Anatomy of an Agentic Purchase
One transaction. Eight steps. Six protocols. Three architectural layers. Watch how a shopping agent buys trail running shoes autonomously - from human delegation to delivery confirmation - and see where trust propagates, transforms, and can break at every protocol handoff.
Use the controls below, arrow keys, or press Space to advance. Press P to play/pause. For the full analysis, read Multi-Agent Orchestration (Issue 068).
Human delegates purchase to shopping agent
Sarah tells her shopping agent: "Find me trail running shoes, women's size 8, under £120, delivered by Friday." The agent connects to commerce tools via MCP, translating the natural language request into structured delegation parameters.
Technical Detail
MCP establishes a tool-use session between the agent and external commerce APIs. The agent's context window now includes product search capabilities, price comparison tools, and delivery estimation services. Each tool connection is authenticated via MCP's capability negotiation.
Handoff
MCP passes structured tool capabilities to the agent. The agent now has access to query commerce catalogues.
Failure Risk
If MCP fails to negotiate tool capabilities, the agent cannot access any commerce data. The transaction dies at step one.
The Full Transaction Timeline
Eight steps, six protocols, approximately 11.5 seconds from delegation to confirmation. Click any step to jump directly to it.
What This Transaction Reveals
Trust propagation is the real design challenge
Sarah's delegation constraints - budget, size, delivery deadline - must survive 7 protocol handoffs across 3 architectural layers. Each handoff is a potential trust leak. The transaction succeeds not because each protocol works individually, but because constraints propagate faithfully across every boundary.
Observability determines trust recovery
When the agent adjusts the shoe size from UK 8 to UK 8.5, it must explain why. Without observability into the agent's reasoning chain, Sarah cannot calibrate her trust in future delegations. The size adjustment is correct - but if the agent cannot explain it, trust erodes regardless of outcome.
Failure modes are protocol-boundary failures
Every failure risk in this walkthrough occurs at a protocol handoff - not within a single protocol. UCP returns unstructured data. A2A loses delegation context. VIC cannot verify expired credentials. The protocols work. The boundaries between them are where transactions break.
No orchestration layer exists yet
This transaction required 6 protocols but no single layer ensured trust propagation across all of them. Each protocol maintained its own context. The missing piece - the orchestration layer that governs cross-protocol trust - is the central question of agentic commerce architecture.