Projected agentic commerce volume by 2030
Annual online sales potentially agent-driven by 2030
of routine purchases could be fully automated by 2028
AI agent instances projected to be active globally by 2027
AI agents are beginning to handle product discovery and comparison. Early adopters in retail and travel are seeing agent-initiated sessions. The infrastructure for agent identity and delegated payments is being built.
Merchants begin optimising for agent visibility. New 'agent-optimised' product data standards emerge. Payment networks introduce agent authentication tokens. Early regulatory frameworks are proposed in the EU and UK.
Dedicated payment rails for agent-initiated transactions. Issuer risk models updated to incorporate agent trust scores. First wave of agent-specific dispute resolution frameworks. Widespread consumer adoption of delegated purchasing.
Agentic commerce becomes the dominant mode for routine purchases. The 5-party model is the standard. Brands that have not adapted to agent-native commerce face structural decline. A new generation of 'agent-first' merchants emerges.
The most important infrastructure investment of the next three years is not AI capability — it is agent identity. Who is this agent? On whose behalf is it acting? What are the limits of its authority? These questions must be answered before agentic commerce can scale safely.
In a world where AI agents make purchasing decisions, the quality, completeness, and machine-readability of your product data is your most important marketing asset. Structured data, verified reviews, and real-time inventory signals become the new brand equity.
The existing consent frameworks — cookie banners, terms of service, PSD2 SCA — were designed for human decision-makers. Agentic commerce requires a new consent architecture that captures the scope of delegation, the limits of authority, and the conditions of revocation.
The organisations that will thrive in the agentic era are not those that abandon human-centric design, but those that build experiences that work for both humans and agents simultaneously. The interface layer must serve two audiences with very different needs.