Agentspay

Explainer

AI Agent Governance: Agentic AI Governance, AI Agent Guardrails and Governance Framework

Every agentic AI governance framework published so far governs the same four things: identity, tools, data and prompts. Not one of them carries a budget. Here is what the real frameworks say, which guardrails actually bind at runtime, and what to do about the last mile none of them reach.

Agent Payments Console

Pick an agent

Payment intent

intent:

Policy evaluation

Human approval required

This spend is over your approval threshold. Approve it to issue a scoped card, or deny it.

Scoped virtual card issued

Agentspay

single-use

Wallet budget

spent of

Audit trail

In short

AI agent governance is the set of controls that decide what an autonomous AI agent is allowed to do, prove what it did afterwards, and stop it while it is still acting. The 2026 reference points are the NIST AI Agent Standards Initiative launched February 17, 2026 by the Center for AI Standards and Innovation, the OWASP Top 10 for Agentic Applications published December 9, 2025 covering ASI01 through ASI10, and Gartner’s forecast that 40% of enterprise applications will embed task-specific AI agents by the end of 2026, up from under 5% in 2025. Almost every published framework governs identity, tools, data and prompts, then stops at the point where an agent spends money: none of them defines a budget, a merchant allowlist or an approval threshold. That last control has to be enforced at the payment rail, because a policy document cannot decline a transaction.

What is AI agent governance?

AI agent governance is the practice of deciding in advance, in enforceable terms, what an autonomous agent may do on your behalf, and then proving afterwards what it actually did. Five things are in scope. Identity: which agent this is and which human or team it acts for. Authority: the tools, systems, data and budgets it can reach. Constraints: the limits, thresholds and allowlists that bind while it runs. Oversight: the decisions a person has to approve before they happen. Evidence: a record an auditor can read six months later. Most vendor material stretches the word governance to mean a policy document and a dashboard. The narrower definition is the useful one: a governance control is only real if it can refuse an action while the action is still in flight. Everything else is reporting.

Why is AI agent governance different from AI governance?

Traditional AI governance is model-centric. It asks whether a model is fair, accurate, explainable and documented, and it lives in evaluation suites, model cards, bias testing and the paperwork that satisfies the NIST AI Risk Management Framework or the EU AI Act. All of that still matters, and none of it constrains an agent. A model produces text. An agent holds credentials, calls tools, keeps memory across steps and chains actions together until a goal is met or something breaks. The unit of risk moves from the output to the action, and the question moves from was this answer good to should this system have been allowed to do that. That is why practitioners describe the shift as moving from model-centric governance to authority-centric governance, and from after-the-fact review to real-time enforcement. It is also why an approval workflow that reviews a monthly report is not governance for something that acts in 400 milliseconds.

What does the NIST AI Agent Standards Initiative cover?

NIST’s Center for AI Standards and Innovation, CAISI, launched the AI Agent Standards Initiative on February 17, 2026, the first dedicated US federal program aimed at interoperability and security for autonomous agents. It is organized in three pillars: facilitating industry-led standards development, supporting community-led open-source protocol work, and funding foundational security and identity research. The first pillar runs technical convenings and gap analyses and states an explicit intent to strengthen US participation in international standards setting through ISO/IEC JTC 1. Alongside it, the National Cybersecurity Center of Excellence published a concept paper in February 2026 proposing to adapt existing identity and authorization frameworks to agents rather than invent new ones, and NIST held virtual listening sessions in April 2026 aimed at healthcare, financial services and education. Nothing here is binding yet. If you are building a program in 2026, treat CAISI as the direction of travel for identity and authorization, not as a compliance deadline.

What is the OWASP Top 10 for Agentic Applications?

The OWASP Gen AI Security Project published the Top 10 for Agentic Applications on December 9, 2025, developed with more than 100 industry experts, researchers and practitioners. It catalogs ten risk categories numbered ASI01 to ASI10 that only appear once a model stops generating text and starts acting with delegated authority, credentials, memory and tools. It is the most practical starting point available because it is written as a threat list rather than a set of principles, which means each entry maps to a control you can actually implement. Microsoft maintains an open-source agent governance toolkit on GitHub that claims coverage of all ten. The table below adds the column the security-focused write-ups leave out: what each risk looks like once the agent has a payment method attached.

