DoorDash CEO: AI coding productivity gains require workflow changes, not just code generation

DoorDash CEO Tony Xu says that AI coding alone is not enough to boost engineering productivity. While AI now writes about two-thirds of DoorDash's code and productivity gains are north of 50%, Xu argues that real gains require AI-enabled workflows across product reviews, design meetings, and business alignment. For AI builders and engineering teams, this means investing in cross-functional AI workflows, not just code generation tools.

What happened

In a podcast interview, DoorDash CEO Tony Xu stated that AI is responsible for roughly two-thirds of the company's code and that productivity gains from AI coding are well north of 50% Business Insider. However, he emphasized that code writing accounts for only 25% to 50% of an engineer's time. The rest is spent on product reviews, design meetings, and aligning with business teams Business Insider. Xu said those aspects also need AI augmentation to achieve the hoped-for productivity gains.

During a May earnings call, Xu noted that AI has not yet changed DoorDash's organizational structure Business Insider. He also pointed to physical AI efforts such as food-preparation robots and autonomous delivery vehicles as areas where the company still needs to improve.

Why AI builders should care

Xu's comments highlight a critical gap in the current AI coding narrative. Many teams focus on adopting code-generation models but neglect the workflows that surround coding. If product reviews, design alignment, and cross-functional coordination remain manual, the overall productivity gain is capped. Xu argues that companies need to become AI-native in how they operate, not just in how they develop code Business Insider.

Other tech leaders share this caution. Wix CEO Avishai Abrahami said the industry is overselling AI, and Andrej Karpathy, who coined the term vibe coding, described AI-generated code as messy, brittle, and bloated Business Insider. For builders shipping AI products, this means that code generation is only one piece of the puzzle.

Practical implications

Engineering teams should invest in AI-enabled workflow orchestration tools that automate or assist with product reviews, design alignment, and cross-team communication. Simply adding a code-generation model without changing how teams collaborate will leave productivity gains on the table.

DoorDash's experience also suggests that physical AI and operational automation remain important for companies with real-world logistics. Builders working on enterprise or delivery platforms should consider how AI can augment non-coding tasks across the organization.

Caveats

Xu's claims about two-thirds of code being AI-written and productivity gains north of 50% are based on DoorDash's internal metrics and have not been independently verified. The company has not disclosed specific productivity measurement methods. Additionally, AI-generated code quality remains a concern: Karpathy noted that it can be bloaty and brittle, which may offset some gains if not carefully reviewed Business Insider.

Xu also acknowledged that AI has not yet changed DoorDash's organizational structure, suggesting that the full impact of AI on team design is still uncertain. The broader industry debate shows that while AI coding is advancing rapidly, its integration into workflows and team structures is still evolving.