RecruitWhiz
Hire people who use AI and still own the work.
Candidates work a real scenario from your job with their own AI tools. You see how they worked, what they caught, and how far they got past the model. Then a person decides.
- 0:14T4 caughtAsks the stakeholder about renewals in progress
- 0:31T1 missedCopies an existing filter into the new resolver
- 0:38T3 caughtChecks the repo for an API her AI suggested
- 0:52T2 missedAdds a column in a way that locks a large table
- 1:24Submits with a five-step rollout plan
Why now
The job changed. The coding test did not.
Engineers already work with AI coding tools in their own IDEs. A test that bans them measures a job that no longer exists, and a test that only checks the final code cannot tell judgment from a pasted answer.
| Topic | Classic coding screen | RecruitWhiz |
|---|---|---|
| AI during the test | Banned, or ignored | Allowed in Part 1, and measured |
| What gets scored | The final code | How they break the work down, verify, ask, and defend it |
| Where they work | A browser editor | Their own IDE, connected to an isolated environment |
| The task | A puzzle from a library | A scenario generated from your job description |
| The live interview | Starts from zero | Starts from what they did and what they missed |
How it works
Two parts and a decision.
- 01
Your team sets the bar
About 30 min of your time
Upload the job description. HR fills in a five-minute form and the technical interviewer talks with a copilot for 15 to 20 minutes. Nothing is built until a person approves what gets measured.
- 02
Part 1: the work, with AI
Async · about 90 min
The candidate connects their own IDE to an environment we run and works a scenario built for the role. Planted traps and a simulated stakeholder show how they verify, ask, and decide.
- 03
Part 2: the conversation, without AI
Live · 45 to 60 min
The interviewer asks the base questions every candidate gets, plus questions generated from this person’s submission.
- 04
A person decides
Every time
The report shows the scores, the evidence behind them, and the comparison with a frontier model working alone. It recommends. Someone on your team records the decision.
The report
Every score sits next to what a frontier model does alone.
Before a test reaches candidates, a frontier model takes it with no human. The report shows the distance between that baseline and the person, and the planted traps that explain it.
- T1Tenant data leakMissed
- T2Blocking migrationMissed
- T3Invented APICaught
- T4Unstated exceptionCaught
What we capture
We record the work. That is all.
Recorded
- Code diffs
- Commands
- Test runs
- SQL queries
- Calls to our tools
Never recorded
- Their prompts
- Their screen
- Files they read
- Their machine
Join the pilot waitlist.
The first pilot covers backend and fullstack TypeScript roles. We will email you when a seat opens.