabrication. Not legal-specific.

    Layer scorecard, contested vs. open

    SublayerAI-first occupantsState
    L1b Proprietary DatavLex/Vincent (acq. Clio), Regbase, BlueshoeScarce → bought
    L1d Outcome DataEvenUp, Supio, Eve, Darrow, Theo AiOpen & fundable
    L2b Specialized & Fine-Tuned ModelsLuminance, PaxtonSqueezed middle
    L3a Compliance & Export ControlsNorm Ai, Delve, Greenlite, Hadrius, RegbaseOpen & fundable
    L5a Domain Execution & Tool UseHarvey, Legora, EvenUp, Supio, Eve, Luminance, Ivo, Wordsmith, Definely, LegalFly, Lawhive, DeepIP, Patlytics, Solve, immigration trioMost crowded
    L5c RAG / ResearchPaxton, Alexi, Vincent, Midpage, Wexler, NLPatentMost crowded
    L7c Embedded CopilotSpellbook, Genie, Definely, DeepIP, Pincites†, Robin†Squeezed middle
    L8d Institutional KnowledgeHarvey, Legora, Alexi (aspiring)Open & fundable
    L8e Learned World ModelsTheo Ai (litigation outcomes)Open & fundable

    Where the next winner comes from

    1 · L8d Institutional Knowledge, the unclaimed crown.
    Turn execution exhaust into structured, queryable memory. Everyone aspires; nobody owns it.
    2 · L1d / L8e litigation-outcome data.
    Theo Ai & Darrow are nearly alone in owning hard-to-copy case-outcome data. Outcome data beyond personal injury is wide open.
    3 · L3a Compliance & Export Controls.
    Norm Ai, Delve, Greenlite are early and few, an upstream, recurring, Bloomberg-shaped layer.
    4 · The AI-native firm (Crosby, Garfield, LegalOS).
    Doesn't sell software, replaces the firm. Captures full margin + outcome data. Highest variance.
    † Pincites → acquired by Filevine (Dec 2025); Robin AI → distressed sale to Microsoft (Mar 2026). Funding from primary announcements + TechCrunch, Bloomberg, LawNext/LawSites, Artificial Lawyer, Sifted, Crunchbase. Valuations point-in-time (2025-26).