Shareable

    Twenty posters of the framework.

    The stack, the laws, the archetypes, the dynamics, every poster downloads as a watermarked PNG or PDF. Use them in decks, on LinkedIn, on your wall. Citation-ready, no signup.

    Back to the framework

    I, The Stack

    Three ways to see all ten layers at once.

    00

    The Framework · One Image

    At-a-Glance Summary

    The single image to answer 'what is this framework?', 10 layers with a one-liner each, plus the 4 laws strip. Built for screenshot + paste.

    The Framework · At a Glance

    Supply Chain of Intelligence™

    The 10 layers of the generative AI stack, not logistics, not freight.

    L8MemoryRetention, learning, compounding context. What the system remembers.
    L7SurfaceInterface, presentation, experience. How the user meets the intelligence.
    L6OrchestrationWorkflow, routing, coordination. How skills compose into outcomes.
    L5ExecutionApplied skills and capabilities. Doing the actual work.
    L4AccessConnectivity, permissions, integrations, the pipes layer.
    L3GatesTrust, acceptance, approval. Can the system be allowed in?
    L2ModelsIntelligence refinement. Rent early, build custom at scale.
    L1DataThe raw input. What data do you have that nobody else can get?
    L0InfraThe shovels. Chips, data centers, networking, cloud, edge, what is needed to process intelligence.
    L−1ResourcesWhat supports the chain. Energy, water, fabs, materials, skilled trades, the inputs the entire stack consumes.

    The Four Structural Laws

    I

    Intelligence commoditizes downward.

    II

    Value accrues at bottlenecks.

    III

    Surface captures attention; chain captures power.

    IV

    Generation and verification must be separate.

    10 layers · 50 sublayers · 4 laws. The map for every AI strategy conversation.

    SupplyChainOfAI.com

    01

    Magazine Cover

    The Hero Brand Poster

    Dark, editorial, scroll-stopping. The single image to anchor a LinkedIn post or open a deck.

    A Framework for AI Defensibility

    The Supply Chain
    of Intelligence.

    10 layers · 50 sublayers · 4 structural laws

    L-1
    L0
    L1
    L2
    L3
    L4
    L5
    L6
    L7
    L8

    below the line

    above the line, where intelligence compounds

    By Anand Arivukkarasu · supplychainofai.com

    02

    Reference Sheet

    The 10 × 50 Grid

    Every layer and every sublayer on one page. The pin-on-the-wall version. ★ marks structurally defensible sublayers.

    Supply Chain of Intelligence™ (SCoI)

    The 10 layers · 50 sublayers of the generative AI stack.

    L8
    Memory
    L8a
    Session & Short-Term Memory
    L8b
    User & Entity Profiles
    L8c
    Aggregated Network Learning
    L8d
    Institutional Knowledge
    L8e
    Learned World Models
    L7
    Surface
    L7a
    Conversational
    L7b
    Visual Interfaces & Media
    L7c
    Embedded & Embodied AI
    L7d
    Transaction Surface
    L7e
    Async & Ambient Surfaces
    L6
    Orchestration
    L6a
    Agent Loops
    L6b
    Human-in-the-Loop
    L6c
    Role Routing & Task Decomposition
    L6d
    Context & State Management
    L6e
    Runtime Assurance & Learning Loops
    L5
    Execution
    L5a
    Domain Execution & Tool Use
    L5b
    Decision Frameworks & Reasoning Scaffolds
    L5c
    Retrieval-Augmented Workflows
    L5d
    Operating Playbooks
    L5e
    Interaction Skills & Actuation
    L4
    Access
    L4a
    API & Integration Layer
    L4b
    Agent Interface Protocols
    L4c
    Access Governance & Agent Commerce
    L4d
    Real-Time Interaction Infrastructure
    L4e
    Agent Identity & Provenance
    L3
    Gates
    Gatekeeping
    L3a
    Compliance & Export Controls
    L3b
    Quality Gates
    L3c
    Safety, Security & Provenance
    L3d
    Editorial Gates
    L3e
    Distribution Gates
    L2
    Models
    L2a
    Foundation & Multimodal Models
    L2b
    Specialized & Fine-Tuned Models
    L2c
    Embedding & Retrieval
    L2d
    Model Routing & Composition
    L2e
    Reasoning & World Models
    L1
    Data
    L1a
    Public & Open Data
    L1b
    Proprietary Data
    L1c
    Behavioral & Sensor Data
    L1d
    Outcome Data
    L1e
    Synthetic & Simulation Data
    L0
    Infra
    Infrastructure
    L0a
    Silicon & Memory
    L0b
    Data Centers
    L0c
    Interconnect Fabric
    L0d
    Compute & State Infrastructure
    L0e
    Edge & On-Device Compute
    L-1
    Resources
    L-1a
    Energy & Grid Interconnect
    L-1b
    Thermal & Water Management
    L-1c
    Fabrication & Foundry
    L-1d
    Critical Materials & Supply Chain
    L-1e
    Skilled Trades & Human Capital

