Agentic commerce: payments, trust rails, and where moats move (www.thoughtworks.com)

OpenAI and Stripe’s instant checkout announcement and the broader push for an “Agentic Commerce Protocol” mark a shift from conversational AI as an information layer to AI as a transactional environment: agents that reason, plan and complete purchases end-to-end on behalf of users. Unlike today’s rule- or human-driven tools, agentic AI can find products, compare merchants, apply loyalty and pay autonomously, turning search, recommendation and checkout into a single seamless flow. Google Cloud’s Agent Payments Protocol (AP2) and moves by Visa, Mastercard and PayPal to expose agent-friendly APIs show payment networks racing to make their rails agent-ready. Technically, this requires new trust and identity infrastructure: ways to authenticate an agent’s delegated authority, prove a request reflects genuine user intent, and assign accountability for mistakes or fraud. Merchants must make catalogs machine-readable, expose reliable APIs and instrument KPIs around discoverability, data quality and verification scores rather than clicks. Emerging patterns include B2A (merchant-to-agent) feeds, branded A2C assistants, and A2A replenishment flows; platforms will likely productize identity, consent, and agent-to-agent settlement as new revenue streams. Risks are real: scalable social-engineering, disputed chargebacks, and opaque decisioning call for stronger auth, audit trails and explainability. European rules (EU AI Act, PSD3, DSA) push accountability and could shape trust frameworks, but winners will balance speed with interoperable standards. Payments firms should build trust rails, co-create protocols and partner across ecosystems now—deployment could accelerate in months, not years.