methodicaltrades
← All insights
Weekly newsletter

When AI Gets a Wallet

Agentic commerce could change who controls discovery, payment and customer trust. The investment question is who owns the permission to act.

Sent 26 September 2026●5 min read
An AI shopping assistant connects a customer, merchant and payment network

The handoff that matters

AI already helps shoppers compare products. The harder step is letting it complete a purchase. That turns a recommendation engine into an agent with authority over money, identity and returns.

The decisive advantage may belong to whoever makes a machine-initiated transaction safe enough for consumers and merchants to trust.

Goldman Sachs estimated a $2.6 trillion high-likelihood US spending pool potentially addressable by agentic commerce. That is an estimate of eligible spending, not a forecast of sales captured by AI agents. The gap between those two figures is the central investment question.

A purchase needs a mandate

A useful shopping agent needs explicit limits: what the customer wants, how much it may spend, which merchants it may use and what happens if the product is wrong. Without those rules, a smooth checkout can produce an expensive dispute.

IntentCan the system distinguish a firm instruction from a casual question?
ScopeDoes the customer approve a product, a price ceiling and acceptable substitutions?
IdentityCan the merchant identify the agent and verify its authority?
RecourseWho handles refunds, fraud, mistakes and delivery failures?

Visa’s Trusted Agent Protocol is one example of the infrastructure being built to identify agents and convey purchase context to merchants. Standards can lower friction, but they do not solve every question of consent, liability or customer service.

Who controls the customer?

Agentic shopping can shift traffic away from a retailer’s website. That makes product data, inventory accuracy and access to checkout more valuable. It also threatens a familiar advertising model: a machine comparing offers has less reason to admire a sponsored banner.

Retailers still own useful assets. First-party purchase history, fulfilment, loyalty benefits and returns can help them keep the customer relationship, even if discovery starts inside a general-purpose assistant. Payment networks and platforms may capture value by authenticating the agent and routing the transaction. The winners need not be the makers of the largest models.

There is a second-order risk: if an agent optimises only the visible price, merchants may respond with fees, poorer service or restricted access. The best agent must optimise for the whole transaction, including fit, availability, delivery and recourse.

Signals that adoption is real

Watch for repeat purchases completed without manual checkout, measurable reductions in abandoned carts and dispute rates that remain manageable. Merchant acceptance matters more than product demos. So does the share of transactions where consumers set durable permissions instead of approving every step.

The near-term race is to become the trusted permission layer between a shopper’s intent and a merchant’s fulfilment. The spending opportunity is large; the amount that becomes agent revenue is still unknown.

Sources and method

  1. Goldman Sachs, Americas Technology: Entering the Agentic AI Era for the Commerce Landscape, 24 September 2026 (user-supplied research PDF; spending estimate and adoption discussion).
  2. Visa Developer, Trusted Agent Protocol overview
  3. Visa Developer, protocol specifications

This article provides general information only and is not personal financial advice. It does not consider your objectives, financial situation or needs. Nothing here is an offer or recommendation to buy or sell a financial product. Investing involves risk, including possible loss of capital.

Explore more weekly newsletters.

Browse all insights →