Summary
Sierra has released Poppy 1.0, a protocol for letting personal AI agents identify themselves to companies, obtain permission and carry out customer-service tasks. Shopify has enabled it across all stores on its platform, while Meta Muse and more than 100 design partners are also adopting or testing the standard.
A protocol for agents acting on a customer’s behalf
Sierra has released version 1.0 of its Personal Agent Protocol, known as Poppy, in an effort to establish a common way for personal AI agents to interact with companies. The protocol addresses a practical problem emerging as AI assistants move beyond answering questions: how should a business identify an agent, explain what the agent is allowed to do and ensure that the customer remains in control of consequential actions?
Most websites were designed around a human user who logs in, navigates a page and clicks through a defined workflow. A personal agent changes that model. It may search for information, retrieve an order, compare options or begin a service request on behalf of a person. Without a shared interaction framework, each website may expose different login procedures, permissions and interfaces, making it difficult for an agent to understand what is available and difficult for the customer to see what has actually been authorized.
Poppy is intended to provide that common layer. According to the source material, a company can use the protocol to identify itself to a visiting agent and specify which actions the agent is permitted to perform, such as checking an order or booking a flight. Customers authenticate through standard logins and decide what their agent can access.
The distinction between access and action is central. An agent might be allowed to read an order status without being allowed to cancel the order. It might be able to search flight options without being able to pay for a ticket. A protocol that makes those differences explicit could give businesses and users a more predictable basis for using agents across multiple services.
Shopify’s broad rollout does not replace its commerce stack
Shopify has switched Poppy on for every store on its platform. This gives the protocol a large potential distribution channel and means that individual merchants do not have to configure the initial capability one by one, according to the report.
Poppy is not, however, replacing Shopify’s existing shopping infrastructure. The protocol handles the moment when an agent arrives at a merchant’s site and identifies itself. Shopify’s existing shopping protocol continues to handle carts and checkout. That division suggests that Poppy is primarily an agent-to-business communication and authorization layer, rather than a complete payment or commerce system.
The source also says that an agent must ask the shopper before taking certain actions. That requirement is significant because an agent’s ability to navigate a site should not automatically be treated as permission to complete a transaction. The user may authorize product discovery or order tracking while withholding permission for checkout, account changes or other actions with external consequences.
For merchants, this model could reduce the need to build a separate agent experience for each provider. For users, the benefit would depend on whether permissions and approval prompts are presented consistently and clearly. A protocol can define the exchange of information, but it cannot by itself ensure that every business implements the controls in the same way.
Early use cases span banking and air travel
Some companies are already using Poppy for live or early-stage workflows. Banking firm Mercury allows agents to check balances and start payments, while those payments remain subject to the account’s existing approval rules. Turkish Airlines allows agents to check fares and find business-class upgrade opportunities, with customers approving the payment before it is made.
These examples illustrate two different categories of agent activity. The first is information retrieval: reading an account balance, checking a fare or finding an available service. The second involves an action that may create a financial or contractual consequence. The latter requires a stronger control model, including explicit customer confirmation and enforcement of the account or merchant’s approval rules.
That distinction is especially important in financial and payment-related workflows. A personal agent may be able to move through a service more quickly than a human, but speed does not remove the need for authentication, authorization or review. The examples described by Sierra indicate that Poppy is designed to preserve existing approval structures rather than bypass them.
For businesses, implementation questions will include how an agent’s permissions are expressed, how approval is recorded and what happens when a requested action fails. For users, the important questions include what information the agent can see, how long authorization lasts and whether it can be revoked. These issues become more consequential when the agent has access to account data or can initiate a payment workflow.
Adoption is expanding, but interoperability is not automatic
Meta’s Muse agent now looks for Poppy first on every site it visits. The source says that more than 10 million people have set up Muse. Sierra also says that more than 100 companies have signed up as design partners, including Delta Air Lines, Intuit, Salesforce and Zoom. The company said that 60 additional companies had joined, bringing the total above 100.
The range of participants suggests that personal-agent interoperability could extend well beyond online retail. Airlines could use agents to retrieve travel information and identify service options. Financial companies could expose account data and tightly controlled operations. Software and enterprise-service providers could use the model for customer support, account queries and routine business processes.
However, a design partnership, an early integration and a fully standardized customer experience are not the same thing. Broad interoperability depends on companies using compatible permission models, agents interpreting those models consistently and users receiving understandable confirmation prompts across different services. Shopify’s adoption may lower the barrier for merchants, but it does not automatically resolve those broader coordination problems.
There is also a difference between an agent finding a protocol and an agent safely completing a task. Discovery can be standardized relatively easily: an agent checks whether a website supports Poppy and learns how to communicate with it. Execution is more complex. It requires the agent to understand business-specific limits, preserve the customer’s intent and avoid taking an action that the user did not approve.
Version 1.0 remains a draft
Despite the 1.0 label, Poppy is still marked as a draft specification. The source material says that the rules can change and that some changes could break existing compatibility. This gives early participants an incentive to monitor revisions and maintain their implementations as the protocol evolves.
That status is not unusual for an emerging standard, particularly one that touches identity, customer data, service access and payments. Real-world deployments are likely to reveal cases that are difficult to define in advance: how an agent should interpret a narrow permission, how to handle a failed or duplicated action, how to pass approval between systems and how to create a reliable record of what the user authorized.
The next stage of Poppy’s development will therefore involve more than adding companies. The specification will need to clarify how businesses expose capabilities and limits, how agents request additional permission and how users can understand or withdraw authorization. Stronger operational guidance may also be needed for actions that are difficult to reverse or that affect financial accounts.
A protocol can provide a shared vocabulary, but it does not eliminate the responsibility of each business to implement access controls, authentication and internal approvals. Nor does it guarantee that an agent will interpret every instruction correctly. Those safeguards remain part of the surrounding system.
A changing relationship between websites and software agents
Poppy 1.0 reflects a larger shift in how online services may be built. AI agents are moving from being conversational interfaces on a single website toward becoming software that represents a user across several services. If that model expands, businesses will need to communicate not only with human visitors but also with agents that require machine-readable information about identity, capabilities and authorization.
The promise is a more interoperable web in which a user can delegate limited tasks without learning the details of every company’s interface. The risk is that a poorly understood authorization could allow an agent to expose data or perform an action beyond the user’s intent. The balance will depend on clear permission boundaries, visible approval steps and reliable enforcement by each participating business.
Shopify’s platform-wide enablement, Muse’s preference for finding Poppy and the participation of more than 100 design partners give the protocol a meaningful early base. But Poppy is still a draft, and its long-term significance will depend on whether the specification stabilizes, whether implementations behave consistently and whether users can continue to understand and control what their agents do on their behalf.
Source: link