What are AI agent guardrails?

Guardrails are the runtime half of governance: the constraints that bind while an agent is executing, as opposed to the policies that describe what should happen. In practice they fall into four layers. Input guardrails filter and validate what reaches the model, which is where prompt injection defenses sit. Tool guardrails decide which functions an agent can call at all, usually through an allowlist, least-privilege scoping and short-lived credentials. Action guardrails constrain the effect of a call rather than the call itself: rate limits, transaction caps, merchant allowlists, approval thresholds and a kill switch. Output guardrails check what comes back before anything downstream consumes it. An honest ai agent guardrails example: an agent told to restock inventory has a tool guardrail if it can only call the purchasing API, and an action guardrail only if a 40,000 dollar order it was never meant to place gets declined by something the agent cannot talk its way past.

Which layer can actually stop an agent?

There are three places to enforce a rule, and they are not equally strong. The prompt layer is instruction, not enforcement: a system prompt saying never spend more than 500 dollars is a preference that a goal-hijack attack, a confused plan or a long context window can quietly erode. The tool layer is genuinely strong for access, because an API the agent has no credential for is closed, but it degrades once the tool is legitimately available and the question becomes how much and how often. The rail layer, meaning the payment network, the card, the wallet or the settlement contract, is the only one the agent cannot reason with. It sees the amount, the merchant and the counterparty, and it either authorizes or declines. A useful rule when you design a control: if the agent could argue its way out of it, it is not a guardrail.

Where every agentic AI governance framework stops

Read the specifications rather than the summaries and the gap is specific. In Google’s A2A protocol, a Task and an Artifact have no price field at all, so the core spec has no budget, no cap and no approval primitive. The a2a-x402 payment extension negotiates per request, so it sees one payment at a time and keeps no cumulative memory across tasks. AP2 mandates prove that a human authorized a specific purchase and say nothing about the tenth purchase that day. x402 settles a single request. Visa Intelligent Commerce binds spending limits to the consumer mandate and the card issuer, not to the buying company’s finance team. Amazon Bedrock AgentCore Payments checks a maxSpendAmount inside one payment session. Every one of these is a correct design decision at the protocol level and a hole at the program level, because the risk that actually costs money is cumulative: forty small purchases nobody authorized in aggregate, from a merchant nobody vetted, discovered at month end.

Which controls do the payment rails already enforce?

The rails are further along than the frameworks, and it is worth knowing exactly how far. Mastercard Agentic Tokens are network-issued, time-scoped and revocable, and they are scoped per agent rather than per device, so two agents acting for the same person carry two tokens and two policies and revoking one leaves the other running. Visa Intelligent Commerce binds a consumer mandate to a device with FIDO and Visa Payment Passkeys, then issues a merchant-specific token constrained by that mandate. Stripe Shared Payment Tokens carry a scope and published minimums of 0.50 dollars for cards and 0.01 USDC for stablecoin settlement. Virtual cards remain the bluntest and most reliable instrument: one card per agent or per vendor, with a hard cap, an expiry and a merchant category restriction that no prompt can override. The table below sets out which of these enforces what, and the pattern is consistent: rails are excellent at scoping a single instrument and weak at anything that has to hold across a whole fleet.

What platforms provide AI agent governance?

Four categories exist and they solve different problems, so the honest answer is that most enterprises end up with more than one. Cloud platform governance, such as Microsoft’s Cloud Adoption Framework guidance for agents or Microsoft Agent 365, gives you inventory, identity and policy where your agents already run, and is the cheapest place to start if you are on a single cloud. Worth knowing before you rely on it: we read the machine readable Graph schema behind Agent 365 and found 623 properties across its agent governance types, none of which expresses a spending limit. The same is true one platform over: the Salesforce object model documents 178 money objects and 41 agent objects with no overlap at all, which is why Agentforce spend controls stop at capability rather than amount. Google is the exception worth knowing about: the Gemini Enterprise spend controls we measured do enforce a hard monthly ceiling, but it is a Cloud Billing budget whose nine filter dimensions contain no way to name an agent, so it stops every agent in the project at once or none of them. AI security and posture vendors cover the OWASP threat surface: prompt injection, tool misuse, shadow agents and data egress. GRC and compliance platforms map agent activity to control frameworks and produce the evidence auditors ask for. Payment-side controls, which is the category this site sits in, bind budget, merchant scope and approval thresholds to the money itself. Ask each vendor the same question: can this decline an action in real time, or does it tell me afterwards? Both answers are legitimate, and confusing the two is how programs end up with excellent dashboards and no brakes.

