Sierra Releases Poppy v1.0: Preparing for Agent-to-Agent Commerce

The Touchpoint Shift: From Human Clicks to Agent Handshakes

For three decades, digital business rested on a single premise: a human being opens a browser or an application, navigates visual menus, and clicks buttons. That interface model is reaching its limits. With the release of Personal Agent Protocol (Poppy) v1.0 by conversational AI company Sierra, customer touchpoints are shifting from human browsing to direct agent-to-agent (A2A) negotiation.

Instead of a customer manually searching for flights, comparing banking fees, or filling out ecommerce checkout forms, an external personal agent performs these tasks autonomously. When consumer AI agents become the primary visitors to digital properties, corporate websites and self-service portals cannot rely exclusively on visual user interfaces designed for human eyes.

Poppy v1.0 provides an open standard that governs how an individual customer's AI agent connects to enterprise systems. Built largely on existing foundations—the Model Context Protocol (MCP), OpenAPI, and the Universal Commerce Protocol (UCP)—Poppy sets formal boundaries for identity, discovery, and execution.

Adoption has moved quickly: - Shopify has activated Poppy across its entire merchant base by default, linking store inventories directly to external agents. - Meta has configured its consumer AI assistant, Muse, to scan for Poppy protocol files first whenever it lands on any website. - Over 100 design partners have joined the initiative, including Salesforce, Twilio, Delta Air Lines, Affirm, Hyatt, Freshworks, and Zoom. - Early adopters like Mercury and Turkish Airlines are already processing authenticated financial and travel queries routed directly from personal agents.

To engage with this shift, enterprise leaders must understand how the protocol works and evaluate whether their internal systems are ready to receive external AI agents.

---

How Poppy v1.0 Bridges Personal Agents and Corporate Backends

When a personal agent attempts to accomplish a goal on behalf of a user, it faces three technical hurdles: discovering what the business offers, authenticating identity without exposing credentials, and invoking backend tools safely. Poppy structures this exchange across defined operational stages.

+-------------------------------------------------------------+
|                      Customer (User)                        |
+-------------------------------------------------------------+
                               |
                   Delegates task via prompt
                               v
+-------------------------------------------------------------+
|              Personal AI Agent (e.g., Meta Muse)            |
+-------------------------------------------------------------+
                               |
                    1. Checks Poppy File
                    2. Requests OAuth Permission
                               v
+-------------------------------------------------------------+
|                 Poppy Protocol Gateway / Proxy              |
+-------------------------------------------------------------+
             |                                    |
   3. Pre-transaction Handshake          4. Safe Execution
             v                                    v
+-----------------------------+     +-------------------------+
|   Enterprise AI Agent /     |     |   OpenAPI / MCP Endpoints
|   Customer Service Agent    |     |   & Domain Extensions   |
|   (e.g., Smart Website)     |     |   (e.g., Shopify UCP)   |
+-----------------------------+     +-------------------------+

1. Pre-Transaction Discovery vs. Transactional Execution

A common point of confusion is how Poppy interacts with existing ecommerce and enterprise standards. Poppy handles what happens before transactional execution. When a personal agent arrives at a domain, Poppy defines how the agent identifies itself, verifies which business tools are exposed, and determines what it is permitted to inspect.

Transactional execution is then delegated to domain-specific protocols. In commerce, Poppy links directly to UCP. While UCP handles finding products, filling carts, and running payments, Poppy handles the initial handshake and context exchange. Through Poppy's commerce extension—co-authored by Shopify—a personal agent can talk directly to an online store's agent (via UCP Ask), operate as a guest, or prompt the user for single-sign-on credentials.

2. Guardrails and Permission Delegation

Personal agents cannot be given unchecked API keys or carte blanche access to a user's account. Poppy relies on OAuth authentication, putting the end customer in explicit control of permissions.

Enterprises can require explicit human confirmation for sensitive operations: - Mercury allows personal agents to fetch balances, summarize transactions, or track receipts automatically. However, if an agent initiates an outgoing transfer, Mercury forces the request through the account's existing approval rules before funds move. - Turkish Airlines permits agents to pull booking data via surname and reference code or search for business-class upgrades. But the purchase itself requires the Miles&Smiles member to inspect and approve the transaction before payment completes. - Synchrony is testing agent-led credit and reward recommendations across its merchant network, ensuring credit checks remain strictly bounded by consumer consent.

---

Three-Layer Interface and Permissions Self-Audit Checklist

If your customer service channels and website APIs only support human clicks and basic web scrapers, external agents will fail to interact with your organization.

Use this three-layer checklist to evaluate your enterprise readiness for external personal agents.

