The Bring-Your-Own-Bot Pattern: Why Marketplaces Must Be Agent-Agnostic

A specific design question is going to determine which marketplace operators succeed in agent-to-agent commerce: must participants use the marketplace's agent, or can they bring their own? This paper argues that the answer must be the latter, the marketplace must be agent-agnostic: and explains why this is structurally necessary, what it requires of the marketplace's architecture, and how the implications cascade through the rest of the design.

The closed-platform temptation

Every marketplace operator faces the same temptation early in the design phase: tightly couple the marketplace to a specific agent implementation, ideally one the operator builds itself. The intuition is that this gives the operator control over the user experience, ensures consistent quality, and creates a captive ecosystem.

This intuition is wrong, for the same reasons that walled-garden platforms have repeatedly lost to open ones in adjacent technology categories. The marketplace's value to participants is proportional to the breadth of counterparties available on it. A marketplace that requires consumers to use its specific agent excludes consumers who already have agents from other providers, including the consumer's bank, their advocate provider, their personal assistant. A marketplace that requires brands to use its specific agent excludes brands that have already invested in their own agentic infrastructure or that prefer specialized vertical solutions.

The exclusions compound. Each exclusion reduces network density. Reduced density makes the marketplace less valuable to remaining participants. Less value to participants increases the rate of churn. The marketplace that started by trying to control the agent ends with a captive but small ecosystem and watches a more open competitor compound past it.

The historical pattern is consistent. Closed messaging networks lost to open SMS, then to open internet protocols. Closed search platforms lost to the open web. Closed app stores have repeatedly faced regulatory challenge and competitive erosion from cross-platform alternatives. The agentic-commerce marketplace category will follow the same arc: open beats closed for structural reasons, not for ideological ones.

What "bring your own bot" requires

Designing for agent-agnosticism is not a feature toggle. It is an architectural commitment that has implications throughout the marketplace's stack.

Standardized protocol grammar. Every agent that joins the marketplace must be able to participate using a single, public protocol: covering identity claims, authority declarations, negotiation grammar, and receipt schemas. The marketplace's role is to define and maintain the protocol; the agents' role is to implement it. This is the model that internet protocols use, and it is the model that produces the largest networks.

Open identity acceptance. Any agent identity issued by a recognized identity provider must be acceptable to the marketplace. Lock-in to a single identity issuer would recreate the closed-platform problem at the trust layer. The marketplace's role is to maintain a list of accepted identity providers and the verification logic for each; the identity providers' role is to issue identities to qualifying agents. This model has worked in the financial industry for decades (Visa accepts identities issued by member banks) and will work in agentic commerce.

Transparent reputation scoring. Each agent on the marketplace accumulates a reputation score based on its behavior, honored-commitments rate, dispute outcomes, response latency, principal-confirmation compliance. The scoring methodology must be public and auditable. The marketplace's role is to compute and publish the scores; the agents' role is to behave well. Closed scoring methodologies invite regulatory scrutiny and erode counterparty trust.

Inspectable receipt logs. Every transaction's signed receipt must be accessible to both principals (and, with appropriate authorization, to regulators) in a standard format. The marketplace acts as a notary, not a gatekeeper. Closed receipt formats produce vendor lock-in that the operator may want in the short term and the participants will route around in the medium term.

These four design commitments: standardized protocol, open identity, transparent reputation, inspectable receipts, are the architectural expression of agent-agnosticism. They are also the design commitments most likely to be undermined by short-term commercial pressure on the marketplace operator. Vendors that hold the line on all four will compound; vendors that erode them will lose to vendors that did not.

What the marketplace operator actually owns

A natural objection: if the marketplace is agent-agnostic and based on open standards, what does the marketplace operator actually own? The answer is the position, not the protocol.

The marketplace operator owns:

The convening function, being the place where the highest-quality consumer agents and the highest-quality brand agents have already chosen to meet. The convening is itself the value, and it compounds with participation in a way that no single agent provider can match.

The trust-aggregation function: maintaining the reputation scores, the dispute-resolution mechanisms, and the appeals processes that make the marketplace credible to participants on both sides. This function is operationally intensive and requires institutional credibility that takes years to build.

The negotiation-corpus function, the aggregated dataset of signed traces across every transaction the marketplace has cleared. This corpus is structurally inaccessible to any participant or competitor and is the most defensible asset the marketplace accumulates over time.

The standards-stewardship function: being the institution that maintains the protocol grammar, evolves the identity-acceptance logic, and updates the receipt schemas as the category matures. This is unglamorous work that earns durable institutional credibility.

None of these positions require the marketplace operator to build or own any specific agent. They are positions of network architecture, not product control. The operators that focus on these positions and resist the temptation to compete with their participants on the agent layer will compound the longest.

Implications for participants

The bring-your-own-bot pattern has implications for the participants on both sides of the marketplace.

For consumer-side vendors. Build a great agent and integrate with the marketplaces that achieve liquidity. Resist the temptation to build a closed end-to-end experience that requires consumers to use only your agent on only your marketplace. The closed pattern caps your TAM at the size of the captive ecosystem you can convince consumers to join, which is much smaller than the size of the open ecosystem your agent can serve if it can transact across multiple marketplaces.

For brand-side vendors. Build agents that are interoperable with the major marketplace protocols from day one. The marketplaces that achieve liquidity will become the channel through which the highest-intent customer requests reach your brand. Brands whose agents are not present on those marketplaces will lose customer interactions to competitors whose agents are present.

For consumers and the businesses that represent them. Choose agents that work across marketplaces, not agents tied to a single platform. The cross-marketplace agent will give you the best coverage of brand interactions over time, even if a platform-tied agent is initially more polished within its ecosystem.

Closing

The agent-agnosticism design choice is the load-bearing decision for marketplace operators in agentic commerce. Operators that get it right will compound through network effects and define the standards for the next decade. Operators that get it wrong will spend years building closed ecosystems that lose to open competitors.

The pattern is not new. It has played out the same way in messaging, in search, in app distribution, and in every previous network-effect category. The agentic-commerce marketplace will follow the same trajectory. The question is not whether the open pattern wins; the question is which specific operator commits to it most credibly and earns the trust of the participants on both sides.

We expect the leading marketplaces of 2030 to be operators that committed early to bring-your-own-bot, invested in the four architectural commitments above, and resisted the temptation to leverage marketplace position into agent dominance. The temptation is real and continuous; the discipline to resist it is what separates durable platforms from short-lived consortia.