What belongs in an AI agent governance framework?

A framework that survives contact with a real deployment has eight parts. A registry of every agent in production, because you cannot govern a fleet you cannot enumerate, and shadow agents are the normal failure. A named owner per agent, a person rather than a team inbox. A tier assigned by blast radius, so a research agent that reads public data is not reviewed like one holding a corporate card. Least privilege on tools and data, with short-lived credentials rather than static keys. Action limits that bind at runtime: caps, rate limits, allowlists. Approval thresholds that route the small set of decisions worth a human to an approval queue and let everything below them run untouched. A kill switch that revokes credentials and payment instruments in one action, tested rather than assumed. And a complete audit trail that reconstructs what happened without asking the agent to explain itself.

Who is responsible for AI agent governance?

Responsibility is shared across model providers, platform operators and the deploying organization, but the deploying organization keeps the primary share, because it defines permissions, approves use cases and decides how much authority an agent receives. The practical recommendation across the current guidance is to assign agent governance to the same leaders who already own cloud governance, security and compliance, rather than creating a parallel committee. Inside that, the split that works is functional: security owns identity and credentials, the platform team owns runtime enforcement, and finance owns the budget, the merchant allowlist and the approval thresholds, because those are already finance controls applied to a new kind of buyer. The failure mode to avoid is the one where every function assumes another one set the spending limit. In most organizations that has been running agents for a quarter or more, nobody did.

How do you audit an AI agent?

An audit of an agent has to answer six questions from evidence rather than reconstruction: which agent acted, which human or team it acted for, under what authority, what it did, what it cost, and who approved it. That is harder than it sounds, because agent activity is usually spread across model traces, tool logs, cloud audit logs and a card statement that arrives weeks later with a merchant name and no context. The fix is to record the authorization at the moment of the action rather than assembling it afterwards. If your agents transact, attach the agent identity to the payment instrument so the transaction record itself carries the attribution, which is what per-agent virtual cards and per-agent network tokens both make possible. Reconciliation stops being an investigation and becomes a lookup.

Agentic AI governance and risk management strategy for enterprises

The strategic problem is scale rather than principle. Gartner expects 40% of enterprise applications to embed task-specific AI agents by the end of 2026, up from fewer than 5% in 2025, and in the same body of research projects agentic AI driving roughly 30% of enterprise application software revenue by 2035, above 450 billion dollars, from 2% in 2025. A governance approach built on human review, a spreadsheet inventory and team-by-team policy implementation works at a dozen agents and collapses somewhere in the hundreds. Two design choices survive that transition. Tier by blast radius so the review effort concentrates where the damage would be, and make the controls that matter automatic and non-bypassable so they scale at zero marginal cost. A spending limit enforced at the rail costs the same to run with 5 agents or 5,000. A weekly review meeting does not.

A 30-day AI agent governance rollout

If you are starting from nothing, this order gets a defensible baseline in a month without stalling the teams shipping agents. Week 1: enumerate. Find every agent in production, including the ones running on a team credit card, and record owner, purpose, tools and whether it can spend. Week 2: tier and scope. Rank by blast radius, revoke static credentials in favor of short-lived ones, and cut every tool allowlist to what the agent demonstrably uses. Week 3: bind the money. Give every spending agent its own instrument with a hard cap and a merchant scope, set an approval threshold above which a human signs off, and confirm you can revoke in one action. Week 4: prove it. Run a tabletop where an agent tries to exceed its limit and buy from an unapproved merchant, and check that the decline appears in the audit trail with the agent named. If the tabletop passes, you have governance. If it only produces an alert, you have monitoring.

Whatever standard moves the money, Agentspay is the rail-neutral control plane that keeps it governed. See how it works and the control surfaces that enforce policy, approvals, and audit on every transaction.

The threat list

OWASP Top 10 for Agentic Applications 2026, and what each risk costs

Published December 9, 2025 by the OWASP Gen AI Security Project. The third column is ours: the same risk restated for an agent that holds a payment method.

