Supply Chain of Intelligence™(SCoI)
    Full framework
    railroad

    Layer L4

    Access

    Connectivity, permissions, integrations, the pipes layer.

    Why it matters

    Grammarly survived because it had railroad tracks (plugins) into Word, Gmail, Chrome. Jasper had no tracks. Deep integrations and agent identity create switching costs.

    The Railroads & Transport

    Refined gold needs to move, by rail, armored truck, secure vault. In AI: APIs, MCP, real-time pipes, and agent identity move intelligence between systems. Grammarly survived because it had tracks into every workflow. Jasper had none.

    The 5 sublayers

    L4a

    API & Integration Layer

    REST/GraphQL endpoints, SDKs, webhooks connecting AI to systems

    L4b

    Agent Interface Protocols

    MCP, tool-use specs, agent-to-agent communication standards

    L4c

    Access Governance & Agent Commerce

    Who can use what, RBAC, scoping, audit trails, and agent-payment rails (Stripe/Visa/Mastercard agent-pay, spend limits, programmatic checkout, machine-to-machine billing)

    L4d

    Real-Time Interaction Infrastructure

    Streaming, voice pipelines, video, low-latency modality transport

    L4e

    Agent Identity & Provenance

    Verifying which agent acted, credential chains, trust signatures

    , Layer diagnostic card · SCOI v1

    Is a company really at L4?

    The pipes: APIs, MCP, integrations, agent identity, real-time transport, how intelligence reaches systems and users.

    Inclusion tests · include if ALL

    • Owns the integration surface other products must use (auth, RBAC, audit, transport).
    • Sits between two systems that would not otherwise talk.
    • Removing it forces every consumer to rebuild the same plumbing.

    Exclusion tests · exclude if ANY

    • Builds integrations into someone else's distribution surface (then *they* own L4).
    • Pure SDK with no hosted runtime, no identity, no governance, closer to L6.
    • MCP server with no enforced auth or audit, L4 in shape, not in trust.

    The L4 removal test

    Remove L4 and the agent has hands but no arms. It can think, but it cannot reach the system of record.

    Economic work this layer does

    Provides the universal connective tissue so L5/L6/L7 don't each rebuild every integration.

    Canonical examples

    • Stripe

      The canonical payments rail. Every L7 commerce surface rides it.

    • Microsoft 365 / Google Workspace

      Own the workplace surface, every copilot rents distribution from them.

    • Cloudflare

      Network edge + zero-trust + AI gateway. The connective rail for the agent web.

    Anti-examples · look-alikes that fail

    • MCP demo servers

      Protocol example without governance, not a real L4.

    • Zapier-clone startups

      L4+L6 with no distribution. Absorbed by whichever L7 owns the user.

    • iPaaS without auth ownership

      Pipes that don't carry identity = pipes someone else will replace.

    Disagree with a classification?Open the classification table →

    Who's playing here

    AWSSnowflakeSupabaseTwilio

    Verdict: Load-bearing walls. Invest accordingly.

    Case studies touching L4

    Jasper, Grammarly, Copilot in Word: Same Category, Three Structural Fates

    All three help you write. Jasper owned only the surface (L7c) and dissolved when the model went free. Grammarly owned distribution (L4a + L7c embedded copilot) into every browser and editor, until a bigger L4a owner, Microsoft, integrated the model directly into Word, Outlook, and Teams. Same layer. Bigger railroad. The market is repricing layer ownership, not ARR.

    Gamma at $2.1B: The Thin-Layer Graveyard in Real Time

    Presentation generation lives at L7b, a single thin slice of the stack. Claude, Copilot, and Gemini now do it for free inside surfaces 100× larger than Gamma's. The Intelligence Cube predicted this before the market priced it in: when your entire product is one prompt away from being free inside an L4 you don't own, the valuation is a liability, not a moat.

    Sierra's Memory Moat: Why L8 Beats Salesforce's Agentforce

    Sierra and Salesforce Agentforce look like the same product on stage, an AI agent that resolves customer issues. The Cube projection shows they are structurally opposite. Sierra was architected as L1c behavioral data + L5d operating playbooks + L8c network learning from day one: every resolution compounds into per-customer memory. Agentforce is L5 bolted onto Salesforce's existing L1, with no compounding loop. Same demo, opposite trajectories.

    BloombergGPT: Why a 50B-Parameter Model Beats GPT-4 in Finance

    Bloomberg trained its own 50B-parameter model on 40 years of proprietary financial data. Smaller than GPT-4. Better at finance tasks. The reason isn't the model, it's that Bloomberg owns the terminal (L4a API + L7c embedded), the data (L1b proprietary), the compliance posture (L3a regulated), and now the model (L2b specialized). Four sublayers in one regulated vertical.

    L3 GatekeepingL5 Execution