Pricing in ecommerce has always been a monologue. Retailers set prices, customers accept or leave. That model is ending. AI agents on both sides of a transaction now negotiate the final price and promotional terms before a human sees the offer.
The implications for margin strategy, promotional planning, and platform architecture are significant enough that treating this as a future concern puts you behind retailers who are already deploying and developing retail ai agents.
The Dual-Agent Architecture
The architecture driving this shift splits into two distinct roles. Buyer agents operate on behalf of consumers: they interpret purchase intent, set budget parameters, compare options across merchants, and negotiate terms without the customer doing any of it manually.
Seller agents operate on behalf of retailers, optimizing pricing and promotions in real time before responding to buyer agents with the most competitive offer the retailer’s parameters allow.
Researchers have described this as “a fundamentally different market dynamic than traditional e-commerce,” one in which agent-to-agent transactions happen largely independently of direct human involvement.
By the time a customer reviews a product recommendation, the price and any promotional offer have already been negotiated between machines.
The supporting infrastructure is live. OpenAI and Stripe launched the Agentic Commerce Protocol in September 2025 to connect buyer agents to merchant backends. Stripe powers 78% of the Forbes AI 50 and launched an Agent Toolkit in 2024 enabling agents to process payments and create virtual cards.
Visa launched “Intelligent Commerce” in the same year – infrastructure specifically designed for AI agents to transact securely. These are not pilots. They are the payment and communication rails of a new commerce layer.
Procurement: What Walmart Proved at Scale
The clearest proof that AI negotiation agents deliver measurable outcomes comes from procurement, not the consumer side. In 2021, Walmart deployed Pactum AI’s autonomous negotiation platform to handle supplier contract renegotiations with tail-end vendors – suppliers on default, unnegotiated terms that human buyers couldn’t cost-effectively reach.
The platform closed agreements with 68% of suppliers, achieved an average 3% savings per contract, and compressed negotiation timelines from weeks to an average of 11 days. Payment terms extended by 35 days.
Two outcomes stand out. First, 75% of suppliers preferred negotiating with the AI rather than a human, citing consistency and transparency. Second, Pactum’s agents conducted negotiations with up to 2,000 suppliers simultaneously – a throughput no procurement team could match. Walmart has since expanded the program to mid-tier vendors and route-rate transportation negotiations.
Across its client base, Pactum reports an average 4.2% increase in profitability. The mechanism is multi-dimensional negotiation: rather than optimizing a single variable, the agent evaluates thousands of term combinations – pricing, payment schedules, volume commitments, service levels – and identifies proposals that create genuine value for both parties. A human negotiator working through a spreadsheet simply doesn’t have that evaluative range.
Customer-Facing Negotiation: A Different Problem
Procurement is a controlled environment. Both parties are sophisticated, terms are bounded, and the scope is defined in advance. Consumer-facing negotiation is harder because the parameters are wider, negotiation failures are visible, and outcomes affect brand perception.
Google Cloud describes the consumer-facing version in concrete terms. A customer specifies requirements in natural language – a jacket that is warm when windy but breathable in the sun, packable, with pockets, under $100.
A buyer agent identifies options across retailers while simultaneously negotiating with brand-side seller agents for personalized promotions. Mr. Brown, Global Managing Director at Google Cloud, summarized the trajectory in early 2026: “We’re moving from a one-to-many pricing strategy to a dynamic, one-to-one model where the negotiation happens between two machines before a human even sees the final offer.”
McKinsey projects the global agentic commerce market at $3 to $5 trillion by 2030. Morgan Stanley estimates the US market alone at $190 to $385 billion over the same period.
Adobe data cited by BCG found that traffic to US retail sites from AI browsers and chat interfaces grew 4,700% year-over-year in July 2025. These numbers reflect a market shift in progress, not a speculative future.
For promotional strategy, the practical consequence is direct. A seller agent doesn’t broadcast a promotion to a segment. It calibrates an offer against a specific buyer agent’s parameters – budget, delivery preference, purchase history.
A price-sensitive buyer with a $100 ceiling receives a different offer than one with looser constraints and a preference for next-day delivery. Promotional mechanics become individual rather than cohort-based, which changes how promotional budgets are allocated and measured.
Three Risks Retailers Are Underestimating
Regulatory exposure. Algorithmic pricing is under active oversight in both the US and EU. The European Commission launched a public consultation under the Digital Fairness Act in July 2025, specifically identifying dynamic pricing as requiring stronger consumer protections.
In the US, the FTC launched a Section 6(b) investigation into surveillance pricing in July 2024, examining how behavioral data influences prices. New York’s S 3008, effective July 2025, requires businesses to disclose when algorithmic systems use personal data to set prices. Any retailer deploying AI pricing agents without legal review of these frameworks is carrying compliance exposure.
Model capability disparity. Research published in 2025 benchmarking agent-to-agent negotiations found that more capable AI models consistently outperform weaker ones – earning higher profits as sellers and securing better terms as buyers.
If your seller agent runs on a less capable model than the buyer agents your customers deploy, your negotiation outcomes will be systematically worse. Model selection and fine-tuning become commercial decisions, not IT ones.
Brand disintermediation. BCG warns that without deliberate intervention, retailers risk becoming background utilities in agent-controlled marketplaces. Customers arriving through AI agents are 10% more engaged than traditional visitors but arrive with weaker brand loyalty and stronger intent to transact on price and specification.
Promotional investment that historically built brand preference needs to either influence buyer agent behavior directly or reach consumers before they delegate shopping decisions to an agent at all.
The Infrastructure This Requires
Running effective seller agents requires more than a pricing model. The agent needs live access to inventory data, competitor pricing signals, margin floors by SKU, customer lifetime value scores, and promotional budget availability – simultaneously, with latency low enough to respond during an active buyer agent session. That is a data integration problem before it is an AI problem.
Retailers who haven’t rationalized their commerce infrastructure are at a structural disadvantage. Effective retail AI agent development depends on a clean, integrated data layer that most mid-market retailers haven’t yet built. An agent is only as effective as the context it can access at the moment of negotiation. If your inventory, pricing, and CRM systems don’t exchange data in real time, the agent is making decisions with incomplete information and losing deals it could have won.
The same constraint applies to the commerce platform itself. Seller agents need to write back to cart, pricing, and promotion systems in real time. Retrofitting legacy platforms for bidirectional, agent-to-agent interaction is not a configuration change — it is an architectural rebuild. Organizations investing in ecommerce software development services for their next-generation platforms need to treat agent-readiness as a design requirement from the start, not a feature to be added in a later phase.
Procurement-side AI negotiation is already delivering measurable returns at Fortune 500 scale. Consumer-facing agent-to-agent negotiation is commercially operational in early deployments and, by McKinsey and BCG’s timelines, mainstream within two to three years. Retailers building the data infrastructure, the platform architecture, and the seller agent capability now will be in a materially stronger position than those waiting for the market to force the investment.