ID Official name What it looks like once the agent can spend
ASI01 Agent Goal Hijack The objective is rewritten, so the purchase is real, authorized and wrong
ASI02 Tool Misuse and Exploitation A legitimate purchasing API called far outside its intended pattern
ASI03 Identity and Privilege Abuse One shared credential means no transaction can be attributed to one agent
ASI04 Agentic Supply Chain Vulnerabilities A compromised third-party skill inherits your payment authority
ASI05 Unexpected Code Execution Arbitrary execution inside a process that already holds card credentials
ASI06 Memory and Context Poisoning A planted merchant or invoice persists across sessions and looks routine
ASI07 Insecure Inter-Agent Communication A delegated task arrives with no proof of who authorized the spend
ASI08 Cascading Failures One bad signal replicated across a fleet, so the loss multiplies
ASI09 Human-Agent Trust Exploitation Approval fatigue, where a reviewer waves through the tenth request
ASI10 Rogue Agents An agent nobody registered, spending on an instrument nobody scoped

Enforcement, honestly

What each governance layer can and cannot refuse

The distinction that matters in procurement: detection tells you afterwards, enforcement declines while it is happening.

Layer Typical control Can it decline in real time? Where it breaks
System prompt Written instruction to the model No Goal hijack, long context, plan drift
Tool allowlist Which functions the agent may call Yes, for access Silent once the tool is legitimately allowed
Identity and credentials Short-lived, scoped tokens Yes, for access Says nothing about amount or frequency
Observability Traces, logs, anomaly alerts No Tells you after the money left
Approval workflow Human sign-off above a threshold Yes Fails to approval fatigue if the threshold is too low
Payment rail Cap, merchant scope, revocation Yes Usually scoped per instrument, not per fleet

The money layer

Which agent controls each payment rail actually enforces

Checked against each provider’s own documentation in August 2026. Yes means the mechanism is published and enforceable, not that a control of that name exists in marketing copy.

Mechanism Per-agent identity Hard spend cap Merchant scope Cumulative budget Instant revocation
Mastercard Agentic Tokens Yes, per agent Issuer policy Issuer policy No Yes
Visa Intelligent Commerce Yes, per mandate Consumer mandate Merchant-specific token No Yes
AP2 mandates Yes, signed Per purchase Per purchase No Not applicable
x402 and a2a-x402 Per request Per request No No Not applicable
Stripe Shared Payment Tokens Per session Scoped to token Scoped to token No Yes
Bedrock AgentCore Payments Per session maxSpendAmount No No Yes
Per-agent virtual cards Yes, per card Yes Yes, by category Yes, per card Yes

Frequently asked

Questions people ask about AI agent governance

What is AI agent governance?

AI agent governance is the set of controls that decide what an autonomous AI agent may do, prove what it did, and stop it while it is acting. It covers identity, authority over tools and budgets, runtime constraints, human approval for high-impact decisions, and an audit trail. It differs from AI governance, which is model-centric and concerns fairness, accuracy and documentation rather than authority.

What is the difference between AI governance and agentic AI governance?

AI governance evaluates a model: is it accurate, fair, explainable and documented. Agentic AI governance constrains an actor: which tools it may call, how much it may spend, when a human must approve, and what evidence exists afterwards. The practical shift is from model-centric review to authority-centric enforcement, and from periodic audit to real-time control.

What are AI agent guardrails?

Guardrails are the constraints that bind while an agent runs, in four layers: input guardrails that filter what reaches the model, tool guardrails that limit which functions it can call, action guardrails such as spend caps and approval thresholds that limit the effect of a call, and output guardrails that check results. A guardrail that the agent can reason its way past is a preference, not a guardrail.

What platforms provide AI agent governance?

Four categories: cloud platform governance such as Microsoft’s Cloud Adoption Framework guidance, AI security and posture vendors covering the OWASP threat surface, GRC platforms that map agent activity to control frameworks, and payment-side controls that bind budget and merchant scope to the money. Most enterprises run more than one. Ask each whether it can decline an action in real time or only report it.

What is the NIST AI Agent Standards Initiative?

NIST’s Center for AI Standards and Innovation launched it on February 17, 2026 as the first dedicated US federal program for AI agent interoperability and security. It runs three pillars: industry-led standards development, community-led open-source protocol work, and foundational security and identity research, with explicit intent to strengthen US participation in ISO/IEC JTC 1.

