Ampersend + Amazon Bedrock AgentCore Payments pay-per-intelligence
Ampersend used Amazon Bedrock AgentCore Payments and the x402 protocol to enable two-hop.
TL;DR
- 01Ampersend used Amazon Bedrock AgentCore Payments and the x402 protocol to enable two-hop.
- 02The Ampersend team completed the full integration in under two weeks; without AgentCore Payments they estimated wallet custody and spending controls would have taken 3–4 months of engineering work.
- 03Behind that summary are explicit components and constraints.
Ampersend used Amazon Bedrock AgentCore Payments to build a pay-per-intelligence routing layer that lets AI agents pay per request, operate within session budgets, and settle on-chain in USDC on the Base network. The Ampersend team completed the full integration in under two weeks; without AgentCore Payments they estimated wallet custody and spending controls would have taken 3–4 months of engineering work.
How does the two-hop payment flow work?
The two-hop pattern routes a single agent request through Ampersend to an upstream model provider while separating buyer and seller settlement: the agent pays Ampersend, and Ampersend pays the model provider. An agent calls Ampersend, receives an HTTP 402 (payment required), invokes AgentCore ProcessPayment with x402 details and a signed USDC authorization, retries the request with the proof, and Ampersend verifies settlement on-chain before auto-paying the provider.
Behind that summary are explicit components and constraints. The application backend creates a Payment Manager that defines wallet connections and spending policies. Before the agent starts, the backend opens a Payment Session with a budget cap (the post gives $0.05 as an example). AgentCore signs payments using connected wallet credentials so the agent never touches private keys. Two settlements occur in each transaction: agent to Ampersend, then Ampersend to the model provider.
How does AgentCore Payments manage the payment lifecycle?
AgentCore Payments provides managed wallets, deterministic spending governance, native x402 signing, and observability so neither Ampersend nor agent builders must build custody or signing infrastructure. AgentCore connects to Coinbase Developer Platform (Coinbase CDP) or Stripe Privy wallets for custody; Ampersend connected credentials once and had funded wallets ready to transact.
Session-level budgets are enforced at the infrastructure layer. The backend sets limits; if the budget is exhausted the next payment is rejected cleanly. AgentCore handles x402 protocol negotiation, wallet authentication, stablecoin payment, and proof delivery without interrupting the agent’s reasoning loop, and it supports both v1 and v2 of the x402 protocol. Observability is unified: every transaction flows through the same logs, metrics, and traces developers already use in AgentCore.
What Ampersend adds and what the integration delivered
Ampersend provides model routing, a provider marketplace, the two-hop settlement pattern, and operational tooling that abstracts provider complexity from agents. Ampersend’s SDK handles upstream settlement to model providers such as BlockRun, while AgentCore handles managed wallet custody, signing, and spending guardrails. The combined stack lets an agent discover, select, and pay for intelligence services through a single integration point, with Ampersend settling upstream and verifying payment proofs on-chain (Base network, USDC).
Kevin Jones, who led the integration, summarized the developer experience: "Building multi-agent systems is genuinely complex, and we expected payment infrastructure to be the hardest part of the entire project. AWS AgentCore Payments completely changed that." Rodrigo Coelho, CEO of Edge & Node, said Ampersend was built as the control layer for agent payments and described AgentCore Payments as a natural fit because of its managed wallets, guardrails and x402 settlement.
Why it matters
This pairing separates payment plumbing from routing logic, shrinking the engineering barrier to agentic commerce. By offloading wallet custody, signing, and budget enforcement to a managed service, agent builders can integrate paid models without bespoke billing, credential handling, or per-provider contracts. That reduces integration time dramatically: Ampersend’s end-to-end integration took under two weeks versus an internal estimate of 3–4 months to build the same wallet and governance features.
What to watch
Watch for broader adoption signals: more model providers integrated into Ampersend’s marketplace, additional wallet providers beyond Coinbase CDP and Stripe Privy, and wider support for x402 v1 and v2. The AgentCore CLI quickstart, samples and workshops on GitHub will show working buyer/seller implementations and will be the next public testbed for production use cases.
Written by The Brieftide · Source: AWS Machine Learning
The Brieftide Daily · 06:00
Briefs like this one, in your inbox every morning.
Continue reading
More in Coding AgentsData2Story: CSV-to-article pipeline with seven AI agents
A Claude Code skill runs seven specialist agents to turn a CSV into a verifiable, interactive news article with an Inspector panel.
Vibe Coding: AI evaluation for greenfield software engineering
Callum Barbour's arXiv paper tests 'vibe coding' on isolated Python greenfield tasks using a custom evaluation suite.
CODA-BENCH benchmark: testing code agents on data tasks
CODA-BENCH places agents in a Kaggle-based Linux sandbox with 1,009 tasks across 31 communities and an average of 980 files per task.
SWE-Explore: benchmark shows AI coding agents miss key lines
SWE-Explore isolates code search from repair and finds agents hit the right files but cover only 14–19% of the lines that matter.