Dordash reveals agent-driven commerce harvest, only hundreds of orders in a sea of millions of deliveries
The operational figures released by the US-based Dordash platform reveal a huge gap between the expectations surrounding commerce that relies on smart agents and the actual consumer behavior on the ground. In the disclosure of the company’s third-quarter results, CEO Tony Shaw announced that the total orders the platform received through third-party AI agents such as “Muze”, “Instinct” and “Foo” did not exceed a few hundred transactions. This result is striking when compared with the platform’s daily volume; its 2025 annual report records an average of three million deliveries per day, meaning that the contribution of independent software to purchases remains practically zero.
Between Yielding to Forces of Change and Avoiding Computational IllusionsShaw articulated his management philosophy for dealing with emerging technology, confirming that his company does not seek to resist technological gravity, but at the same time refuses to be swept up by estimates and fantasies disconnected from commercial reality. Although an overwhelming majority of customers continue to purchase directly through the mobile app, the platform chose to build flexible working relationships with independent agent developers rather than cut them off, while planning to capture advertising opportunities within those agents’ environments in the future if they become primary search and request gateways.
Dordash did not stop at waiting; over the past months it rushed to build its private proxy system to protect the direct link with consumers and the business sector. The company launched the Ask Dordash robot in June, a conversational assistant that allows customers to place orders through interactive text andDordash did not stop at waiting; over the past months it rushed to build its private proxy system to protect the direct link with consumers and the business sector. The company launched the Ask Dordash robot in June, a conversational assistant that allows customers to place orders through interactive text and image analysis, followed in July by offering a proxy that works via a command-line interface, and then launching a proxy for ordering by text messages in September. To support the corporate and office sector, the platform provided a model-context protocol, known as the MCP system, enabling external enterprise chatbots to complete group food orders without leaving the work environment.
The split among trade giants defines the contours of the upcoming battle.While e-commerce sites are perplexed about allowing such software to browse their databases, Amazon chose to block personal agents to protect user data and its direct advertising channels, whereas Walmart opened its doors fully to receive their requests in order to capture additional market share from competitors. This contrast reveals the fierce fight over ownership of the most valuable step in the value chain: who retains the interaction interface and the moment of payment, and who becomes merely a logistical network executing algorithms from abroad.
This scenario carries direct operational implications for those who run fast-commerce platforms or retail chains in the Gulf, Egypt and the Levant; the rush to redesign the digital architecture to accommodate individual consumer agents appears premature and drains budgets, given that the end user clings to familiar applications. The strategic priority today requires preparing standard integration protocols for enterprise and corporate orders, while strictly preserving consumer-facing interfaces and advertising networks within apps, because an early concession of screen prominence to external software strips platforms of commercial sponsorship revenue and turns them into low-margin freight carriers.