What is the OWASP Top 10 for Agentic Applications?

It is a threat list published December 9, 2025 by the OWASP Gen AI Security Project, developed with more than 100 contributors, covering ASI01 Agent Goal Hijack through ASI10 Rogue Agents. It is written as risks rather than principles, so each entry maps to an implementable control, which makes it the most practical starting point for an agent governance program.

Who is responsible for AI agent governance?

The deploying organization holds primary responsibility because it defines permissions, approves use cases and sets how much authority an agent receives, though model providers and platform operators share part of it. Current guidance recommends assigning it to the leaders who already own cloud governance, security and compliance rather than forming a separate committee.

How do you set a spending limit for an AI agent?

Bind the limit to the payment instrument rather than the prompt. Issue a scoped instrument per agent, whether that is a virtual card, a network token or a session token, with a hard cap, an expiry and a merchant scope, then route anything above a threshold to a human approver. A limit written into a system prompt is advisory and will not decline a transaction.

Can an AI agent governance framework stop an agent from overspending?

Not by itself. Published frameworks including the OWASP list and the NIST initiative govern identity, tools, data and prompts, and none of them defines a budget, a merchant allowlist or an approval threshold. Those are enforceable only at the payment rail, which sees the amount and the merchant and can decline. A framework tells you which control you need, it does not apply it.

How do you audit an AI agent?

Record the authorization at the moment of the action instead of reconstructing it later. An audit has to show which agent acted, who it acted for, under what authority, what it did, what it cost and who approved it. Attaching agent identity to the payment instrument makes the transaction record carry the attribution, which turns month-end reconciliation into a lookup instead of an investigation.

Do AI agents need their own identity?

Yes. A shared service account means no action can be attributed to a single agent, no limit can be scoped to one, and revoking access stops the whole fleet. OWASP lists Identity and Privilege Abuse as ASI03 for this reason, and NIST’s NCCoE published a February 2026 concept paper on adapting existing identity and authorization frameworks to agents rather than inventing new ones.

Keep reading

More explainers

ServiceNow AI Control Tower

ServiceNow AI Control Tower

ServiceNow AI Control Tower is the most complete agent inventory and risk console a large US enterprise can buy, and it now reaches across AWS, Google Cloud and Azure. We read the schema ServiceNow ships to developers to answer the one question the rollout meeting always ends on: can it stop an agent from spending money? It cannot, and the reason is written into the data model.

Read

Gemini Enterprise

Gemini Enterprise

Google did something in August 2026 that the other agent platforms have not done: it shipped a hard monthly spend cap that genuinely stops usage instead of emailing you about it. That deserves credit, and it also moves the interesting question one step along. A cap that stops something is only as useful as the thing it is scoped to, so we went and measured what Google can actually point that cap at, in the API model Google publishes for anyone to read.

Read

Salesforce Agentforce

Salesforce Agentforce

Agentforce is the largest agent platform any US enterprise is likely to already own, and it moved to consumption billing, which means the meter now runs on what your agents do rather than on how many seats you bought. That raises a finance question the rollout deck rarely answers: when an Agentforce agent is loose in production, what actually stops it spending. We went and measured the answer in Salesforce own published object model rather than guessing at it.

Read

AWS AgentCore

AWS AgentCore

Amazon shipped the missing piece in August 2026. Bedrock AgentCore Payments went generally available, and it is a real payments product: an agent can now hold a wallet, meet an HTTP 402, pay, and carry on reasoning without a human in the loop. So the question a platform lead has to answer stopped being whether AWS gives agents money and became a narrower, more awkward one: how much of a spend policy did AWS actually ship? We went and measured it, property by property, in the API model AWS publishes.

Read

Microsoft Agent 365

Microsoft Agent 365

Microsoft shipped a control plane for AI agents, and it is a good one. It gives every agent an identity, a registry entry, an owner, a sponsor and a Conditional Access policy. Then somebody in finance asks the obvious follow-up question: fine, but what stops the agent from spending money? This page answers what Agent 365 costs, what it governs, and what we measured when we went looking for a dollar amount anywhere in Microsoft's agent governance surface.