Audit Layer	Core Objective	Key Verification Questions	Required Standards & Tech
Layer 1: Discovery & Identity	Expose your public capabilities and agent identity to inbound AI traffic.	• Does your website publish a standardized agent manifest (e.g., Poppy file)? • Can inbound assistants like Meta Muse determine what services your business offers? • Do you provide machine-readable guidelines for conversational constraints?	Static manifest hosting, OpenAPI metadata, Sierra validation tools
Layer 2: Capability & Tooling	Expose structured backend actions without breaking core business logic.	• Are self-service functions documented as tool calls using MCP or REST schemas? • Can an external agent query your product catalog or FAQs without scraping HTML? • Can your enterprise agent route complex tasks to customer service agents?	Model Context Protocol (MCP), UCP, structured JSON schemas
Layer 3: Security & Authorization	Restrict agent execution and mandate human approvals for risky actions.	• Do you use OAuth scopes to separate read actions from write actions? • Are high-value steps (payments, account edits) protected by human-in-the-loop gates? • Can an audit log trace which agent initiated an action on behalf of a specific user?	OAuth 2.0 / OIDC, policy approval engines, audit logging

Layer 1: Discovery and Identity Layer

Personal agents need a machine-readable directory to know your rules of engagement. 1. Manifest Availability: Determine if your root domain hosts a machine-readable configuration file that personal agents can parse upon arrival. 2. Guidance and Feedback Channels: Ensure your server can provide operating instructions to an external agent (such as clarifying question formats or return policies) and capture feedback when an agent fails to complete a task. 3. Validator Testing: Verify that your manifest passes standard protocol validation tools, ensuring formatting matches v1.0 specifications.

Layer 2: Capability and Tooling Layer

Exposing HTML to personal agents results in brittle, error-prone interactions. Agents require structured endpoints. 1. Standardized Tool Exposure: Convert legacy customer service scripts into structured MCP tools or OpenAPI endpoints. 2. Domain-Specific Extensions: If you run an ecommerce business, ensure your store connects to protocols like UCP so personal agents can navigate inventory and carts programmatically. 3. Agent-to-Agent Bridging: Ensure your customer-facing digital touchpoints—such as a website AI advisor embedded in a Smart Website—can converse directly with an incoming personal agent, resolving ambiguities before escalating to human staff.

Layer 3: Security and Authorization Layer

You must protect both your systems and your customers from unauthorized automated commands. 1. Granular OAuth Scoping: Configure authentication policies so personal agents receive the minimum necessary privileges (e.g., read:orders versus write:transfers). 2. Action Tiering: Classify actions into two distinct buckets: - Autonomous Actions: Safe tasks an agent can execute without real-time human intervention (e.g., checking flight status, tracking order delivery, checking return eligibility). - Confirmation-Gated Actions: Irreversible or financial tasks requiring out-of-band user approval (e.g., booking a non-refundable ticket, changing shipping addresses, initiating a refund). 3. Identity Verification and Session Revocation: Ensure customers can view and revoke permissions granted to specific personal agents at any time.

---

Architectural Implications for Enterprise Customer Service

The emergence of standards like Poppy signals a major transformation for enterprise customer service and digital marketing teams. For years, digital strategy focused on driving human eyeballs to owned web assets. As personal agents take over routine queries, website traffic composition will change dramatically.

Traditional Customer Touchpoint:
Consumer  ──>  Web Browser / App UI  ──>  Enterprise Backend

Agent-to-Agent Touchpoint (Poppy Standard):
Consumer  ──>  Personal Agent (Muse, etc.)  ──[Poppy / MCP]──>  Smart Website / Enterprise Agent  ──>  Enterprise Backend

Transforming Customer Service into Agent Gateways

Customer service portals have historically been built as defensives moats: phone trees, human chat queues, or rigid rule-based chatbots. When a consumer's agent interacts with these systems, standard chat widgets fail because they expect human latency and human reading patterns.

By adopting standardized agent protocols, enterprises convert customer service into structured programmatic endpoints. Instead of answering repetitive questions about order statuses or return windows over live chat, enterprise customer service systems can resolve those requests in milliseconds through direct agent negotiation.

Modernizing Digital Assets with FDE

Preparing an enterprise for agent-to-agent operations requires bridging legacy databases with modern AI standards. At Starbox AI, this architectural transformation is addressed through Enterprise AI Adoption (FDE).

FDE guides enterprises through the practical steps of agent readiness: - Upgrading Web Infrastructure: Moving from static portals to dynamic Smart Websites equipped with a website AI advisor that can negotiate with external personal assistants. - Exposing Internal Knowledge: Structuring legacy databases and brand guidelines into clean, MCP-compatible formats that external agents can reliably interpret. - Deploying Multi-Agent Systems: Deploying multi-agent collaboration frameworks internally so incoming personal agent requests are routed automatically to the correct department agent—whether for commerce, billing, or technical support.

Organizations that implement these architectures early will lower their cost to serve and establish direct, frictionless relationships with their customers' AI representatives.

---

Moving Forward: Phased Adoption for 2026 and Beyond

You do not need to rewrite your entire digital stack immediately to prepare for agent-to-agent commerce. Sierra designed Poppy v1.0 to support staged adoption:

1. Phase 1: Publish Metadata and Read-Only Tools. Begin by cataloging your existing public APIs and documentation. Expose safe read-only queries—such as product searches, flight timetables, or policy documents—via standardized formats.
2. Phase 2: Establish Agent Handshakes. Implement protocol files on your primary web assets. If your company operates on hosted platforms like Shopify, verify that protocol files are actively publishing and properly mapped to your inventory.
3. Phase 3: Connect Transactional Workflows with Approval Gates. Integrate OAuth-backed write operations for core account functions, starting with low-risk tasks and maintaining strict user confirmation gates for payments and contract changes.

As consumer AI agents like Meta Muse become standard fixtures on consumer devices, businesses without agent-ready interfaces will effectively become invisible to automated shoppers. Auditing your interfaces today ensures that when your customer's agent arrives, your business is ready to complete the handshake.