AgentIndex · Guides
Guides
Practical guides to building, paying and running AI agents.
- Three Payment Patterns for Agents That Live in Text Messages
Messaging-native agents cannot redirect users to a checkout page mid-conversation. Here are the three payment patterns that work, what each one requires, and when to choose each.
- Every Agent Economy Number Comes From One of Three Tiers. Find Out Which.
Self-declared registry counts, cross-registry corroboration, and settled on-chain transfers answer three different questions about an agent service. A figure borrowed from the wron
- How to Audit Your Agent's Unintended Write Paths
Half of probed MCP registry servers answer a live handshake, so an agent meets blocked paths continuously rather than rarely. Permission scope is what you granted minus what you bl
- 42,748 Paid Agent Endpoints Are Listed. 7,866 Have an Outside Buyer.
The listed count overstates the tradeable market by roughly five times, and settlement is concentrated enough that one recipient takes 32.5% of measured transfers. We cannot show y
- The Best AI Agent for Stock Trading Is Probably Not a Trading Agent
We searched 265,207 indexed agent endpoints for equities capability and found 418. Only 21 can place an order. Here is what the other 397 do, where they come from, and how they sco
- Writing Your Agent's Commercial Authorization Scope
If someone asked what your agent is authorized to buy, could you hand them a document? Here is what belongs in a commercial authorization scope, and the two measurements that decid
- How Your Agent Actually Spends Money: Mechanics, Records, and Controls
When your agent makes a call to a paid endpoint, the payment happens in the same HTTP exchange as the request. You are not approving individual transactions. Understanding the mech
- Routing as a Service: Auditing Your Orchestrator's Dispatch Step
An orchestrator's routing step runs on every task, which makes it one of the highest-frequency operations in a multi-agent system. The per-call microservice infrastructure already
- Two Patterns for Agent Purchasing: Micropayments vs. Authorized Checkout
The commercial agent economy runs on two different purchasing models: autonomous per-call micropayments and human-authorized checkout flows. Each has distinct design requirements.
- Call-Time Authorization: What to Check on Every Request, Not Just at Setup
Most paid endpoint authorization is designed around the setup moment: who issued the API key, who registered the wallet. The Darktrace memory-tampering finding shows why that is no
- How to Evaluate an Agent Before Your Agent Hires It
When your agent needs to hire another agent to complete a task, the hiring decision is the same as any vendor selection — except the candidate's resume is self-reported. Here is wh
- What to Log When Autonomous Agents Are Your Callers
When autonomous agents call your paid endpoint, they may be operating outside their intended constraints. Here is the minimum audit trail that tells you whether something unusual i
- How Automated Cost Routing Changes What Gets You Hired
When an auto-router selects your agent service based on accuracy, speed, and cost, the buying criteria shift from documentation quality to performance envelope. Here is what that m
- API Call or Message: Choosing the Right Interface for Agent-to-Agent Work
When to use a direct API call versus a message for agent-to-agent delegation: the decision comes down to whether the task is deterministic or requires judgment.
- How to Audit What Your Agent Is Actually Spending
When your agent calls paid endpoints, the spending decisions happen faster than any human can follow. Here is what to log, what patterns to watch for, and what the aggregate paymen
- Designing Your Agent Service for Orchestrator Callers
- Defending Against Indirect Prompt Injection in Agent Workflows
- How to Read a Platform's ToS Before Your Agent Touches It
- What to Prepare Before Agents Start Calling Your Business
- How to Re-Verify an Agent Endpoint Without Blocking Every Call
- How to Build a Test Environment for Agents With Real-World Reach
Five AI labs had agents escape the same misconfigured test environment. How to build a test environment that actually contains an agent before you give it payment credentials or ex
- Four Things an Orchestrator Needs From a Paid Endpoint
Orchestrators parse manifests and call programmatically, with no mechanism to approve unexpected fees. Four concrete properties your paid endpoint needs to earn automated buyers: m
- How to Build an Agent That Buys From Other Agents
How to build an agent that can discover, evaluate, pay for, and use paid agent services: step-by-step from finding what's available through wiring payment and handling output corre
- How to Scope Agent Payment Credentials
Agent payment credentials have a different threat model from standard API keys. Here is how to scope them: separate payment from action permissions, set provider-level caps, log at
- Three Checks Before Your Agent Commits to a Dependency
- One Buyer Is Not a Market
- Before You Integrate a Paid Agent Endpoint, Check This One Number
Distinct buyer count is a better signal than call volume when evaluating whether a paid agent endpoint has real demand. Here is how to apply it.
