DoorDash CEO: AI Coding Tools Alone Cannot Drastically Boost Developer Productivity
DoorDash CEO Tony Xu stated that AI coding tools cannot yet drastically improve overall developer productivity because coding represents only a fraction of a software engineer's daily responsibilities. He noted that while AI writes over half of DoorDash's code, engineers still spend significant time on meetings, product reviews, and alignment. This perspective highlights a realistic industry shift, emphasizing that true productivity gains require integrating AI across the entire software development lifecycle (SDLC) rather than just automating code generation. It challenges the hype that AI coding assistants will immediately replace or drastically downsize engineering teams. Xu estimated that actual coding takes up only 25% to 50% of an engineer's time, with the rest spent on collaboration and design. Additionally, AI pioneer Andrej Karpathy has also pointed out that AI-generated code is often bloated, redundant, and poorly structured, despite being functional.
## BACKGROUND
AI coding assistants like GitHub Copilot use large language models to help developers write, debug, and document code. However, modern software engineering involves a broader lifecycle (SDLC) that includes system design, cross-functional communication, and maintenance, leading to the emergence of 'AI-native software engineering' which aims to integrate AI across all these phases.