Written on
Amazon, Walmart, Shopify and Alibaba are deploying shopping agents, and 2025 quietly settled the standards letting them transact at scale. Brands optimised for clicks now must optimise for agents. So who owns the customer when an agent stands between you and them?
We've all been there:
15 tabs open. Comparing prices across sites. Reading reviews. Checking availability. Coordinating delivery windows. It's exhausting. What if you could delegate all of that to an assistant who knows your preferences and your needs, but never gets tired?
That solution exists and it's called agentic commerce.
Shopping is shifting from "I need to buy X" to "I need to solve Y":
Not "find me a hotel", but "plan my anniversary trip"
Not "buy furniture", but "furnish my apartment within budget"
Not "find a jumper", but "my style and needs for autumn"
It has already started. Amazon ('Buy for me'), Walmart (Sparky), Shopify, Alibaba and Instacart are already deploying agent shopping — not “experimenting,” but shifting the shopping interface. McKinsey, BCG, Bain, and Accenture all say agentic commerce will reshape retail quickly, likely faster than mobile did.
In 2025 we got generally recognised standards
MCP — so agents can easily use any tool, eg a retailer's stock or price list
A2A — so agents can easily collaborate with other agents, Google's vision (see below)
ACP/AP2 — so agents can complete payments
Visa and Mastercard have agent payment frameworks, preparing for AI buyers
In 2026 we will see roll outs using these standards. In fact, we can already shop from within ChatGPT. Brands once optimized for clicks. Now they must optimize for AI agents. Retailers risk losing the customer interface to AI platforms.
Google's ADK (agent development kit) launched in May 2025 with their vision; whereby a user submits a request, Google then orchestrates a team of agents to resolve that query. The orchestrator builds a team from retailer's agents, to see which delivers the best solution. The user gets a shortlist of options. Wonderful for the user and Google, but the vendors risk becoming hidden utilities.
Companies spent decades optimizing for "customers who click.". Now they must optimize for "agents who orchestrate.". This isn't just a new channel. It's a fundamental restructuring of how products are discovered, decisions are made, and customer relationships are formed. And it's happening faster than the e-commerce revolution because agents can "ride the rails" of existing infrastructure.
Natural language search across platforms
24/7 instant customer support
Synthesize reviews and recommendations (with supervision)
Fully autonomous complex purchases without oversight
Real negotiation and deal execution
Consistently explaining "why I recommended this"
Capture intent BEFORE comparison shopping begins
New revenue models: negotiation fees, agent subscriptions, data monetization
New marketplaces: agents can transact in different ways to people
Ad revenue decline as agents bypass traditional channels
Disintermediation: agents unbundle vertical platforms
Trust violations could poison the entire ecosystem
Agentic commerce is a sufficiently large shift that we can talk about entirely new business models. We've done a couple of blogs on how agents create new business models, here's what you need to know for agentic commerce:
API-as-Storefront
ChatGPT 'apps' and Claude 'connectors' allow these agents to integrate with business services via MCP, so business are rushing to build MCP's, to deliver their services to customers
For buyers: Faster than web browsing
For vendors: Lower infrastructure costs
Risk: Vendors must maintain human UX too. Services for agents only are not far away
New Types of Marketplace
Existing markets are often unsatisfactory to many participants. Think of all the wasted searching, matching and regret that happens on ebay, and hinge!
Agents find matches between market participants much more effectively than people
These superior matching algorithms, eg 'Gale-Shapley', are already used by agentic pioneers in recruitment: JackAndJill.ai who just won their $20m seed investment.
Such new and more effective markets may cannibalise existing markets, stranded assets
Agent-as-Customer
Y-Combinator recently closed a round of start-up investment where funds were focussed on businesses born to serve AI agents, not people, agents are the customers. SEO organisations have already realised the importance of this as search services go to AI.
For buyers: Agent shops autonomously
For vendors: Must serve agents as primary interface
Risk: Trust infrastructure is critical
Expertise at Scale
Unlike people, AI agents can be trained with multiple areas of expertise on the items be searched or purchased
For buyers: Expert-level guidance for everyone
For vendors: Premium agent subscriptions
Risk: Quality control, liability questions
Agents Solving Problems, not Just Purchasing
Most importantly, agents can work in teams or create teams on the fly, dedicated to the immediate task. e.g. a used car specialist and an insurance agent, to research the best combination of options for you. Especially powerful in the traditional 'agencies' with complex unstructured data to sift and arrange; insurance agent, estate agency, travel agent etc.
For buyers: Set goals, eg get a deal on a complementary set of autumn clothes, then agent executes
For vendors: Capture buyer's intent early
Risk: Agent errors cascade when they are multiple components to a purchase
Tireless Negotiator
The situation may quickly arise where both buyer and seller employ AI agents. We then have potential for price discovery via negotiation. Of course, negotiations take time, but relative to the pricing advantage this time is unimportant to agents
For buyers: Best deals automatically
For vendors: New revenue from negotiation fees
Risk: Agent manipulation, fairness concerns
AI agents are, unsurprisingly, most effective in agent businesses: travel agent, insurance agent, employment agency etc. Agents always existed to solve friction in these sectors where:
Information was fragmented
Paperwork was non-standard and painful
Negotiation and coordination required a human
Trust and relationships mattered
AI collapses 1-3. The model would appear to be a human to manage relationships and build trust, with AI to execute tasks.
You can't win at agentic commerce without solving trust. When agents make mistakes, who's liable? The platform? The brand? The user? Right now, there's no consensus. When an agent "negotiates" a deal, how do we ensure it's acting in the customer's interest and not skewed by hidden incentives?
This is where responsible AI becomes competitive advantage. Businesses need:
Explainability (agents must explain their choices)
Transparency (clear disclosure of how agents are compensated)
Control (users need override capabilities)
Accountability (clear frameworks for when things go wrong)
Auditability (cryptographically signed transaction trails)
The companies that build trust infrastructure—not just agent capabilities—will own this market.
Agentic commerce is coming whether we're ready or not.
---
What we say:
10 Paths to Win with Agentic Commerce
How to Reinvent Markets with Agents
Google's Plan to Replace Your Customers
What the traditional consultancies say:
McKinsey — “The Agentic Commerce Opportunity: How AI Agents Are Ushering in a New Era for Consumers and Merchants” (Oct 2025)
Accenture — “Agentic payments in commerce — the future is here” (Sept 2025)
BCG — “Agentic Commerce is Redefining Retail — Here’s How to Respond” (Oct 2025)
10 Paths to Win in Agentic Commerce
Gartner says a third of enterprise software purchases will involve an AI agent by 2028, and machines are assumed immune to persuasion. We ran 8,000 trials across five frontier models. Which techniques work, which backfire, and what does that mean for selling to agents?
Once agents start finding, negotiating and buying on our behalf, they become participants and the economy shifts. Drawing on MIT and Harvard research, this maps ten templates where new players and markets are emerging. The harder question: which is coming for your business first?
A new MIT and Harvard report reaches back to Coase's 1937 theory of the firm and Gale-Shapley matching to argue the real prize isn't automating old processes. The boundaries between firms and markets start to move. So where does that leave your competitive advantage?