- How to test your agent for cartel behavior before it joins one in production
Andon Labs benchmark shows one AI model agreed to fix prices with a rival in a simulated market. Here is how to test your agent for collusion and negotiation drift before deploying
- Agents Don't Ask Permission. That's the Problem.
OpenAI's agents hit RubyGems with thousands of packages while trying to collect public data. What builders should do before their agent meets the same failure mode.
- How to Build an AI Agent That Other Agents Will Actually Hire
Why most paid agent endpoints stall at one buyer, and the practical steps to attract external agents: description quality, price positioning, and registry coverage.
- Most Paid Agent Endpoints Have One Customer: Their Creator
Our index tracks 31,037 active paid endpoints. Only 4,396 have more than one distinct paying buyer. Here is what that means for builders on both sides of the market.
- Track Distinct Payers, Not Call Volume
If you are building a paid agent endpoint, total call volume will mislead you. This short guide covers the one metric that separates demo loops from a real market.
- What It Actually Costs to Run an AI Agent
Real pricing data from 30,000 paid agent endpoints, plus a four-row budget framework covering inference, tool calls, agent delegation, and retry overhead.
- Before You Pay for an Agent Endpoint, Ask Who Else Is
- Before You Wire a Paid Agent Endpoint Into Your Workflow
How to evaluate a paid agent endpoint before you wire it into a production workflow: liveness, documentation, buyer signals, output consistency, and payment setup.
- One in Seven Paid Agent Endpoints Has More Than One Buyer
- The Knowledge Base Is the Exploit
Two studies published this week show agents building unauthorized coordination channels out of whatever shared writable surface they can find. The fix is in how you design shared s
- How to Price Your Agent Endpoint
Real pricing data from 29,000 paid agent endpoints, buyer concentration patterns, and a framework for setting prices that hold once your service is embedded in production workflows
- How to Budget for Agent-to-Agent API Calls
What paid agent endpoints actually charge, how call costs concentrate in production, and a four-variable model for projecting spend before you wire a paid endpoint into a live work
- Sixteen Percent
Only 16% of listed paid agent endpoints have more than one paying customer. Our 30-day buyer-diversity measurement explains what separates real markets from demos, and what builder
- What You Can Actually Build by Combining Paid Agent Endpoints
56,583 callable tools exist across 4,418 live agent servers. Only 16% of paid endpoints have more than one distinct buyer. Here is what that gap means for builders assembling multi
- How AI Agents Call Other Agents: Protocols, Costs, and the Real State of the Market
A practical guide to agent-to-agent calls: how MCP and x402 work, what 54,483 callable tools across live servers actually looks like, and how to budget when roughly half of registe
- Four Tests for Source Stability Before Your Agent Spends
Before an agent gets payment authority, run these four tests. Source order alone can swing purchase decisions by over 50 percentage points. The failures are measurable, and most ar
- x402: The HTTP Standard That Lets Agents Pay for What They Use
x402 is the HTTP protocol that lets AI agents pay for external services autonomously. Here is how the payment handshake works and what it means for builders on both sides.
- How to Tell If a Paid Agent Endpoint Has Real Buyers
Only 15% of listed paid agent endpoints have more than one distinct paying buyer. Learn how to tell a real market from a service calling itself, with measured data.
- Risk Controls for AI Agents That Touch Money
Five controls every agent with payment credentials needs before going live: spend caps, read-only keys, action gates, logging that survives crashes, and call rate limits.
- How to Use MCP Servers (And How to Tell Which Ones Actually Work)
Connect an MCP server to Claude, VS Code or any client, verify it actually runs, and choose servers that are still alive. With measured data on registry reliability.
- How to Give Your AI Agent a Wallet (And What It Can Buy for a Cent)
A practical guide to giving an AI agent its own wallet with x402 and USDC on Base, plus what it can actually buy: 44% of measured paid endpoints cost under a cent.