DoorDash AI coding interview: 6 candidate reports

Mid-level · Own machine · JavaScript · 60 min

E4 full stack loop in Seattle; the AI Code Craft round was described as a new pilot. In 60 minutes the candidate built a frontend feature, live delivery tracking over server-sent events, step by step on a starter Node backend, using a real AI coding agent (Claude Code, Cursor or Codex) with full screen share; a chat window alone is not allowed. Later milestones added real-time updates. Graded: prompt quality, why you pick each model, engineering judgement, how you validate AI output (unit tests, Playwright) and how you’d structure and review PRs. A separate manager round also asked how they use AI day to day.

AI: Your own AI tools that could read the files, write code, run tests · Task: Build from a spec · Source: LeetCode Discuss post

Own machine

The AI Code Craft round was the same kind of task as the screening Code Craft, but with AI allowed. The candidate spent the first 20 to 30 minutes planning instead of implementing, which the interviewer didn’t seem to like, although the planned code then took about 3 minutes and was easy to extend. Follow-up questions were about how they validate the AI’s code. The rest of the loop: behavioural with a product thinking part, system design and a debugging round.

AI: Your own AI tools that could write code, run tests · Task: Build from a spec · Source: Blind thread

Own machine · 60 min

DoorDash’s own description: traditional coding rounds are replaced by a 60-minute AI-assisted working session on your own machine and editor. Using an editor-integrated assistant is required, every feature including agent mode is allowed, and ‘yes, you are allowed to one-shot problems’. Starter code in your language looks like real DoorDash work, such as extending a dispatch system or automating support requests. Signals: getting oriented in the code, using AI and verifying it, debugging, managing scope and communicating trade-offs. The first pilot’s three-service project was too big, so projects got smaller and candidates now get a guide in advance.

AI: Your own AI tools that could read the files, write code, run tests · Task: Extend existing code · Source: Official page

Own machine · Python · 60 min

Onsite AI Code Craft round in the US. You bring a local editor with a real AI setup (Copilot, Cursor, Claude Code); a ChatGPT tab is not enough. The starter code was a workflow engine for automating support tickets such as late orders, with stubbed functions and one test. Part 1: make the test pass. Part 2: clarify requirements with the interviewer and add a partial refund case, with your own extra tests. Part 3: the engine is too slow, so investigate and improve it. The interviewer sets the pace and you’re not expected to finish every part. The candidate, who has interviewed others, warns against using the LLM as a crutch.

AI: Your own AI tools that could read the files, write code, run tests · Task: Extend existing code · Source: LeetCode Discuss post

Mid-level · Own machine

E4 backend loop after the Code Craft screen: AI Code Craft, debugging, a project discussion plus system design, and engineering values, about 5 hours in total. For the AI Code Craft round you download an IDE such as Cursor or GitHub Copilot, prompt it, run the code locally and get to a working solution.

AI: Your own AI tools that could read the files, write code, run tests · Source: Blind thread

DoorDash’s separate policy on AI in interviews: don’t use AI in live interviews unless the interviewer invites you to, and undisclosed use can disqualify you. The AI working session is the exception, so check which of your rounds allow it. A June 2026 candidate who had to type code by hand in an early HackerRank round confirms the split.

AI: No AI · Source: Official page