Developer roles have changed.Your tests should too.

Hire developers who move fast with AI but don't take its word for it. In 25 minutes your candidate reviews an AI agent's pull request against a spec, AI allowed. You see if bugs, weak design or security holes get through.

First test free. No contract.

The test01

Your candidate reviews an AI agent's pull request

A short spec on one side, the agent's code on the other. They comment on the lines that need changes, then approve or request changes.

Pull request by an AI coding agent: feat(auth): self-service password reset

Caught of 3 in this file

src/routes/passwordReset.ts

1import { Request, Response, Router } from "express";
2import { logger } from "../lib/audit";
3import { asyncHandler, HttpError } from "../lib/http";
4import { RateLimiter } from "../lib/rateLimit";
5import { confirmPasswordReset, PasswordResetDeps, requestPasswordReset } from "../services/passwordReset";
6
7const HOUR_MS = 60 * 60 * 1000;
8
9const normalizeEmail = (email: string) => email.toLowerCase();
10
11const ACCEPTED_MESSAGE = "If an account exists for that address, we've emailed a link to reset the password.";
16 unchanged lines
28
29 const email = typeof req.body?.email === "string" ? req.body.email : "";
30 const input = { email, ip: req.ip, baseUrl: `${req.protocol}://${req.get("host")}` };
31
32 res.status(202).json({ message: ACCEPTED_MESSAGE });
22 unchanged lines
55
56 router.post("/request", asyncHandler(request));
57 router.post("/confirm", confirm);
58 return router;
59}

Sample review. The candidate is fictional; the task and the planted problems are real. See the report

What it shows02

Developers who lead the AI.

AI tools are allowed, like on any normal workday. The test shows what the developer does with the agent's work.

Requirements
They check the code against the spec and notice what is missing or different.
Bugs
They find logic errors, race conditions and broken edge cases before they reach production.
Design
They spot duplicated logic and code that ignores the patterns already in the codebase.
Security
They catch leaked tokens, missing checks and input the code should not trust.

The report03

You get a report in plain English

  1. What they caught, in their own words.
  2. What they missed, and why it matters.
  3. Questions for the interview, built from their review.
  4. How a plain AI review did on the same task.
Read the sample report

Code review test result

Sam Lindqvist (fictional)

9 to 10out of 10

Caught nearly every planted problem and explained why it matters, including one that needed a careful read of the spec.

Caught 5 of 6

  • Reset link origin taken from the request Host header
  • Existing sessions are not revoked after the reset
  • Confirm handler not wrapped in asyncHandler
  • Token lookups can't use the index (user_id leads it), and hashes aren't unique
  • Single-use test cannot fail

Missed 1

  • Re-implements normalizeEmail with different behaviour

    Instead of reusing the project's existing helper for cleaning up email addresses, the code adds its own slightly different copy. Two versions of the same rule drift apart, and here the copy forgets to strip spaces, so the per-address limit can be dodged by adding a space.

For the interview

Walk me through src/routes/passwordReset.ts around line 5. Does it do what the spec asks? What would you change?

For comparison: a plain AI review of the same task caught 61% of the planted problems.

Sam Lindqvist found 1 of the 2 problems a plain AI review usually misses.

How it works04

How it works

  1. 01

    Send the test

    Pick a task by role and level and enter the candidate's email. Sending is free.

  2. 02

    They review the pull request

    About 25 minutes in the browser, AI tools allowed. No account and no code to write.

  3. 03

    Read the report

    Graded in minutes. You pay only when a result is graded.

Try it yourself05

Take the test yourself first.

25 minutes, no signup. You see what you caught and what you missed.

Start the challenge

Why Majken06

“When anyone can vibe code, it's harder to tell who's a good hire. I'm a senior developer who hires developers, so I built the test I wish I'd had.”

Julia Wallin, founder

Read the full note

Pricing07

Pay when a result is graded

Sending a test is free, and a test nobody takes costs nothing. Prices in euros.

First test
€0
Free, no card needed
Single test
€19
One graded test
10 tests
€129
€12.90 a test, credits never expire
Team
€99
A month, up to 30 tests
See pricing

Questions08

Questions

Do you test prompting?

No. The test looks at what a developer does with code an AI agent has already written: whether they check it against the requirements and find the bugs, design problems and security holes in it. How they use AI tools along the way is up to them.

What about candidates who use AI?

AI tools are allowed, as on a normal workday. The candidate can note which tools they used, and the report shows it. Each candidate gets a different mix of planted problems, and the report shows how a plain AI review scored on the same task. If a review closely matches a plain AI review or another candidate's, the report says so and suggests a follow-up question.

How do you handle candidate data under GDPR?

The candidate sees who invited them, that the result goes only to you, and when it is deleted. We keep their review and result for 180 days after grading, then delete them. A candidate can ask us to delete their data sooner by writing to hello@majken.ai. A person on your side makes every hiring decision.

Which languages and stacks do the tests cover?

TypeScript and Node.js backend tasks today. Python and React frontend tasks are coming.

How long does the test take?

About 25 minutes. There is no hard timer, and the time they spend is recorded. The candidate can pause and come back while the link is valid, which is 7 days.

All questions