“My take on all things technology is you never want to fight gravity. But at the same time, you don’t want to be delusional.”
Redeployed is a weekly newsletter that breaks down one important AI story at a time for product builders and engineering leaders. Every issue explores what the shift means for technology companies and how leaders can respond.
On October 29, DoorDash CEO Tony Xu disclosed that the company had received “maybe several hundred orders in total” from some third-party AI agents.
The number stands out because DoorDash processed 993 million orders during the third quarter. These figures cover different periods, so they cannot be used to calculate an adoption rate. Still, the contrast is revealing: third-party agents have generated very little ordering activity on DoorDash so far.
As TechCrunch reported, DoorDash remains open to working with outside assistants. Xu said the company does not want to resist changes in consumer behavior, but warned against becoming “delusional” about current demand.
The low volume may reflect limited availability, low awareness, or customer hesitation. Whatever the explanation, it exposes a question for companies building agentic products: just because an AI agent can complete a transaction, does that mean customers want it to?
Assistance Is Winning Before Autonomy
The promise of agentic commerce is straightforward. Instead of opening an app, comparing products, building a cart, and checking out, customers could describe what they need and let an assistant handle the transaction.
The technology is becoming capable of doing that. Models can search catalogs, compare products, call tools, and initiate purchases. But completing a transaction requires more than technical execution.
Ordering food is already relatively easy for many DoorDash customers. They know the interface, can browse menus visually, review substitutions, and confirm the total before paying. Moving that experience to an external assistant may introduce uncertainty without eliminating enough effort to justify the change.
There is also a difference between trusting an AI recommendation and trusting an AI purchase. A bad recommendation can be ignored. An incorrect order can charge a card, overlook a dietary restriction, or create a problem that requires customer support.
DoorDash's own AI results offer an interesting contrast.
The company has introduced Ask DoorDash for in-app discovery and shopping assistance, Text DoorDash for conversational ordering, and an MCP interface that external agents can use.
According to DoorDash's reported results, grocery baskets created with Ask DoorDash are built five times faster than conventional baskets. The company also reports nearly 50% higher average basket value and approximately 60% more unique items compared with traditional orders from the same consumers.
These are company-reported figures from different Marketplace samples collected between June and early September 2026. They do not establish long-term adoption or prove that the AI experience caused every improvement.
They do, however, suggest a useful product direction. AI can reduce friction inside an experience customers already understand, using the preferences, inventory information, and fulfillment systems the platform already controls.
A grocery customer might appreciate an agent rebuilding a recurring cart while still wanting to review substitutions and confirm payment. That is a smaller behavioral change than asking an external assistant to independently decide what to buy.
For technology leaders, the opportunity may be to improve existing customer journeys before trying to replace them.
Autonomy Has to Be Earned
Product teams should treat delegation as a progression rather than a feature that is either enabled or disabled.
An agent that recommends products requires relatively little authority. One that creates a cart needs access to customer preferences and inventory. Checkout introduces payment permissions, while proactive purchasing requires the system to decide what to buy and when to buy it without a new request.
Each step changes the cost of an error and the controls needed to prevent it.
In production ecommerce and retail systems, companies can introduce autonomy function by function, with human approvals around consequential actions such as large purchases, disputed transactions, or customer-facing price changes.
The engineering challenge is to make those boundaries reliable. Customers need to understand what an agent selected, approve important decisions, set spending limits, and correct mistakes without navigating a complicated support process.
The same principle applies beyond commerce. A travel assistant might compare itineraries before being authorized to book a nonrefundable ticket. A procurement agent might prepare a purchase order but require approval before committing company funds.
Teams should also measure each stage of delegation separately. Successful tool calls and available integrations indicate technical capability, but they reveal little about customer adoption.
More useful measures include completed transactions, repeat usage, incremental conversion, reversals, support contacts, and retention.
Repeat usage may be the strongest early signal. A customer might try autonomous purchasing because it is new. The more meaningful result is whether that customer trusts the agent with the same decision again after seeing what happened.
This creates a practical test for product teams: identify one decision customers already make repeatedly, determine how much of it they are willing to delegate, and measure whether that behavior continues.
This issue of Redeployed is brought to you by Tecla Labs: The difficult part is rarely proving that an agent can complete an action. It is identifying a workflow people are willing to delegate, the data and permissions the agent needs, the approval boundaries that preserve trust, and the operating owner responsible when something goes wrong. Tecla Labs maps how work moves through a company, identifies where AI can reduce friction or produce measurable value, and defines the first system worth building. Start with a free AI Opportunity Assessment to explore a prioritized roadmap grounded in your workflows, data, and decision points.
Build for Agents Without Betting the Company on Them
DoorDash is keeping its options open. It is making its platform accessible to external agents while continuing to improve the customer experience inside its own products.
That is a useful approach for other technology companies. Structured product data, reliable APIs, scoped permissions, and observable transactions can prepare a business for agent-driven interactions without requiring it to rebuild the entire customer experience around a behavior that has yet to demonstrate broad adoption.
There is a longer-term distribution risk worth watching. If external assistants eventually control product discovery and purchasing, they could influence which brands customers consider and which companies receive transactions. Businesses may become increasingly dependent on interfaces they do not control.
For now, the more immediate opportunity is to learn where customers actually want assistance and how much authority they are willing to give it.
The first winners in agentic commerce may not be the companies that automate the most. They may be the ones that identify a small but valuable decision customers are willing to delegate repeatedly, prove its reliability, and expand autonomy as trust develops.
Connect With Other Technology Leaders
If you want to exchange practical ideas with senior technology and product leaders navigating the AI era, check out the upcoming GILD Forums. They bring together experienced operators for peer discussions around the technology and business decisions they are working through right now.
More to come…
Recommended Reads
✔️ DoorDash says only several hundred orders came from AI agents — TechCrunch
✔️ DoorDash Q3 2026 Earnings Call — SEC
✔️ Skip The App: Now You Can Just Text DoorDash — DoorDash
– Gino Ferrand, Founder @ Tecla Labs