CrackingWalnuts
Checked 3 August 2026. OpenAI does not publish its interview structure. Everything on this page is commonly reported by candidates, so treat it as a guide rather than a specification.
| Round | Duration | What it tests |
|---|---|---|
| Phone Screen | 60 min | Practical coding plus background. |
| Practical Coding | 90 min | Build a small system or library. |
| AI Infra Design | 60 min | Inference, RAG, evals, or training infra design. |
| System Design Deep | 60 min | AI-flavored end-to-end design with depth. |
| Behavioral / Taste | 60 min | Engineering judgement and decision-making. |
| Team Match | 45 min | Final team fit. |
IC4 is the senior level. The bar is "owns features end to end, has strong engineering taste, and can design AI infra without being walked through it".
The loop adds a dedicated AI infra design round and a deeper system design round on top of practical coding. The shift from IC3 is accountability: you are expected to make the call on inference-serving trade-offs, not just implement someone else's call.
Targeted guidance per round in this loop. Each section links to the matching track on the site. There is no shortcut list, the bar moves loop to loop, so build the pattern recognition and the stories from real practice.
Work the pattern templates until the recognition is instant. Trace one mock problem per pattern. Practice stating time and space complexity out loud before you write code.
OpenAI IC4 (Senior Software Engineer) mixes classical system design with AI-specific design. The AI engineering round is: AI Infra Design (60 min): Inference, RAG, evals, or training infra design.. The system design round is: System Design Deep (60 min): AI-flavored end-to-end design with depth.. Bar is owning a non-trivial design end to end, with capacity numbers, a clear data model, and one deep dive you can defend under push-back. Prepare in both the AI Engineering track and the System Design Problems track.
Senior bar. The rubric shifts from execution to ownership at scope, judgment, cross-functional impact, and being calibrated on what is hard. Every story needs alternatives considered and what you would do differently. Frame stories around safety-mindedness, mission alignment, and judgment under uncertainty about capabilities.