    ★ = structurally defensible sublayer. Below L1 = foundation. Above = where intelligence compounds.

    Anand Arivukkarasu · SupplyChainOfAI.com

    03

    One Square

    The Stack, on a Square

    Sized for LinkedIn, X, decks, Pinterest. Ten chips, the tagline, the attribution, nothing else.

    Supply Chain of Intelligence™ (SCoI)

    The 10 layers of the generative AI stack.

    L8Memory
    L7Surface
    L6Orchestration
    L5Execution
    L4Access
    L3Gatekeeping
    L2Models
    L1Data
    L0Infrastructure
    L-1Resources

    Not logistics. The generative AI stack.

    Anand Arivukkarasu

    04

    Mental Model

    Above / Below the Line

    The single most teachable diagram in the framework. Below the line: inputs consumed. Above: intelligence compounded.

    The Mental Model

    Above the line, intelligence compounds.Below the line, the inputs get consumed.

    ▲ Above, value compounds

    data · trust · distribution · workflow · context · surface · memory

    L1
    Data
    L2
    Models
    L3
    Gates
    L4
    Access
    L5
    Execution
    L6
    Orchestration
    L7
    Surface
    L8
    Memory
    The Line

    ▼ Below, inputs the chain consumes

    power · water · fabs · chips · data centers

    L-1
    Resources
    L0
    Infrastructure

    Below the line is consumed. Above the line is accumulated. Build above. Hedge below.

    Anand Arivukkarasu · SupplyChainOfAI.com

    II, The Arguments

    The posters that win the disagreement.

    05

    Why It Matters

    10 Benefits of the Framework

    The answer to 'why should I care?' One artifact, ten precise reasons. Best for product leaders, founders, and skeptical investors.

    Why The Framework Exists

    10 things you cannot say without Supply Chain of Intelligence™.

    01

    A Common Language

    Replaces vague words, 'agentic', 'wrapper', 'AI-native', with layer-level precision.

    02

    Predicts Who Gets Absorbed

    Law I: surface-only products become features inside the model layer below them.

    03

    Names the Bottleneck

    Law II: durable value sits at the scarce layer, data, trust, distribution, memory.

    04

    Decodes 'Agent'

    An agent is not a layer. It is L5 + L7 (+L8) packaging. Decode it; never quote it.

    05

    Separates Surface From Chain

    Beautiful UIs capture attention. Deep chains capture power. The framework draws the line.

    06

    Shows Where Intelligence Compounds

    L1 → L8 is gravity-fed downward, value-fed upward. Memory (L8) is the strongest moat.

    07

    Maps Trust to a Layer

    L3 is not a feature, it is a structural gate that the model layer cannot legally cross.

    08

    Forces an Honest Moat Conversation

    Founders, investors, PMs ask: 'which layers do we actually own?' No more hand-waving.

    09

    Replaces Hype With Diagnosis

    Every announcement maps to a layer. Every threat maps to a law. Every category maps to an archetype.

    10

    Shifts From Generation To Accumulation

    Most AI thinking optimizes for output. The framework optimizes for what persists.

    Vague words let weak strategy hide. The framework forces it into the light.

    Anand Arivukkarasu · SupplyChainOfAI.com

    06

    The Physics

    The Four Structural Laws

    Wrappers get compressed. Bottlenecks compound. Surface captures attention; chain captures power. Memory is the final moat.

