Knowledge Hub

Agent Commerce Stack.

The 9-layer map of agentic payment infrastructure — from the moment an agent decides to spend, to the moment money settles. Where the gaps are, and who is filling them.

Founder goal

To become the go-to map for the agentic payments stack — where builders, payment companies, and infrastructure teams can understand who is building what, where the protocol gaps are, and what safe agent transactions require.

AI agents are becoming economic actors. I’m mapping the payment infrastructure they’ll need to transact safely.

Layer 1 · Intent Layer 2 · Identity Layer 3 · Authorization Layer 4 · Credentials Layer 5 · Funding Layer 6 · Checkout Layer 7 · Rails Layer 8 · Settlement Layer 9 · Risk

The 9-Layer Framework

Layer	What it does	Key Capabilities	Example Players	Agent Value	Gaps / Opportunities
1 Intent / DecisionWhere the spend decision is made → deep dive	Agent determines whether, what, and how much to spend based on context, preferences, and policy constraints.	LLM reasoning, task planning, budget modeling, preference inference	OpenAI Agents Anthropic CrewAI LangChain	Decides whether, what, and how much to spend	❗ No standard for agent spend policies — LLM budget logic is opaque, non-auditable, and varies by model
2 Identity & TrustWho is acting — human, agent, or bot → deep dive	Verify agent and user identity, detect automated abuse, and establish trust before payment access is granted.	KYC/KYB, bot detection, device fingerprinting, agent attestation, W3C DID	Skyfire KYAPay Forter Persona Socure	Establishes trusted identity before payment access is granted	❗ No universal "proof of agent" standard — existing KYC frameworks were written for natural persons, not software principals
3 AuthorizationWhether this payment is permitted → deep dive	Policy engine that decides whether a specific payment is permitted — within what limits, under whose delegation, and with what approval chain.	Spend policy rules, delegated authorization, approval flows, 3DS/SCA adaptation	Visa TAP MC Agent Pay AP2 ACP	Controls agent spend scope and prevents unauthorized payments	❗ Platform war in progress — Visa, Mastercard, and crypto protocols competing to own delegation standards; 3DS was built for humans, not agents
4 Payment CredentialsThe instrument the agent uses	Issue and store the payment instrument the agent actually presents — card token, wallet key, or signed credential.	Tokenization, PCI-scoped vaulting, virtual card issuance, wallet signing	Nekuda Basis Theory Stripe Issuing Privy Prava	Gives the agent a spendable payment instrument	Most solutions are human-UX virtual card products adapted for agents; no native agent credential spec exists
5 Funding SourceWhere the money comes from	Connect the agent's spending credential to a real funding source — bank account, card line, or stablecoin treasury.	Bank connectivity, card funding, stablecoin on-ramp, multi-currency treasury	Stripe Plaid Circle Coinbase	Links agent credentials to real money	❗ Multi-source routing unsolved — agent needs to select between fiat, crypto, and credits in real time; no unified abstraction layer
6 Checkout / ExecutionCompleting the purchase	The agent completes the purchase — navigating checkout APIs, cart flows, or headless commerce interfaces.	Checkout APIs, cart automation, headless commerce, purchase orchestration	Stripe Agent Toolkit Shopify APIs Browser Use Rye	Translates agent intent into a completed merchant transaction	No agent-native checkout API standard — most flows still require browser automation or custom merchant integrations.
7 Payment RailsHow money moves → deep dive	The actual movement of money — card networks, bank rails, real-time payments, or on-chain transfers.	Card network clearing, ACH/RTP/FedNow, stablecoin settlement, on-chain transfer	Visa / MC FedNow UCP Solana Pay	Moves value from agent wallet to merchant	Rail fragmentation remains severe — card networks, real-time bank rails, and on-chain systems are still incompatible, with no cross-rail routing layer.
8 Settlement & ReconciliationMoney lands, books close	Money lands and books close. Automated reconciliation, ledgering, and audit trails for agent-initiated spend.	Ledgering, automated reconciliation, treasury reporting, real-time audit trails	Stripe Treasury Modern Treasury Adyen	Closes the financial loop — records spend, triggers reporting	❗ T+1 card settlement was designed for human-frequency transactions; agent-initiated micro-transactions at high frequency need different clearing windows and reporting hooks
9 Risk / ComplianceWho bears the risk	Monitor for fraud, enforce AML compliance, manage chargebacks, and assign liability when an agent-initiated payment goes wrong.	AML/BSA monitoring, fraud scoring, chargeback management, liability allocation	Sardine Alloy Unit21	Prevents losses, ensures compliance, determines liability	❗ Visa/MC operating regulations designate the cardholder as a natural person — no dispute or chargeback framework exists for autonomous agent transactions

Deep Dives

Detailed breakdowns of the layers with the biggest open questions.

Layer 1

Intent / Decision

How agents decide to spend: budgeting, preference modeling, policy constraints, and the missing standard for portable spend logic.

LLM Planning Budgets Autonomy

Explore →

Layer 2

Identity & Trust

Who is the agent? Proof-of-agent, KYC/KYB adaptation, and the trust rails that payment systems will require.

Skyfire KYAPay KYC

Explore →

Layer 3

Authorization

Visa TAP, Mastercard Agent Pay, AP2, and crypto protocols competing to define who can delegate spend and under what conditions.

Visa TAP MC Agent Pay AP2

Explore →

Layer 4

Payment Credentials

The spendable instrument itself: tokens, virtual cards, wallet keys, and the missing credential model designed for software agents.

Nekuda Basis Theory Privy

Jump to layer →

Layer 5

Funding Source

Where the money sits before an agent spends it: bank accounts, card lines, stablecoin treasuries, and cross-source routing logic.

Stripe Plaid Circle

Jump to layer →

Layer 6

Checkout / Execution

How agent intent turns into a completed purchase: APIs, cart flows, browser automation, and merchant-side execution logic.

Rye Shopify APIs Browser Use

Jump to layer →

Layer 7

Payment Rails

How money actually moves: card networks, real-time bank rails, stablecoin systems, and on-chain settlement compared.

FedNow Solana Pay Visa / MC

Explore →

Layer 8

Settlement & Reconciliation

What happens after payment executes: ledgering, treasury visibility, reconciliation, and machine-readable audit trails for agent spend.

Modern Treasury Adyen Ledgering

Jump to layer →

Layer 9

Risk / Compliance

Fraud, AML, chargebacks, and liability assignment once AI agents begin initiating more transactions on their own.

Sardine Alloy Unit21

Jump to layer →