AI-assisted engineering assessment

Show how you think, not how fast you type

AI is allowed. What gets evaluated is what you add on top of it.

HR New Way puts you on a real engineering task with an AI assistant at your side, and the platform records evidence of thinking — it does not try to catch you out.

Why it's built this way

Most hiring tests measure what AI can already do — syntax, boilerplate, a working function on the first try. This one measures what you add on top of that: how you frame the problem, where you push back on the AI's output, how you verify it, and how you explain the trade-offs. AI writes the code faster than any of us now; the part that still tells you who to hire is everything around it.

How it works

1
Work a real task with an AI assistant. A Workbench Assistant sits alongside you inside the scenario; an Interviewer runs it and may check in as you go.
2
The process is recorded. Prompts, replies, edits — logged for private, internal analysis. No camera, microphone or screen proctoring.
3
You get a growth report. Verdicts across six dimensions, each backed by evidence from what you actually wrote — not a score pulled from nowhere.

What gets measured

Problem Framing: Do you understand what's actually being asked before you start typing?
AI Collaboration: How you prompt, steer and push back on the assistant.
AI Output Review: Whether you catch what the AI got wrong before it ships.
Debugging & Verification: How you check that what you built actually works.
Communication & Ownership: Do you explain the trade-offs, or just hand over code?
Risk & Production Readiness: Would this survive contact with a real system?

Sample growth report

Example only — an invented candidate, an invented run. This is the shape of what you get, not a real result.

Example — not a real run

AI Output Review

Adequate

Evidence

"The suggested query looks right, but I want to check it against the actual index before I trust it — running EXPLAIN now."

What to develop

Name the specific risk before you accept an AI suggestion, not just after something breaks.

More than one reviewer

Every run is reviewed several times, not once. Where the reviews agree, you see a single verdict. Where they differ, you see all of them side by side, each with its own reasoning — a disagreement is information, not something smoothed into an average.

What we don't do

Not an AI detector — using AI well is the point, not a violation.
No proctoring — no camera, microphone or screen capture.
The platform validates evidence — it does not recommend or rank candidates to anyone.
Nothing leaves the platform without your consent. The raw log is not a public profile.

AI is allowed, ownership remains yours.