Machines of Buying and Selling Grace - Adam Behrens, New Generation

Key Concepts

• AI Commerce
• Merchant Agents
• Consumer Agents
• Intent Infrastructure
• Agentic Transactions
• Delegated Authentication
• Buyer Intent
• Seller Intent
• Unified Product API
• Preference Management
• AI-driven Decision Making
• Generative Interfaces

What is a Store?

• Traditional Store: Described as a physical location where inventory was kept in the back, requiring interaction with a clerk to retrieve items.
• Evolution: The emergence of big-box retailers like Walmart and Costco in the 1950s and 60s, enabled by information systems, shifted inventory to the front, introducing the concept of browsing.
• E-commerce: The internet digitized the store, scaling merchandise and distribution globally, available 24/7.
• Sacrifice: This shift resulted in a "sea of sameness," making it difficult to differentiate between brands.
• Definition: A store is defined as "a location for and a protocol that facilitates transactions" between merchants and buyers, supported by a system for interaction.

The Store with AI: Digitizing Participants and Interactions

• E-commerce vs. AI Commerce: E-commerce digitized merchandise and distribution, while AI digitizes the participants (merchants and consumers) and their interactions.
• Transformation: Static websites evolve into merchant agents, consumers browsing become consumer agents, and low-level payment infrastructure transforms into higher-level intent infrastructure.
• Goal: The fundamental goal remains the same: to facilitate transactions.
• Qualitative Shift: A new generation of consumers (both human and agentic) will interact with dynamic, real-time, and generative interfaces built on new infrastructure.
• Two Possible Futures:
  • AI Agents on Websites: AI agents navigate websites optimized for agent interaction, expressing intent, dynamically filtering product catalogs, and completing checkout processes. Example: Using ChatGPT to find a TV.
  • Programmatic Access: Agents use APIs to programmatically access merchants, enabling reasoning over API calls and generating UI elements on the fly.

From Static to Agentic Interactions: The Path Forward

• Goal: High-quality conversion with satisfied users and minimal returns.
• Payment Intent: In code, a payment is represented by a payment intent, which undergoes transformations during checkout.
• Challenge 1: Software Clicking the "Buy" Button: Current e-commerce websites often error out when software attempts to click the "buy" button.
  • Solution 1 (Current): Stripe SDK solution where the software provider (e.g., ChatGPT) spins up a virtual card to make the purchase on behalf of the user.
  • Solution 2 (Preferred): Delegated authentication (Visa) allows the agent to use the user's actual credit card for checkout.
• Challenge 2: Fuzzy Intent to SKU-Level Item: Converting vague user intent (e.g., "I want running shoes") to a specific product.
  • Current Solution: Requiring users to provide a product detail page URL.
  • Emerging Trend: AI channel users exhibit higher conversion rates, dollar values, and lifetime value, potentially justifying relaxed fulfillment costs.
• Challenge 3: Product Availability Across Stores: Determining if a specific item is available across thousands of stores.
  • Solution 1 (Suboptimal): Using existing product feed infrastructure (e.g., Google), requiring individual integration with each merchant.
  • Solution 2 (Suboptimal): Scraping product data from websites, leading to repetitive work and bot traffic.
  • Solution 3 (Preferred): A unified API to access product data across all merchants (like Plaid for product data).
• Buyer and Seller Preferences:
  • Today: One-sided, narrow, siloed user accounts, limited LLM memory, and limited business intent sharing.
  • Future: Two-sided, expansive, rich user context, and real-time strategic goal expression from businesses.
  • Challenges: Complex, changing, and conflicting preferences, and disincentives for honest reporting.
  • Current Solutions: Naive trust in provided information, vulnerability to prompt injection.
  • Proposed Solution: Third-party institutions and market makers (similar to finance) to manage differences between buyers and sellers.

The Frontier: Adding Intelligence

• Need: Intelligence for consumers and merchants to reason over and negotiate intents and preferences.
• Infrastructure: Moving from market making to coordination and reasoning over participants.
• Logic: Generating real-time interfaces tailored to each participant's needs.

Real-World Application: Samsung Case Study

• Fortune 500 Companies: Despite their size, these companies are forward-thinking due to their need to adapt to technological and societal shifts.
• Samsung's Evolution: From a fish merchant to a technology giant, Samsung is exploring how its brand evolves in the AI era and how to bridge e-commerce to the agentic future.
• Steps:
  1. API and MCP Server: Creating an API for any chat client to use, abstracting complex product systems into a consistent API.
  2. Data Connection: Connecting product data with other data sources (e.g., brand and design systems) to construct seller intent.
  3. Experimentation Container: Creating an AI subdomain for rapid experimentation of generative interfaces that ingest product and brand data.
  4. Agentic Transactions: Enabling payment flows on new surfaces for bot traffic.
• Benefits: Higher intent, deeper funnel, and better conversion rates from AI chat users.
• Recommendation: Every retail brand and merchant should adopt this posture.

Questions and Answers

• Machine Customers: Projection on when machine customers will be seen daily.
  • Answer: It's happening fast, with products like ChatGPT starting to bring shopping experiences. The missing piece is the full journey within the application. Brands want to own a surface in this new world, starting with a web-like surface that can be transported into chat products.
• Payment Mechanisms: Are credit cards the right payment mechanism for the agentic economy?
  • Answer: Conceptually, stable coins and crypto are a strong argument as the native payment rail. Practically, consumers use credit cards, making them the most likely bridge. A third alternative is the agent owning a perpetual credit card.
• Super Apps: Will Claude and ChatGPT try to become super apps?
  • Answer: That's their goal. The question is how merchants have control in that environment. Model providers are open to getting the user to the right outcome, creating alignment between merchant and chat product goals.
• Revenue Share: Will there be revenue sharing?
  • Answer: Definitely. It will likely be affiliate revenue or a portion given back to merchants for providing high-quality data.

Conclusion

The future of commerce is moving towards agentic interactions, where AI agents represent both buyers and sellers. This requires new infrastructure, including unified product APIs, delegated authentication, and mechanisms for managing buyer and seller preferences. While challenges exist, early adopters like Samsung are already experimenting with AI-driven interfaces and seeing promising results. The key is to start with the "what" and "how," recognizing that stores will evolve rather than disappear, returning to their original form as conversations.