    The Physics of the Stack

    The Four Structural Laws.

    ILaw I

    Intelligence Commoditizes Downward

    Wrappers don't survive. Wrappers become features.

    L7-only → absorbed by L2

    IILaw II

    Value Accrues at Bottlenecks

    Find the scarce layer. Own it. Everything else is rent.

    L1b · L3 · L8

    IIILaw III

    Surface Captures Attention; Chain Captures Power

    Beautiful UIs get users. Deep chains keep them.

    L4 + L5 + L6 + L8

    IVLaw IV

    Memory Is the Final Moat

    What the system remembers about the user, no one else can rebuild.

    L8 compounds

    Laws describe what the market will do, not what you wish it would do.

    Anand Arivukkarasu · SupplyChainOfAI.com

    07

    The Moat Shape

    The Defensibility Triangle

    L1b × L3 × L8. Own two corners and you're defensible. Own three and you're uncopyable. Everything else is rent.

    Where Moats Live

    The Defensibility Triangle.

    Own two corners = defensible. Own three = uncopyable.

    uncopyableALL THREEL1bPROPRIETARY DATAL3TRUST GATESL8MEMORYDATA × MEMORYDATA × TRUSTTRUST × MEMORY

    The model layer can replicate the surface. It cannot replicate your data, your trust, or your memory.

    SupplyChainOfAI.com

    08

    Buzzword, Decoded

    The Agent Decoder

    'Agent' is not a layer. It's L5 + L7 (+L8) packaging. Use this whenever someone announces an agent, and ask which three layers.

    Buzzword, Decoded

    "Agent" is not a layer.

    Agent=
    L5Orchestration
    L7Surface
    (+
    L8Memory
    )

    Without L5

    It's a chatbot.

    Without L7

    It's a script.

    Without L8

    It's a demo.

    When someone says "we built an agent," ask which three layers.

    SupplyChainOfAI.com

    III, The Models

    The frameworks-within-the-framework.

    09

    Pattern Recognition

    The 6 Archetypes

    Every AI company collapses into one of six shapes: Fortress · Refinery · Railroad · Memory · Surface · Agent. Find yours.

    Pattern Recognition

    The 6 Archetypes of AI companies.

    Every AI company collapses into one of six shapes. Find yours.

    Fortress

    L1L3L8

    Multi-layer ownership. Built to outlast the model layer.

    e.g. Bloomberg · Epic · Tempus

    Refinery

    L1

    Owns the data well. Sells refined intelligence upward.

    e.g. Apollo · Scale · Getty

    Railroad

    L0L4

    Owns the rails. Every workload pays the toll.

    e.g. NVIDIA · AWS · Stripe

    Memory

    L8

    Compounds context per user. Switching cost becomes painful.

    e.g. Notion AI · Granola · Linear

    Surface

    L7

    Beautiful, exposed, structurally absorbable by L2.

    e.g. Jasper · Copy.ai · Gamma

    Agent

    L5L7

    Packaged as 'an agent.' Decode it: L5 + L7 (+L8) only works with L8.

    e.g. Sierra · Lindy · Cognition

    Archetype is not a vibe. It is the layer-set you own.

    Anand Arivukkarasu · SupplyChainOfAI.com

    10

    Three Dimensions

    The Intelligence Cube™

    Layer × Sublayer × Depth. Defensibility is volume, not feature count.

    The Intelligence Cube™

    Defensibility has three dimensions.

    L8L7L6L5L4L3L2L1← 5 ACROSS · SUBLAYERS →← 8 LAYERS TALL →DEPTH →

    Layer × Sublayer × Depth. Volume = total defensibility.

    SupplyChainOfAI.com

    11

    For Non-Technical Readers

    The Gold Mining Analogy

    The whole stack told as one extended metaphor, from the land itself to the polished jewelry. The poster to hand a board member.

    For Non-Technical Readers

    Supply Chain of Intelligence, told as a gold mine.

    L8
    Memory

    The Record Book, Compounding Knowledge

    The jeweler keeps records: which designs sold, which metals each customer prefers.

    L7
    Surface

    Wearing the Jewelry, The Moment of Experience

    People see the ring on the finger, the surface, the sparkle, the emotional moment.

