DoorDash's Early AI Lessons: Native Tools Outpace Third-Party Agents
DoorDash data reveals that while third-party agents remain rare, native AI tools like Ask DoorDash build carts 5x faster and significantly expand basket values.
While autonomous AI agents have captured substantial industry attention, direct consumer adoption for automated ordering remains in its earliest phases. In discussions surrounding automated commerce, DoorDash leadership highlighted a clear distinction between independent third-party agents placing orders and platform-native AI features designed to assist buyers. While external agent usage might currently seem negligible, DoorDash's internal deployments reveal that AI-assisted purchasing is already altering how people shop.
The Early Trajectory of Automated Ordering
Addressing the current wave of third-party ordering agents, DoorDash executive Xu expressed a grounded perspective on early adoption metrics. Rather than buying into unrestrained hype or dismissing the trend altogether, Xu highlighted the importance of balancing realism with technological inevitability. "My take on all things technology is you never want to fight gravity," Xu noted. "But at the same time, you don’t want to be delusional."
This measured posture reflects the historical arc of digital commerce. The nascent state of AI-driven food ordering draws parallels to Pizza Hut’s initial ventures into digital ordering in 1994. At that time, placing a pizza order over the internet was an experimental novelty that few could have anticipated becoming an industry standard. Today, more than half of Pizza Hut orders occur online. A similar transition could unfold for agentic ordering over the coming years, meaning today’s modest figures for autonomous agents might eventually be seen as the humble starting point for widespread habit change.
Native AI Delivers Clear Engagement Gains
While external third-party bots have yet to take over delivery checkouts in significant volume, DoorDash’s internally built generative AI system demonstrates tangible behavioral shifts among users. Rolled out over the summer, the company's "Ask DoorDash" feature acts as an embedded shopping assistant capable of interpreting multimodal inputs and unstructured requests.
The tool allows customers to upload a photograph of a recipe to automatically build a digital shopping cart with the required ingredients. Alternatively, users can enter broad, conversational prompts—such as requesting "dinner for a family of four"—and receive tailored meal suggestions.
According to metrics shared by DoorDash, Ask DoorDash alters buyer behavior in several measurable ways:
- Assembly Velocity: Baskets generated via Ask DoorDash are assembled five times faster than traditional shopping carts.
- Higher Order Value: Orders processed through the tool register an average basket value that is nearly 50% higher than standard orders from the same consumers.
- Expanded Basket Diversity: Carts built with the feature contain approximately 60% more unique items compared to the consumers' routine buying habits.
- Merchant Discovery: Nearly half of the transactions initiated through Ask DoorDash direct orders toward local eateries and shops the consumer had never tried before.
These figures show that while consumers may not yet delegate complete checkout authority to standalone external agents, they actively embrace generative tools that streamline discovery, interpretation, and cart configuration directly within the existing application workflow.
What it means for developers
For software engineers and product teams designing commerce-focused AI applications, DoorDash's findings offer key operational takeaways:
First, integration context matters more than complete autonomy. Users are proving far more receptive to native, embedded conversational assistants than to separate, third-party agents attempting to operate on their behalf from the outside. Creating friction-free paths between prompt interpretation and inventory generation delivers immediate commercial uplift.
Second, multimodal capability is becoming essential for practical consumer agents. The ability for Ask DoorDash to convert a recipe photograph directly into a cart highlights the importance of combining computer vision with structured product catalogs. Developers building similar assistive commerce features can experiment with leading multimodal systems and compare model outputs cost-effectively using one API key through https://apixoai.online, which provides pay-per-token access to models like Claude, GPT, and Gemini.
Third, discovery drives economic value. The fact that nearly 50% of assisted orders go to previously unvisited merchants proves that AI query interpretation works well as an discovery engine. By mapping vague consumer intent—such as family meals—to specific merchant offerings, developers can boost both user convenience and platform revenue.
Ultimately, building agentic systems requires heeding Xu’s advice on avoiding delusion while respecting market gravity. The technology is advancing, but immediate value lies in concrete, guided workflows that save users time and solve clear practical bottlenecks.
Source: Basically no one is using AI agents for their DoorDash orders — Ars Technica AI. Written by the Apixo team from that report.
One key for Claude, GPT, GLM, DeepSeek and more. Pay per token with crypto.
Get your API key
