Google AP2: The Agent Payments Protocol Freelancers Should Know

The core primitive: Mandates

AP2 introduces a primitive called a Mandate: a tamper-proof, cryptographically signed digital artifact that records the user's authorisation of an agent's action. There are three flavours:

• Intent Mandate — captures the initial user request ("find me a French copywriter for SaaS landing pages, max $2,000 budget") and the auditable context the agent will operate inside.
• Cart Mandate — signed after the agent presents a proposed cart, creating an unchangeable record of exact items and price the user has approved.
• Delegated Task Mandate — for autonomous purchases ("book the deposition transcript when it goes live, up to $500"), the user-signed rules-of-engagement: limits, timing, conditions.

The whole stack creates what Google describes as "a non-repudiable audit trail establishing authorisation, authenticity, and accountability." In plain English: the merchant or service provider can prove the customer actually intended this purchase, the agent was authorised to make it, and the cart was what the human signed off on. Disputes get cheaper. Chargebacks get rarer. Trust gets verifiable.

Why a freelancer should care

Freelancers sell to humans today. Within 18-24 months, a non-trivial slice of inbound discovery and booking will route through AI agents acting on behalf of those humans. That is not a hypothesis — it is what Stripe's Link agent wallet, Anthropic's Computer Use, OpenAI's ChatGPT agent, and Google's Gemini Intelligence all explicitly plan for.

When a client's AI agent goes shopping for a Webflow developer, a paralegal, a transcriptionist, or a fractional CFO, the agent will look at three things:

1. Is there a price the agent can quote? (Productised services and listed packages win; "let's hop on a call" loses.)
2. Is there a checkout the agent can transact against? (Branded portal with one-click pay wins; "send me an invoice via email" loses.)
3. Does the checkout accept AP2 Mandates or equivalent agent credentials? (Agent-aware payment surfaces win; legacy redirect flows that require a human to click through 3DS in a browser lose.)

The first two are already obvious to most freelancers. The third is new. AP2 is the protocol that turns the third from a technical curiosity into a competitive advantage.

Stripe, Mastercard, Visa — how the rails connect

Stripe announced at Sessions 2026 that Link's agent wallet lets agents pay on a user's behalf with one-time issued cards per task and per-payment user approvals (Stripe, Everything we announced at Sessions 2026). The mechanism is compatible with AP2 Mandates: the user-signed intent (Intent Mandate) maps to the spending-policy approval, and the per-task one-time card is the issuance event that satisfies the Cart Mandate.

Visa and Mastercard have both shipped their own agentic-commerce surfaces — Visa Intelligent Commerce and Mastercard Agent Pay — and both have aligned on AP2-style verifiable credentials for representing user authorisation across processors (Digital Commerce 360, Visa and Mastercard both launch new agentic AI payments tools, October 2025).

The convergence is what matters: by late 2026, the question for a freelancer's booking surface stops being "which rail" and starts being "do you support the agent authorisation envelope at all?"

What freelancers should actually do

Three concrete moves.

1. Productise pricing now. If an agent cannot read your rate, it cannot quote you, and it cannot book you. A Notion page with three productised packages at fixed prices is more agent-readable than a Calendly link to "discuss scope."

2. Run checkout on a real surface, not over email. A branded portal with Stripe, PayPal, or another card-rail processor — with structured invoice data exposed to the buyer's stack — is what an agent can verify and transact against. PDF invoices attached to emails are a wall.

3. Add structured data to your service pages. Schema.org Service, Offer, and PriceSpecification markup. AI agents reading your site for a client are going to lean on structured data the same way Google's crawler does. The freelancers who have it indexed in 2026 are the ones agent buyers find first.

Cross-references in this batch

For the Stripe-side of agentic commerce, see Stripe Sessions 2026 — the five launches that actually change freelance cash flow. For the broader pricing pressure agentic buyers are creating, see AI-native vs AI-using freelancers — the rate gap explained. And for the underlying data on freelance-spend redirection, see the Ramp data on AI replacing freelancers.

Delivvo's branded client portal exposes structured invoice data to the underlying payment rails (Stripe, PayPal, Tap, Telr, PayTabs, Checkout.com) on the freelancer's own gateway — and never takes a cut. As agent-initiated checkout becomes a meaningful share of bookings, the surface you take payment on matters more, not less. See how it works →

The takeaway

AP2 is the boring infrastructure layer that will quietly decide which freelancers are visible to AI buyers in 2027 and which are not. The protocol itself is not what freelancers should evaluate — the question is whether the booking surface, the checkout, and the structured-data layer they sit on can accept an AP2-style Mandate when an agent shows up.

The freelancers who treat 2026 as the year to productise pricing, move checkout onto a real surface, and add structured data are the ones who will be in the agent-shopping result set. Everyone else is going to spend 2027 wondering where the inbound went.