    L6
    Orchestration

    The Jewelry Store & Workshop

    A single ring is useful.

    L5
    Execution

    The Master Jeweler

    A jeweler takes refined gold and crafts rings, necklaces, watches, each requiring specialized skill.

    L4
    Access

    The Railroads & Transport

    Refined gold needs to move, by rail, armored truck, secure vault.

    L3
    Gates

    The Hallmark & Assay Office

    Before gold enters the market, the assay office verifies purity and the hallmark guarantees quality.

    L2
    Models

    The Smelter & Refinery

    Raw ore becomes pure gold through smelting.

    L1
    Data

    The Raw Gold Ore

    The unrefined material pulled from the earth.

    L0
    Infra

    The Shovels & Mining Equipment

    Before anyone finds gold, someone has to build the pickaxes, drill rigs, and mine shafts.

    L-1
    Resources

    The Ground Itself, Land, Power, Materials

    Before the gold rush, you need land, water rights, ore deposits, and the miners who work the seams.

    Every gold rush enriches the shovel sellers, the assayers, and the refiners, long after the miners are gone.

    Anand Arivukkarasu · SupplyChainOfAI.com

    12

    Self-Diagnostic

    Where Does Your Company Live?

    A fill-in-the-blank stack chart. Print it, tick the layers you actually own, and answer three honest questions.

    Print · Pin · Answer

    Where does your company actually live?

    Mark the layers you own. Not what you touch. Not what you integrate. What you own.

    L8Memory

    Retention, learning, compounding context. What the system remembers.

    L7Surface

    Interface, presentation, experience. How the user meets the intelligence.

    L6Orchestration

    Workflow, routing, coordination. How skills compose into outcomes.

    L5Execution

    Applied skills and capabilities. Doing the actual work.

    L4Access

    Connectivity, permissions, integrations, the pipes layer.

    L3Gates

    Trust, acceptance, approval. Can the system be allowed in?

    L2Models

    Intelligence refinement. Rent early, build custom at scale.

    L1Data

    The raw input. What data do you have that nobody else can get?

    L0Infra

    The shovels. Chips, data centers, networking, cloud, edge, what is needed to process intelligence.

    L-1Resources

    What supports the chain. Energy, water, fabs, materials, skilled trades, the inputs the entire stack consumes.

    Q1

    Which layers do you own?

    Tick boxes →

    Q2

    Which is your bottleneck?

    The scarce one, Law II.

    Q3

    What happens if L2 ships it free?

    If nothing, you're inside Law I.

    If the boxes you tick are only L7, the model layer below you owns your roadmap.

    Anand Arivukkarasu · SupplyChainOfAI.com

    IV, The Dynamics

    Where the framework becomes an operating system.

    13

    Where The Meter Runs

    The AI Tax Map

    Seven taxes every AI product pays, token, context, orchestration, inference margin, API dependency, model lock-in, distribution. Most founders only price for two.

    Where The Meter Runs

    The AI Tax Map.

    Every AI product pays seven taxes. Most founders only price for two.

    L0

    Inference Margin Collapse

    Tax #01

    Every call = compute cost. Margin shrinks with usage, not scale.

    L2

    The Token Tax

    Tax #02

    $/token × every user × forever. The meter never stops running.

    L2

    Model Lock-In Tax

    Tax #03

    Switch providers = rewrite prompts, re-eval, re-validate. Real cost.

    L4

    API Dependency Tax

    Tax #04

    Rate limits, deprecations, ToS shifts. Your roadmap = their decisions.

    L5

    Context Tax

    Tax #05

    Bigger windows = bigger bills. Context cost compounds per turn.

    L5

    Orchestration Tax

    Tax #06

    Each tool-call = a model hop. Latency × cost × failure surface.

    L7

    Distribution Tax

    Tax #07

    App stores, ad platforms, search, they price you. You don't price them.

    If you can't name your tax stack, your margin isn't real, it's just unbilled.

    Anand Arivukkarasu · SupplyChainOfAI.com

    14

    Law I, Visualized

    The Stack Compression Map

    When the model layer ships your feature for free, who holds and where the counter-move sits. The visual proof of Law I.

    Law I, Visualized

    The Stack Compression Map.

