Anthropic's real-world testing of the AI-powered trading market: Six steps to let AI automatically sell products, quote prices, and negotiate

Internal testing of fully AI-mediated trading at Anthropic, six steps for AI to autonomously complete quoting and buying/selling. Seeing the trend of AI agent payments, payment giants are actively deploying AI commerce; and blockchain and cryptocurrencies, with their low-cost features, are expected to become the backbone supporting AI payments in the future?

Anthropic’s internal AI trading market test, six steps to achieve fully automated trading

Without manual intervention, is the future where AI agents (AI Agents) help us shop and sell getting closer? Recently, Anthropic conducted an internal experiment called “Project Deal,” allowing AI to autonomously buy and sell items. The company invited 69 employees, each given a $100 budget, ultimately completing 186 transactions with a total value of over $4,000.

In the test, employees’ AI agents completed the process from negotiation to delivery. The official article showcases transactions between employee Rowan and Gabby, which involved six specific steps:

1. Initial interview (Intake interview): Participants converse with AI to set the items for sale, ideal prices, and negotiation styles. For example, Rowan requested the AI to negotiate in a down-and-out cowboy tone.
2. Agent assignment: Based on the interview, participants receive a customized, dedicated Claude agent.
3. Deploy agents: The system deploys these agents into four parallel Slack communication channels, ready to interact.
4. Agents negotiate: AI autonomously posts item listings, makes offers, and counters without human intervention. Rowan’s agent pitches a $10 handmade painting in cowboy style, while Gabby’s agent expresses a love for horror-themed art and negotiates the price.
5. Deal execution: Both sides’ agents reach consensus in the channels and officially finalize the transaction.
6. In-person exchange: After online agreement, participants meet in person at the office to exchange physical items according to the AI agents’ negotiated terms.

Image source: Anthropic　Internal AI trading market test at Anthropic, six steps to complete fully automated trading

International payment giants compete in the AI commerce arena

Anthropic’s experiment demonstrates an initial attempt for humans to free their hands through AI agents to automatically buy and sell goods. Amid the AI agent boom, international payment giants have already entered the AI agent payment field.

Mastercard Taiwan General Manager Chen Yiwen said that agent-based AI has integrated into daily payments. Mastercard recently partnered with Taipei Fubon Bank to complete the first agent-based AI transaction booking airport transfers.

Mastercard also collaborates with AI agent payment platform Lobster.cash, integrating its Mastercard Agent Pay and Verifiable Intent framework into an open-source AI agent platform. After integration, users don’t need to apply for new wallets or cards; existing cards can serve AI agents.

Visa, Mastercard’s competitor, is also active, besides the command-line interface “Visa CLI” designed specifically for AI agents, it also cooperates with Coinbase on the x402 protocol.

Related reports:
AI agent payments are booming! Visa launches CLI payment tool, enabling AI to automatically swipe cards

Mastercard advances into AI card payments! Completes Taiwan’s first AI agent payment with Fubon Bank, integrating Lobster.cash

Blockchain and cryptocurrencies may become the backbone of AI payments

Renowned venture capital firm a16z crypto predicts that by 2030, the scale of autonomous AI trading will reach $30 trillion. While payment giants are deploying AI agent payments, the blockchain industry has also found potential entry points.

Traditional finance faces high transaction fees and limitations on issuing cards for AI, but cryptocurrencies’ 24/7 operation and low costs make them promising as the future backbone of AI agent payments.

Cloud service provider Cloudflare has announced the upcoming launch of the micro-payment stablecoin NET Dollar and has joined the x402 protocol promotion; meanwhile, Cardano founder Charles Hoskinson has focused on privacy, launching the sidechain Midnight at the end of March this year, introducing the privacy stablecoin USDCx.

From Anthropic’s internal experiments to the proactive deployment by traditional payment giants, AI agents have demonstrated a vast potential to go beyond simple command execution and directly participate in economic activities.

In this shift, blockchain and cryptocurrency technologies are poised to fill gaps that traditional finance cannot support, such as micro-payments and privacy protection, becoming key infrastructure for agent-on-agent commerce.

As related payment protocols and privacy networks mature, autonomous transactions by AI agents will move beyond experimental stages, further reshaping global business operations and driving a massive new economy.

Further reading:
Preparing for the AI agent payment era: Killer crypto applications are already here, just not as you imagine