Read

QuickBooks MCP Server

QuickBooks MCP server

Connecting an accounting system to an AI assistant is now a ten minute job. Deciding what that assistant is allowed to do once it is connected is the part nobody writes about, and it is the part your controller will ask about first. This page compares what the official QuickBooks, NetSuite and Xero MCP servers actually hand a model, measured rather than summarized from marketing pages.

Read

Payment MCP Servers

payment MCP servers

Every large payment company shipped an MCP server in the last eighteen months, and almost every write-up of them is a setup tutorial. The setup is the easy part. The question worth answering before you connect one to a production account is narrower and much less comfortable: what, exactly, can the model on the other end of that connection do to your money?

Read

PayPal Agentic Commerce

PayPal Agentic Commerce

PayPal made a bet that most merchants would rather not implement a commerce protocol at all. Where Stripe and OpenAI shipped a spec for you to build against, PayPal shipped two products that sit on top of the checkout you already have, and then bought a company to make the catalog half work. That choice is the whole story: it explains why Agent Ready needs almost no engineering from you, why there is nothing for an agent to discover about your store on the open web, and why the thing PayPal will not do for you is the thing that gets expensive later.

Read

Shopify Agentic Commerce

Shopify Agentic Commerce

Shopify switched agentic commerce on by default, so your store is probably already selling to AI assistants whether or not anyone on your team configured it. Instead of restating the announcement, we checked something you can check too: on September 2, 2026 we requested the machine-readable capability file that Shopify publishes for real storefronts, on fourteen well-known US brand domains, and read what it exposes to an agent. Eleven answered correctly. The three that did not share one trait, and it is quietly costing them agent traffic.

Read

Web Bot Auth

Web Bot Auth

Web Bot Auth is the reason your agent either gets served or gets throttled with the scrapers. Almost everything written about it repeats the same architecture diagram, so we did something different: on September 1, 2026 we fetched the published key directories of more than twenty major AI operators and infrastructure vendors to see who is genuinely signing their traffic. Four were. The results are in the first table.

Read

Tempo Blockchain

the Tempo blockchain

Tempo is the payments chain Stripe and Paradigm built, and it shipped with a protocol that lets software pay for things on its own. It settles machine payments in under a second. It has nothing at all to say about whether your agent should have paid.

Read

A2A Protocol

A2A Protocol

Most explanations of the A2A protocol stop at the sentence that agents can now talk to each other. That was true in April 2025 and it is no longer the interesting part. A2A shipped version 1.0 in April 2026 under Linux Foundation governance, it runs in production inside Azure AI Foundry and Amazon Bedrock AgentCore, and the questions engineers actually get stuck on are narrower: what an Agent Card commits you to, when to reach for MCP instead, and what happens the first time one of your agents has to pay another one for the work. That last question has a specific answer, and it is not in the core spec.

Read

Mastercard Agent Pay

Mastercard Agent Pay

Nearly every article about Mastercard Agent Pay is a retelling of one press release from April 2025, the one where Mastercard said AI agents would be able to shop with Agentic Tokens and named Microsoft as the first platform. That was sixteen months ago, and four more things have shipped since. Reading only the launch coverage leaves you with roughly a quarter of the picture, and the missing three quarters are the parts that decide whether you can actually put this into production.

Read

Visa Intelligent Commerce

Visa Intelligent Commerce

Almost everything written about Visa Intelligent Commerce is a retelling of the April 2025 announcement, when Visa said AI agents would be able to pay with a Visa credential. Three more things have shipped since, including an open agent-identity protocol built with Cloudflare that most coverage does not mention at all. This page is the current version, checked against Visa’s own developer documentation and newsroom in August 2026.

Read

Stripe agentic commerce

Stripe agentic commerce

Most writing about Stripe and agentic commerce is still a retelling of the September 2025 launch week, when Stripe and OpenAI shipped Instant Checkout and published the Agentic Commerce Protocol together. Stripe has built a good deal more since then, and some of it points in a direction the launch coverage never anticipated. This page is the current version, checked against Stripe’s own documentation in August 2026.

Read

ChatGPT Instant Checkout

ChatGPT Instant Checkout

Almost every guide to ChatGPT Instant Checkout still reads like it was written the week it launched, walking merchants through how to apply and what the fee will be. OpenAI changed course in March 2026. Here is the accurate version: what Instant Checkout was, what the numbers actually looked like, what replaced it, and which parts of the stack are still very much alive.

