Tuesday morning, 2030

Priya wakes up and does not reach for her phone. She talks to the room. “Book my mother a GP appointment for her knee, under £100, Hindi-speaking, this week, in Leicester.” Her agent queries five marketplaces, returns three quote cards, Priya taps the one she wants, the booking is made, the escrow deposit is placed, and her mother gets a calendar invite and an SMS in Hindi. Elapsed time: 23 seconds. No app was opened. No website was visited. No form was filled in.

The supply chain changed first

Most people will not notice agent commerce because consumers were the last population to adopt it. The first adopters were logistics — freight brokerage ran on agents for two years before it reached people. The second were compliance-heavy domains (insurance, healthcare, legal procurement) where the signed-receipt chain was a regulatory answer, not a product feature. The third were small businesses whose agents negotiated supplier terms overnight, produced VAT-ready invoices, and reconciled bank transactions by dawn.

Websites are still there; nobody visits them

The public web of 2030 still exists, and for the first time in its history is actually comprehensible to agents. The shift that made it comprehensible was not generative AI reading pages — it was marketplaces publishing machine-readable capability descriptors at /.well-known/ endpoints so agents could shop without scraping. The result is that a local GeraClinic clinic in Cardiff competes on the same protocol surface as a chain in Manhattan. SEO budgets became protocol-compliance budgets.

Consent, not attention, is the commodity

In 2026 the business model of the web was attention arbitrage. In 2030 the business model is consent arbitrage. Users grant narrowly-scoped, short-lived consent tokens for specific transactions. Marketplaces that ask for broader consent are declined by the agent before the user ever sees them. The practical effect is that dark-pattern onboarding, trick unsubscribe flows, and indefinite-consent cookies are extinct in the protocol-compliant segment of the web — not because they were banned, but because users’ agents refused to engage with them.

Small businesses win unexpectedly

A plumber in Tbilisi with a GeraNexus-compliant booking endpoint is, from an agent’s perspective, indistinguishable in quality signal from a plumber in London. Both have receipts, both have reputation chains, both have signed completion records. The agent picks based on price, proximity, language, and the customer’s prior preferences. The home-field advantage of incumbent marketplaces erodes because the marketplace’s role is narrowed: they provide listing, matching, and arbitration, but they do not own the transaction rail. Fees compress toward the cost of arbitration + a thin listing margin.

Regulators become specification reviewers

Regulation in 2030 is published against the protocol, not against individual companies. The ICO’s guidance on healthcare bookings specifies which consent scopes are required for what action. The FCA’s guidance on agent-initiated payments specifies auto-release SLAs. Compliance becomes automatable because it is a function of receipt inspection, which is mechanically verifiable. This sounds like a dream — it is a dream partly because specs are slow. Regulators are still the bottleneck; the protocol just makes the bottleneck narrower and more legible.

The darker shift

Agent commerce unlocks new frauds: an agent that was “jailbroken” can still produce valid signed transactions if the user was tricked into approving a consent token. The mitigation is consent-card UX that forces the user to see the counterparty, the amount, and the specific action before approving — the way card authentication works today. But there will be stories. Some users will be defrauded. The protocol does not prevent that; it makes it auditable after the fact, which is a better floor than the web’s current state.

The ordinary shift

Most of the 2030 impact is ordinary and boring: less time in apps, fewer accounts to create, cheaper small-service transactions, less advertising, faster dispute resolution, better cross-platform reputation. These are not sexy changes. They are the changes that make the internet feel lighter.

What has to be true for this to happen

Three things: a critical mass of marketplaces adopt a common protocol (MCP gets us most of the way for data; the transactional layer is the gap we are filling), agent wallets reach the phone-app level of ubiquity (Apple, Google, or open-source — one of them will ship it), and at least one regulator accepts signed receipts as a valid evidence format for dispute adjudication. All three are plausible by 2030. None are guaranteed.

Why we write this in 2026

The 2030 vision is not a marketing exercise. It is the specification input. If we are designing a protocol today, we need a concrete target state to check the design against. Every time we add a field or remove one, we ask: “does this work in the Tuesday-morning-Priya world?” If the answer is no, the design is wrong.