Agent Spending Without Controls — Rodrigo Coelho & Pranav Maheshwari, Edge & Node

Ampersand presents the critical need for a new payment infrastructure for AI agents, moving beyond traditional human-centric rails to address machine-speed transactions and stringent enterprise compliance requirements. The presentation highlights how agents require access to paid tools via a "skill file" and demonstrates agentic e-commerce, culminating in a crucial demo of real-time sanctions screening for wallet addresses.

September 1, 2026 5 min read

Agent Spending Without Controls — Rodrigo Coelho & Pranav Maheshwari, Edge & Node — source: https://www.youtube.com/watch?v=ZyGMqdIpPoE

The Missing Infrastructure Layer for AI Payments: Agent Spending with Compliance Controls

Rodrigo Coelho (CEO of Edge & Node, creators of The Graph) and Pranav Maheshwari introduce Ampersand, a project addressing the critical gap in payment infrastructure for the burgeoning agentic economy. They argue that as Large Language Models (LLMs) and autonomous agents begin to transact at machine speed, the existing payment rails—designed for human interaction—are fundamentally inadequate, especially regarding compliance.

The Dawn of Agentic Commerce and Early Micropayment Initiatives

Edge & Node has been at the forefront of agentic commerce, having developed a micropayment system for queries as early as 2021 for The Graph protocol. This system even referenced the HTTP 402 “Payment Required” status code spec years before Coinbase released x402. This early work laid the foundation for their contribution to the x402 spec and their understanding of nano-payments, a field also explored by Circle. The speakers emphasize that while the agentic commerce landscape is rapidly expanding with numerous companies building infrastructure, payment protocols, and governance, a crucial piece for enterprise adoption remains missing.

The Challenge of Traditional Payment Rails vs. Machine Speed Transactions

Coelho highlights that traditional financial payment systems are built on the assumption of a “human in the loop” for decision-making. However, AI agents transact “at machine speed and around the clock,” rendering these human-centric controls obsolete. This fundamental mismatch means that existing policies and rules, designed for human processing times, cannot accommodate the rapid, autonomous nature of agentic transactions. This necessitates a completely new method for managing payments.

Compliance as the Enterprise Bottleneck

The primary barrier to enterprise adoption is not throughput, but compliance. Financial institutions operate within a stringent regulatory environment where identity verification and sanctions screening are paramount. When an agent initiates a transaction, the counterparty often appears as “just a wallet address with no kind of background information on that.” This lack of inherent identity poses a massive challenge for Chief Legal Officers or Chief Policy Officers, who bear the ultimate responsibility for signing off on systems. Getting sanctions screening wrong can lead to fines “into the tens, hundreds, even billions of dollars.” Therefore, a robust compliance layer is non-negotiable for large enterprises to feel comfortable adopting agentic technologies.

The Necessity of Paid Tools for Powerful Agents

Pranav Maheshwari further elaborates on the concept that “agents are only as good as the tools they can pay for.” While many “MCP servers” (micro-computation processing servers or similar tools) are currently free, this will not remain the case. Useful and powerful MCP servers will inevitably become paid services. Expecting users or agents to manually enter credit card details for each service is impractical and inefficient.

Ampersand addresses this by providing a marketplace and an Ampersand skill file for agents. This skill file acts as an aggregator, allowing agents to access a suite of paid MCP tools and handle payments seamlessly in the background.

Ampersand Demos: From Data Retrieval to Compliant Transactions

Maheshwari demonstrates Ampersand’s capabilities through several practical scenarios:

Paid Data Retrieval:
- Scenario: Finding the email address of the head of crypto and blockchain at Mastercard.
- Comparison: One terminal (without Ampersand skill file) returned only the company’s email format and an invitation to figure out the rest. The other terminal (with the Ampersand skill file) paid a fraction of a cent to a metered endpoint and immediately provided the specific email address, the person’s location, and their Twitter handle.
- Implication: This vividly illustrates that valuable, precise information often resides behind paid services, and agents equipped with a payment mechanism (via the skill file) are significantly more capable.
Agentic E-commerce with Budget Constraints:
- Scenario: An agent buying a Father’s Day gift under a $10 budget.
- Process: The agent, leveraging Ampersand’s wallet, autonomously located suitable items, processed the order, and handled the payment without manual intervention. The agent maintained “memory” of preferences and personal information, streamlining the checkout process.
- Vision: This demo showcases the potential for automated, personalized e-commerce where agents handle the entire transaction, moving beyond simple gift-buying to a broader vision of “agentic e-commerce.”
Critical Compliance and Sanctions Screening:
- Scenario: Two agents attempting to use a metered web scraping service, one from an ordinary wallet (“good claw”) and another from a simulated sanctioned wallet address (“bad claw”).
- Demonstration:
  - With compliance screening disabled, both transactions were authorized.
  - With compliance screening enabled (integrated with services like TRM to scan wallet addresses), the “good claw” transaction proceeded normally, while the “bad claw” transaction was instantly rejected due to being flagged as a blocklisted address.
- Conclusion: This demo underscores Ampersand’s crucial role in integrating enterprise-grade compliance into agentic transactions, ensuring adherence to financial regulations and preventing dealings with sanctioned entities. This capability is vital for widespread enterprise adoption.

The Future of Agentic Commerce: Paid MCPs, Wallets, and Compliance

The speakers conclude by reiterating that agentic commerce is becoming real and agents will require sophisticated payment mechanisms. Traditional payment gateways like Stripe are insufficient; agents will need their own proprietary payment frameworks, likely involving wallets integrated with paid MCPs. The success and widespread adoption of this agentic future hinges on providing agents with powerful, paid tools, underpinned by robust wallet infrastructure and, critically, integrated compliance layers to ensure secure and regulated transactions.