Seven theses for commerce with agents

Where the theses come from

I have been writing about technology and commerce since 2023. “The Yin and Yang of Commerce” explored logic and creativity. My 2025 essays examined the shrinking distance between intent and purchase; the January 2026 strategy book focused on protocols, trust and experience. By September 2026, one question had become central: what needs to happen beneath the interface for an agent to buy reliably? [S10, S13, S03, S24]

These seven theses connect the book. Each offers something you can test in your own business. The table at the end offers practical signals to watch.

1 · Machine readability is the entry ticket, not the victory

An offer an agent cannot understand does not exist for it. Merchants who bury dimensions, compatibility, delivery terms and prices in images, PDFs or prose drop out before selection begins.

The second half is easily missed. Readability gets an offer onto the longlist. Relevance, availability, price, customer preferences and the agent operator’s rules get it onto the shortlist. A spare-parts merchant with perfectly structured data still loses if delivery takes three weeks.

Treat readability as basic hygiene: you cannot compete without it, but it will not win the sale by itself.

2 · Brands become criteria, while remaining a feeling

The fear that agents will flatten brands underestimates customers. Someone who says “only from this manufacturer” or “prefer repairable, even if it costs more” gives the agent a preference shaped by the brand. The agent carries out a preference formed much earlier: in a shop, through a recommendation or in use.

Brands face two tasks. Every promise needs a verifiable form: “durable” becomes “a ten-year warranty, with spare parts until 2036”. The places where preference forms also become more valuable. A strong brand in the agent era can be felt emotionally and justified factually.

3 · Gatekeepers change shape

Open interfaces lower the cost of access. They do not, by themselves, redistribute power over that access. Whoever controls the preferred assistant, product index, identity or payment relationship stands at a new gate. In the last decade, gatekeepers were search engines and marketplaces; in the next, they may be assistants, wallets or procurement networks.

Merchants should build several routes to demand: use open protocols, maintain multiple channels and make the direct relationship strong enough that customers explicitly send their agents to them. Chapter 19 shows how to prepare a second route in practice.

4 · Trust needs evidence rather than a universal score

A single trust score is convenient and usually wrong. An agent buying for a hospital weighs delivery reliability differently from one buying a gift. Useful signals have a source and a period: delivery performance over the past 90 days, common return reasons or documented repair times.

Merchants should maintain and publish these signals themselves rather than wait for a third party’s score. A number without a verifiable basis is decoration with decimal places.

5 · As routine shrinks, experience matters more

Agents take over work people rarely enjoy: reordering printer paper, comparing tariffs or redirecting deliveries. That makes the time customers willingly spend with a brand more valuable: trying an instrument, planning a kitchen or browsing a collection.

The experience moves too. It unfolds in an assistant’s conversation, a browser agent or an interface the merchant does not control. How product knowledge and brand expression reach those places is one of the most interesting open questions in this field; we explore it through the AXP prototype in Chapter 6. My test for a good experience is simple: it should make a decision safer, faster or more enjoyable. [S08, S09, S12]

6 · The shop becomes an execution layer

Agents become the purchasing interface; the shop beneath becomes the execution layer. A conversation can prepare a request. Quantity, price, tax and order become binding in the commerce core, where business rules apply and someone takes responsibility for them. [S22, S24]

Protocols such as UCP, ACP and MCP are the languages between these worlds. They are essential, but do not provide the grammar underneath: who may grant a discount, when an offer is valid or what happens after a timeout. “We support UCP” is therefore only half an answer. The other half is the version, supported capabilities and purchasing journeys actually tested.

7 · Openness makes individuality affordable

For two decades, commerce followed a simple equation: standard was cheap and distinctive was expensive. Many good merchant ideas failed because implementation cost too much. Coding agents such as Codex and Claude Code change that equation. When merchants or agencies can put an agent to work on open code, “that would be nice” can become a functioning prototype within days.

Open interfaces also allow changes on closed platforms. An open core goes further: agents can read the relationships, draft changes and let people examine them through to execution. I therefore expect open source to grow in importance during the agent era. Agencies’ role will change fundamentally, with their value shifting from selling implementation hours towards architecture, acceptance and operations. Part V shows how merchants already use this freedom.

Seven theses in daily business

A thesis becomes useful when it changes a decision. Each of the seven can be followed through specific observations in the business. The last column describes developments that would weaken the relevant assumption.

A spare-parts merchant may follow readability and execution first; an interiors business may focus on brand and experience. The next chapter follows our traveller again: what does she do herself, what does her agent handle, and where does the merchant come in?