    When the model layer ships your feature for free, who holds, and where the counter-move sits.

    ◌ Compressed

    Surface-only
    L7
    Jasper
    Prompt + UI. GPT-3.5 wrapper.
    L7
    Chegg
    Generic Q&A. -99% market cap.
    L7
    Stack Overflow
    Community absorbed into models.
    L7
    Generic Copilots
    Same prompt, free in ChatGPT.
    L7
    Prompt Wrappers
    Feature, not company.
    L5
    Thin Orchestration
    L5 without L1/L8 = a demo.

    ★ Survives

    Bottleneck owners
    L1bL3
    Bloomberg
    40yr data well + regulated trust.
    L1b
    Apollo.io
    Proprietary B2B contact graph.
    L1L4
    Epic
    Owns the hospital workflow.
    L3
    Vanta
    Compliance gate every B2B crosses.
    L1L8
    Tempus
    Clinical data + per-patient memory.
    L8
    Notion AI
    Knows your workspace. Migration = pain.

    The model layer absorbs anything that doesn't sit underneath it.

    Anand Arivukkarasu · SupplyChainOfAI.com

    15

    Four Strategic Arcs

    The Migration Paths

    Companies that survive Law I migrate. Four archetypal arcs, Wrapper to Workflow, Surface to Memory, Tool to System, Data to Platform, cover almost every move worth making.

    Four Strategic Arcs

    The Migration Paths.

    Companies that survive Law I migrate. Four arcs cover almost every move worth making.

    ARC-01

    Wrapper → Workflow

    e.g. Harvey · Cresta
    L7
    L5
    L6
    or die.

    Stop renting features from L2. Embed inside the work.

    ARC-02

    Surface → Memory

    e.g. Notion AI · Granola
    L7
    L8
    or die.

    Trade attention for accumulation. Make leaving painful.

    ARC-03

    Tool → System of Record

    e.g. Linear · Rippling
    L7
    L4
    L5
    or die.

    Become the place the workflow runs through, not on.

    ARC-04

    Data → Platform

    e.g. Apollo · Bloomberg
    L1
    L4
    L2
    or die.

    Open the API. Let others build the surface on your bottleneck.

    Standing still on L7 is the only move that always loses.

    Anand Arivukkarasu · SupplyChainOfAI.com

    16

    How The Stack Moves

    The Gravity Flow

    Data flows down. Value flows up. Memory compounds recursively. The single rule that explains where margins go.

    How The Stack Moves

    The Gravity Flow.

    Data ↓
    L8Memory
    L7Surface
    L6Orchestration
    L5Execution
    L4Access
    L3Gates
    L2Models
    L1Data
    L0Infra
    L-1Resources
    Value ↑
    DATA

    flows down, training, fine-tuning, context fill.

    VALUE

    flows up, margin accrues at orchestration & surface.

    MEMORY

    compounds, every interaction makes leaving more painful.

    Build where things accumulate. Not where they pass through.

    SupplyChainOfAI.com

    17

    The Moat Quadrant

    Open vs Closed Surfaces

    Plot a surface by openness, gate density, and memory depth. The top-left is where AI cannot legally, technically, or commercially compete.

    The Moat Quadrant

    Open vs Closed Surfaces.

    Plot a surface by openness, gate density, and memory depth. The top-left is where AI cannot easily reach.

    MOAT ZONE← CLOSED · OPENNESS · OPEN →← LOW · GATE DENSITY · HIGH →EmailChatGPTSlackQuickBooksSalesforceBloombergEpic

    Read The Map

    • X, how open the data/API is
    • Y, how many trust gates protect it
    • Size, depth of accumulated memory

    The top-left big bubbles are where AI cannot legally, technically, or commercially compete.

    Open is cheap. Closed is durable. Gated + closed + memory-rich is uncopyable.

    Anand Arivukkarasu · SupplyChainOfAI.com

    18

    What Google Actually Said

    GEO is Just SEO, The Five Myths

    Google's Nov 2025 generative-AI search guide, decoded through the 10-layer stack. The AEO/GEO playbook being sold to founders is mostly noise, here's what actually moves the needle.

    What Google Actually Said

    GEO is just SEO.
    The five myths, debunked.

