Companies rolling out AI-enabled interviews have landed on two distinct formats, and they're easy to confuse because both involve an AI assistant. But they test different skills, they're prepared for differently, and walking into the wrong one with the wrong prep is a fast way to underperform.
The first step in preparation is knowing which you're facing — and if the invitation is vague, asking is a legitimate move.
The structured format: a panel with a built-in AI
In a structured interview, the company provides everything: the environment (often CoderPad or HackerRank, sometimes their own in-browser editor), the problem panel, and a built-in AI assistant. You don't configure anything. You work in their sandbox.
The problems typically follow a three-phase arc:
- Bug-fix. You're handed a codebase with a known bug and the AI is available to help you find and fix it. This tests orientation — can you read unfamiliar code and direct the AI to the right area?
- Feature. Add a feature to the existing code. This tests planning and clean extension — do you fit your addition into the architecture, or bolt it on?
- Scale. Extend or optimize for scale. This tests whether you can reason about complexity and steer the AI toward the right approach rather than accepting the first thing it offers.
What makes structured interviews hard is the unfamiliar, constrained AI. You can't bring your own tool, so you're driving a model you may not have used, in an editor you may not know, with whatever shortcuts it has. The skill being tested is adaptability: can you stay effective with an assistant that isn't yours?
The open-ended format: bring your own environment
In an open-ended interview, you share your screen and use your own editor, your own AI tools, your own terminal. There's no problem panel and no built-in assistant — the prompt is typically a greenfield build or a deep design-and-implement exercise, and you drive your entire stack.
Because you bring your own environment, the skill being tested shifts. It's less about adapting to a foreign sandbox and more about:
- How you actually work. Your AGENTS.md, your test commands, your conventions — the interviewer sees your real workflow, not a rehearsed sandbox dance.
- How you architect under time. A greenfield build with an AI is a design-and-implement exercise. Do you set up the skeleton and steer the AI to fill it, or does the AI's first output become your architecture?
- How you handle your own tooling failing. When your model hallucinates or your setup breaks, do you recover calmly or flail?
The trap in open-ended interviews is over-rehearsing a canned flow. The interviewer can see you switching windows, running commands, and thinking in real time — authenticity matters more than polish.
What each actually measures
Here's the honest breakdown of what interviewers tell each format measures:
| Skill | Structured | Open-ended |
|---|---|---|
| Reading unfamiliar code | Strongly tested (bug-fix phase) | Tested (you must orient) |
| Adapting to a constrained AI | The core test | Lightly tested |
| Your real workflow & tooling | Not visible | The core test |
| Architecture & design under time | Tested (scale phase) | Strongly tested |
| Verification & debugging habits | Tested | Tested |
| Communication | Tested | Tested |
Notice what's in the corners: structured is the adaptability test, open-ended is the authenticity test. They're different strengths, and a candidate who shines in one can look ordinary in the other.
How to read the room
If you're unsure which you're facing, the invitation usually tells you — "a CoderPad session with an AI assistant" is structured; "please have your editor and tools ready" is open-ended. If it's ambiguous, asking the recruiter is completely reasonable and reads as preparation, not weakness.
The one decision that matters: prepare in the format you'll face. Structured practice is about getting comfortable with a constrained, unfamiliar AI in a sandbox — the workflow of plan, prompt, verify inside someone else's environment. Open-ended practice is about rehearsing your own stack until driving it is automatic, and being able to narrate the reasoning behind how you build.
If you have time for only one thing
For structured, do timed runs in a CoderPad-like sandbox until the constrained AI feels normal. For open-ended, rehearse your real environment end-to-end — AGENTS.md, test command, clean project — until the mechanics are invisible.
The short version
Structured AI interviews test whether you can stay effective with an assistant that isn't yours; open-ended interviews test whether your real workflow holds up when it's the thing being graded. Both reward the same underlying habits — plan before you prompt, verify everything the AI produces — but they're prepared for differently. Know your format, and practice in it.