A Buyer's Agent Should Negotiate

[c-e-daly]

The traditional commerce model is I list a price, you pay it or I give you a discount. There is well over $100 billion in commerce done annually on platforms like eBay, Poshmark, etc. that enable CONSUMERS to make an offer.

A Buyer's Agent represents the interests of the buyer, not the seller, and that should include the ability to negotiate terms, pricing, shipping/delivery, etc. Part of UCP should be scaffolding the ability of a seller to accept customer or agent generated offers. Whether that is executed through discounts, price markdowns on shipping (eg. $5 flat shipping in the US) or some other value added service would be up to the seller to decide. UCP should surface the ability of a shop to accept and process offers and potentially provide the agent the ability to receive a counter offer and negotiate by giving the agent access points and payload schemas to effect that negotiation.

[Nolpak14]

Great discussion point! The ability for a buyer's agent to negotiate is indeed a powerful extension of traditional commerce, and UCP's extensibility model provides an excellent foundation for this.

From a technical implementation perspective, a "negotiation capability" could be structured as a UCP extension augmenting the dev.ucp.shopping.checkout capability. The extension would need to define:

1. Discovery: A merchant could advertise support for negotiation via an extension in their /.well-known/ucp manifest:

   {
     "extensions": [
       {
         "name": "dev.ucp.shopping.negotiation",
         "version": "1.0.0",
         "parentCapability": "dev.ucp.shopping.checkout"
       }
     ]
   }

2. Negotiation Flow: Similar to how UCP's discount extension works, but bidirectional:

   • Agent initiates with POST /checkout/negotiation with proposed terms (price, shipping, etc.)
   • Merchant responds with counter-offer, accept, or reject
   • State machine: pending -> countered -> accepted/rejected

3. Payload Schema: The extension would need to define negotiation parameters (price adjustments, shipping terms, delivery dates, bundling options) and validation rules.

Challenges to consider:

• Real-time vs async: eBay-style negotiations happen asynchronously over days, while AI agents might expect near-real-time responses
• Escalation to human: When should a negotiation be escalated from automated rules to a human merchant?
• Trust and reputation: How do agents build negotiation credibility? (e.g., successful purchase history)

Implementation approach: For anyone building buyer agents, the Agorio SDK (an open-source TypeScript SDK for UCP/ACP) could be extended to support negotiation. Currently, Agorio handles capability discovery and extension composition - adding negotiation would involve extending the ShoppingAgent with negotiation tools and integrating with the LLM to formulate offers.

The key is ensuring the negotiation protocol remains merchant-friendly. Not all merchants want to negotiate, but for those who do (marketplaces, B2B, high-value items), this could be a significant conversion booster.

Would love to see a working group form around this extension - I'd be happy to contribute implementation insights from the Agorio perspective.

[c-e-daly]

This is correct:

Real-time vs async: eBay-style negotiations happen asynchronously over days, while AI agents might expect near-real-time responses
Escalation to human: When should a negotiation be escalated from automated rules to a human merchant?
Trust and reputation: How do agents build negotiation credibility? (e.g., successful purchase history)

These are the components that will take the most consideration. Agentic Commerce being more automated and built around discovery, but for a true buyer's agent, negotiation is a key point for them and could be a key differentiator for sellers as well.

There are make offer tools out there already. Most of them are a form fill and require a review and then they programmatically append a discount and checkout url for the customer (or agent). If there is a pause, that return path is more difficult to navigate because the agent is likely no longer in session. Those payloads also tend to have a customer email, phone, etc.

My initial thinking on UCP for being able to negotiate was literally the ability for the shop to say "I accept offers" in their UCP manifest and then leave it up to the shop to decide how to fulfill it. The shop can provide through their MCP their capabilities and routes, payloads, etc, for the agent to access, submit and receive responses. That way the agent can decide which shops handle the negotiation the best.

Not just B2C either - you could do B2B and have a different UCP for commercial negotiation. And platforms are pushing hard into B2B digital commerce right now.