Read

Google AP2

Google AP2

Most guides to Google AP2 still describe an Intent Mandate and a Cart Mandate, because most of them are rewrites of the September 2025 launch post. The specification moved. Here is what the Agent Payments Protocol actually defines today, and the one question it deliberately does not answer.

Read

Human in the loop AI

Human in the Loop AI

Every guide to human in the loop AI describes the same shape: the agent pauses, a person decides, the agent continues. The shape is right. What almost none of them ask is a harder question, which is where the pause is enforced, because a pause written into the agent's own code is a pause the agent is trusted to honor.

Read

AI agent cost

AI Agent Cost

Every cost guide for AI agents answers the same two questions: what does it cost to build, and what does it cost to run. Both are answerable, and both are on somebody's invoice. The third question is the one that ends up in a variance report, because the agent also spends your money, and nobody sends you a bill for that.

Read

Agentic checkout

Agentic Checkout

Nearly every guide to agentic checkout is written for the merchant who wants to receive these orders. Far fewer are written for the company whose agents are placing them, which is odd, because agentic checkout quietly removes the one screen where spending used to get a second look.

Read

API monetization

API Monetization

Most guides to API monetization argue about which pricing model wins. The harder question in 2026 is who is calling. An API priced for a signed-up developer with a key behaves very differently when the caller is an agent that showed up once, wants one record, and has no account.

Read

x402 protocol

x402 Protocol

x402 took the one HTTP status code the web never used and turned it into a payment rail machines can drive. The protocol is elegant and genuinely small. The part it deliberately leaves to you is the budget.

Read

AI procurement agents

AI Procurement Agents

Every major procurement suite shipped agents during 2026. Almost none of them answer the question your controller will ask first, which is what happens when the agent is wrong about a purchase and the money has already moved.

Read

Agentic payments

Agentic Payments

Agentic payments move money with no human at the checkout. The rails to do it all shipped during 2026. The part most teams have not solved is deciding, before the money moves, whether the agent was allowed to spend it.

Read

AI agent monetization

AI Agent Monetization

Every AI agent company is rewriting its price list. The models that survive are metered. The ones that quietly fail are the ones where nobody measured what a single task costs to serve.

Read

Agent payment platforms

AI Agent Payment Platforms

Five different kinds of product now call themselves an AI agent payment platform, and they solve five different problems. Picking the wrong category is the expensive mistake, not picking the wrong vendor inside a category.

Read

Universal Commerce Protocol

the Universal Commerce Protocol (UCP)

Google and Shopify shipped UCP as an open standard so an AI agent can check out at any merchant that supports it. Here is what the specification actually defines, where it is live for US buyers, and the one thing it deliberately leaves to you.

Read

MCP Payments

MCP Payments

MCP payments are how an AI agent discovers a payment tool and calls it to move money. The catch: the Model Context Protocol carries the tool call, not the spending decision, so nothing in the stack asks whether the purchase should have happened.

Read

Visa Intelligent Commerce vs Mastercard Agent Pay

Visa Intelligent Commerce vs Mastercard Agent Pay

Visa Intelligent Commerce and Mastercard Agent Pay are the two big card networks racing to let AI agents pay. They take different routes to the same idea, and neither one decides whether a given purchase should have happened.

Read

Agentic Commerce Protocol

the Agentic Commerce Protocol

ACP is the open standard behind agentic checkout in ChatGPT. It tells a merchant how to sell to an AI agent. It says nothing about whether your agent should have made the purchase.

Read

AP2 vs ACP vs x402

AP2 vs ACP vs x402

AP2, ACP, and x402 are the three standards shaping how AI agents pay. They solve different layers of the problem, and most real systems will touch more than one.

Read

Machine payments protocol

Machine payments protocol

As software starts paying software, machine payments protocols define how value moves without a human at the keyboard. The harder question is how to keep that spending governed.

Read

Know Your Agent (KYA)

Know Your Agent

KYA, or Know Your Agent, extends the idea of customer due diligence to autonomous software. When an agent spends, you need to know which agent, on whose authority, and under what limits.

Read

Keep agent spending governed

Add policy, hard limits, human approval, and an immutable audit trail across any protocol or rail. Start in the sandbox today.

Never moves money without policy