AI-native commerce: agents that shop, sell, and negotiate

Market map: AI-native commerce

(1) Thesis AI-native commerce is being restructured from human-browsed storefronts into an agent-mediated layer where autonomous software owns purchase intent, seller interaction, and negotiation—compressing search, sales, and contracting into continuous machine-to-machine workflows.

(2) Segments

1. Consumer shopping / buyer agents
Companies: Perplexity (Buy with Perplexity), OpenAI Operator, Google Project Mariner*, Amazon Rufus, Klarna AI Assistant. Dynamics: These agents compete to own demand-side purchase intent and checkout, but they are constrained by retailer APIs, checkout fragmentation, and consumer trust; distribution through existing search/assistant surfaces is the early moat.
2. Seller / revenue agents
Companies: Salesforce Agentforce, HubSpot Breeze, Intercom Fin, Gorgias, Regie.ai. Dynamics: AI agents are being embedded into sales, support, and e-commerce workflows to lower cost-of-sale and enable 24/7 selling, but this segment is crowded and heavily tied to incumbent CRM/helpdesk distribution.
3. Autonomous negotiation & procurement agents
Companies: Pactum, Vendr, Tropic, Zip, Keelvar, Nibble. Dynamics: This segment spans B2B supplier contracts and retail checkout negotiation, where ROI is directly measured in savings or margin; adoption is gated by integration depth, authority to transact, and liability concerns.
4. Agentic commerce infrastructure: identity, payments, execution
Companies: Stripe, Anon, Skyfire, Coinbase AgentKit, Browserbase. Dynamics: This layer provides the rails for agents to authenticate, pay, and execute web/commerce actions; it has tollbooth potential but faces incumbent competition and lacks agent-native trust and dispute standards.

*I am less sure of the current scale/public availability of Google Project Mariner, Skyfire, and Coinbase AgentKit.

(3) Which segment wins the most value and why Autonomous negotiation and procurement agents win the most value near term because they sit on high-dollar B2B spend with clear, measurable ROI—often captured as a percentage of savings or contract value—and enterprise buyers have strong willingness to pay. Infrastructure may become the ultimate tollbooth, but incumbents like Stripe and PayPal have distribution advantages that limit early startup value until agent transaction volume matures.

(4) White-space opportunities

• Agent-native trust, escrow, and dispute resolution for agent-to-agent commerce. Current players handle identity or payments, but no neutral layer verifies agent authority, holds funds in escrow, manages returns/chargebacks, and maintains cross-agent reputation for fully autonomous buyer-seller agent transactions.
• Consumer negotiation broker for regulated/high-friction personal contracts. No AI-native player currently negotiates healthcare bills, insurance renewals, utilities, rent, tuition, or telecom contracts end-to-end with binding authority across portals and phone trees; existing apps like Rocket Money/Truebill are narrow or human-in-loop rather than autonomous multi-term negotiators.

1) Market thesis

AI-native commerce is shifting from human-led search, browsing, selling, and price discovery toward agent-mediated transactions in which software controls intent, product selection, negotiation, checkout, and post-purchase service.

2) Market map

Segment 1: Consumer shopping and buying agents

Companies:

• Amazon — Rufus: conversational product discovery and recommendation inside Amazon.
• Google — AI shopping / Gemini shopping features: AI-assisted product research, comparison, and discovery across the web.
• Perplexity — Shopping: product research and recommendation integrated into an answer engine.
• OpenAI — ChatGPT shopping features: product discovery and comparison within ChatGPT.
• Daydream: AI-native fashion shopping and discovery.
• You.com: AI search and research with shopping-oriented use cases.

Dynamics: These agents compete to own the highest-value layer—commercial intent—before a user reaches a retailer; the likely market structure is winner-take-most around distribution, proprietary user context, and transaction conversion rather than around raw recommendation quality.

---

Segment 2: Merchant-side sales, marketing, and service agents

Companies:

• Shopify — Sidekick: merchant copilot for store operations, merchandising, analytics, and execution.
• Salesforce — Agentforce / Commerce Cloud: customer-service, sales, and commerce agents for enterprise merchants.
• Intercom — Fin: AI customer-service agent used by online businesses.
• Gorgias: AI support and sales automation for ecommerce merchants.
• Klaviyo: AI-assisted marketing, segmentation, and lifecycle-commerce automation.
• Sierra: enterprise conversational agents for customer service and commerce workflows.