    Google's Nov 2025 generative-AI search guide, decoded through the 10-layer stack. Most of the "AEO/GEO playbook" being sold to founders is noise.

    Write llms.txt files for AI

    Not a Google signal. No special treatment.

    L1

    Ship unique, expert content at L1b.

    Chunk your content for AI parsing

    No ideal length. Systems read context.

    L7

    Write for humans. L7 is downstream of L1.

    Rewrite copy in 'AI-friendly' phrasing

    Models understand synonyms and intent.

    L1

    First-hand POV beats keyword fitting.

    Farm inauthentic mentions across the web

    Spam systems catch it. RAG ignores it.

    L3

    Earn citations through L3 trust gates.

    Pile on structured-data schema everywhere

    Not required for generative search.

    L2

    Technical hygiene + indexability. That's it.

    The Bottom Line

    Google's AI search is grounded in core ranking. The moat is still L1b unique data × L3 trust gates × L8 memory. Everything else is rented surface.

    Source: Google Search Central · "Optimizing your website for generative AI features" (2025).

    Anand Arivukkarasu · SupplyChainOfAI.com

    V, For Your Audience

    Role-specific posters to share alongside your own POV.

    19

    For Product Leaders

    The Product Leader's Lens

    Five questions to ask of any AI feature on your roadmap, mapped to the layer each one stress-tests. The poster that ends 'just a wrapper' arguments in a standup.

    For Product Leaders

    Five questions to ask of any AI feature on your roadmap.

    Before the demo. Before the spec. Before the OKR. If your team can't answer four of five, you're shipping a wrapper, not a product.

    01L2 → L5/L7 compression

    If the model layer ships this for free next quarter, what's left?

    If the answer is 'nothing', you're building inside Law I.

    02L1b · L1c · L1d

    Where does the data come from that no competitor can get?

    No proprietary L1 → no defensible learning loop.

    03L8c · L8d · L8e

    What does the system remember after the session ends?

    Stateless features churn. Memory compounds retention.

    04L3 over L2/L5

    Who verifies the output when it's wrong?

    Generator and verifier must be separate. That's Law IV.

    05L7 alone ≠ defensible

    Is the surface the moat, or just the doorway?

    Beautiful UI gets users. Deep chain keeps them.

    "A roadmap is a layer claim. Most AI roadmaps are claiming L7."

    Supply Chain of Intelligence™, the 10 layers of the generative AI stack.

    SupplyChainOfAI.com

    20

    For Investors

    The Investor's Lens, Rent or Own

    A 10-layer diligence map. Which layers compound, which commoditize, and the five questions every AI deal memo should answer before the check clears.

    For Investors

    Rent or own? A 10-layer diligence map.

    Every AI cap-table claim collapses to one question, which layers does this company structurally own, and which is it merely renting from the layer below?

    Layer
    Structural Verdict
    R / O
    Why
    L8 · Memory
    Compounds
    OWN
    The final moat
    L7 · Surface
    Commoditizes
    RENT
    Doorway, not moat
    L6 · Orchestration
    Becomes a feature
    RENT
    Absorbed by L2
    L5 · Execution
    Durable if deep
    OWN
    Domain > generic
    L4 · Access
    Load-bearing
    OWN
    Pipes the agents ride
    L3 · Gates
    Structurally permanent
    OWN
    Law IV protects it
    L2 · Models
    Winner-take-most
    RENT
    Commodity risk
    L1 · Data
    Defensible if proprietary
    OWN
    L1b/c/d compound
    L0 · Infrastructure
    Shovel sellers win
    OWN
    NVIDIA, fabs, DCs
    L-1 · Resources
    Slow, scarce, real
    OWN
    Power & water

    Five questions for the next deal memo

    1. 1.Which two layers does this company actually own?
    2. 2.What does the L2 roadmap make free in 18 months?
    3. 3.Where does proprietary data come from, structurally?
    4. 4.What compounds the longer customers stay?
    5. 5.If a 'gate' exists above it, who owns the gate?

    Supply Chain of Intelligence™, the 10 layers of the generative AI stack.

    SupplyChainOfAI.com

    More posters as the framework grows. Suggest one →

    Back to the framework