Approximately 3% of transactions involve an AI agent, and this is set to rise. According to Checkout.com, 89% of merchants are already preparing for agentic payments. Stripe, Visa and Mastercard are among the businesses that have now published a way to enable this change:

  • Stripe and OpenAI released Instant Checkout in September 2025.
  • Visa followed the same month with Trusted Agent Protocol, so that merchants could distinguish a legitimate agent from a bot attack.
  • Mastercard and Google opened Verifiable Intent in March 2026.
  • In June 2026, Mastercard released Agent Pay for Machines with more than 30 launch supporters behind it, among them Adyen, Checkout.com, Stripe, Coinbase and Cloudflare.

In response to these developments, Enablis has prototyped three tools that sit across different elements of the agentic purchasing landscape.

Demo One - Buying

Built by Oliver Gilbert, Senior Consultant, Enablis

More people than ever research a purchase with AI, and almost all of them then have to leave that conversation to buy, starting again on a website that knows nothing about the twenty minutes they just spent deciding.

For a retailer that means a lost sale and a customer who was in the market without ever appearing in the funnel. This prototype keeps research, comparison, visualisation and payment inside a single conversation, with the customer approving the order themselves.

A conversational agent handles product research and filtering, then completes a full paid checkout inside the same chat window - running an Agentic Commerce Protocol server, Stripe’s Shared Payment Token and verifiable intent underneath.

Demo Two - Permission

Built by Mike Rossiter, Senior Cloud Engineer, Enablis

Once an agent can pay, the harder question is what it was allowed to pay for, and neither the merchant nor the bank currently has a reliable way to know. That matters because UK payment rules, dispute handling and fraud liability all assume a human gave a specific instruction at the moment of payment.

This prototype gives the customer a signed spending limit that travels with the agent and is checked before the money moves, which is the record a bank, a PSP or a merchant will be asked for when something goes wrong.

Set a permission - “spend up to £200 this week” - and your device signs it. In the demo, a hacked agent tries to raise that signed £200 limit to £1,000 and gets rejected on screen.

Demo Three - Readiness

Built by Stephen Martin, Senior Infrastructure Engineer, Enablis

Most organisations accept that agentic payments are coming and have no way to say what they would need to change to accept one, so the conversation stays at the level of strategy and nothing gets scoped or funded. This prototype scores a merchant against the six capabilities an agent payment actually requires and names the gaps, giving an engineering team and a risk team the same list to work from.

An animated flow puts traditional checkout and the agentic flow side by side, then lands on a scorecard that rates the merchant and names the exact infrastructure gaps to close.