Dynamics: Merchant agents are easier to monetize than consumer agents because they produce measurable outcomes—conversion, support-cost reduction, retention, and average order value—but are likely to be embedded into existing commerce, CRM, and marketing platforms.

---

Segment 3: B2B procurement, sourcing, and negotiation agents

Companies:

• Pactum: autonomous negotiation software for procurement and supplier relationships.
• Fairmarkit: tail-spend sourcing and supplier-intelligence platform.
• Arkestro: predictive procurement and negotiation optimization.
• Keelvar: autonomous sourcing and procurement orchestration.
• Coupa: business-spend management with AI-assisted procurement workflows.
• Ivalua: enterprise procurement and supplier-management software with AI capabilities.

Dynamics: This is currently the clearest economic wedge for autonomous negotiation: purchase values are large, savings are quantifiable, and the buyer has a strong incentive to delegate repetitive supplier interactions while retaining approval controls.

---

Segment 4: Agentic payments, identity, trust, and transaction infrastructure

Companies:

• Stripe: payments, checkout, identity, and emerging infrastructure for agent-mediated commerce.
• PayPal: merchant payments and wallet infrastructure, including initiatives aimed at AI-enabled shopping.
• Visa — Intelligent Commerce: infrastructure and controls for AI agents making payments.
• Mastercard — Agent Pay: tokenization, authentication, and payment support for autonomous agents.
• Adyen: enterprise payments, fraud prevention, and merchant acquiring infrastructure.
• Riskified: ecommerce fraud, identity, and transaction-risk decisioning.

Dynamics: Agents cannot scale without delegated authority, authentication, liability allocation, fraud controls, and payment credentials; this layer is likely to be less visible to consumers but strategically important because it determines which agents are allowed to transact and on what terms.

Note: Some Visa, Mastercard, PayPal, and Stripe agentic-commerce initiatives are recent and evolving; product scope and commercial availability may change.

3) Which segment captures the most value?

Likely winner: agentic transaction infrastructure—payments, identity, trust, and permissions

This segment is best positioned to capture the most durable economic value because:

1. It sits across every transaction, regardless of which consumer or merchant agent wins distribution.
2. Trust is a hard bottleneck: merchants need to know who authorized an action, whether an agent is acting within budget, and who bears fraud or fulfillment liability.
3. It can monetize through transaction volume and risk products, rather than relying only on software-seat pricing or advertising.
4. Network effects favor incumbents: payment acceptance, tokenization, fraud data, merchant relationships, and regulatory infrastructure are difficult for a new agent company to replicate.
5. The agent interface may commoditize, while permissioning and settlement remain embedded infrastructure.

The main caveat is that consumer distribution platforms could capture substantial value first by owning shopping intent and steering demand. In the long run, however, the highest-margin and most defensible layer may be the “control plane” that authorizes, authenticates, finances, and settles agent actions.

4) White-space opportunities

A. Cross-merchant agent negotiation and purchasing network

A neutral platform could let a buyer’s agent solicit and compare offers from multiple merchants, negotiate price or terms, manage substitutions, and execute the transaction under explicit constraints such as budget, delivery date, warranty, and sustainability.

• Existing shopping agents mostly recommend or route users.
• Existing procurement tools mostly serve large enterprises and structured sourcing events.
• There is no widely adopted, consumer-scale, merchant-neutral layer for real-time, multi-seller negotiation with enforceable buyer preferences.

Potential business model: transaction fee, supplier lead fee, or buyer subscription—while avoiding opaque sponsored ranking.

B. Agent accountability, permissions, and post-purchase resolution layer

A platform could provide portable agent credentials and policies covering:

• spend limits and approval thresholds;
• proof of user authorization;
• explainable product and price decisions;
• returns, refunds, warranty claims, and disputes;
• audit logs showing what the agent did and why;
• liability allocation among user, agent, merchant, and payment provider.

Payment networks and fraud vendors address pieces of this problem, but no broadly adopted, merchant-neutral standard appears to cover the full lifecycle from agent authorization through fulfillment, dispute, and remediation.

Potential business model: per-transaction trust fee, enterprise compliance software, or embedded protection/insurance.

The two white spaces are best understood as underserved categories rather than literal claims that no startup has attempted them; early and less-publicized efforts may exist.