Just finished the DoorDash loop (L4 SWE, mobile background but interviewing as backend). Wanted to document what the coding portion actually looks like because there's outdated info floating around.
OA / technical screen coding: I didn't get a separate OA. They scheduled a combined 60-minute technical screen: 30 minutes of coding in a shared editor, then a light system design discussion for the remaining 30. The coding problem I got was graph-based, medium-hard on Leetcode's scale, involving traversal with some state tracking. Not a classic easy problem. But also not a red-black tree implementation.
They use a live interview format, not a timed standalone assessment. This matters because communication counts. I talked through my approach before writing any code and the interviewer flagged one edge case I missed. That helped me fix it before I even ran the tests.
Onsite coding rounds: Two coding rounds during the virtual onsite. One was more algorithmic (BFS/DFS variant), one was more practical, closer to writing actual server-side logic with objects and I/O. Both were in a shared environment where you can run code. I ran my code. Use that. Seeing your output actually helps and it shows confidence.
Language: I did Python. Felt fine. Saw a friend do it in Go with no issues.
Difficulty overall: Honest answer is it sits between "solid Leetcode medium" and "hard if you don't see the pattern." It's not FAANG-tier grinding-on-HARDs territory, but you'll get exposed if you've only done easy problems. Two or three weeks of focused medium-level practice should be enough prep.
One thing they watch: Can you handle feedback mid-problem? Both my interviewers dropped hints at some point. The ones who freeze or argue with the hint probably don't pass.
Any questions on specific formats, happy to answer below.