Work Trial
Sign up

For startups hiring engineers

A work trial that fits in an afternoon

Candidates get a real codebase, the product's data and the AI tools they'd use at work. They figure out what's going wrong, ship a fix and explain their decision. You get everything they did, ready to review with your team.

No account needed for either. You can also create your own workspace.

Length
30 min to 4 h
Setup
About 10 min
AI tools
Allowed
Decided by
Your team

What the candidate does

Think

Dig into the data

They query the product's data in SQL or plain English. The data holds the real answer, a convincing wrong one and plenty of noise, so you can see who checks before they conclude.

Build

Make the change

They work in a real, running codebase from the browser, with an editor, a terminal, the tests and a live preview. Claude Code and Codex are there if you allow them.

Show

Explain the decision

They present what they found and what they shipped, in a few slides or a short video, as if to the person who asked.

Why teams use it

What you find out about a candidate

Algorithm rounds don't look much like the job, and AI can answer them anyway. Paid work trials tell you a lot but take days to run. This takes a few hours and runs in the browser.

  1. 1Do they check the data before they trust a theory?
  2. 2Can they find their way around a codebase they've never seen?
  3. 3Do they ship a small, tested fix or rewrite half the app?
  4. 4How do they use AI? What do they hand off, and what do they double-check?
  5. 5Can they explain a trade-off to someone who wanted a different answer?

After they submit

Everything they did, in one place

Your team reads the work, rates each stage and decides together. Hidden tests run against their final code, and you can see whether their queries found the real cause or stopped at the obvious one.

See a sample review

Code changes

Every change against the starting code, file by file, with their commit messages.

Hidden tests

Tests they never saw, run on their final code when they submit.

Data queries

Each query and question, with the rows that came back.

AI usage

Their prompts to the assistant, Claude Code and Codex, and what the AI did in response.

Decision memo

What they recommend, what they ruled out and what they'd check next.

Presentation

Slides with speaker notes, or a video.

Task library

Start with a ready-made task

Each task comes with a sample company: a full codebase that runs, and 13 weeks of product data with a real problem in it. Use a task as it is, or change the brief, the stages and the time limit. Support for your own codebase is in the works.

Sample company · AI-first renters & home insurance · 2 tasks

Haven

A consumer insurer that sells every policy through an online quote flow.

Codebase haven-webFiles 45Product data 10 tables
Find out why mobile conversion dropped

Mobile quote-to-policy conversion dropped about a quarter in three weeks. The candidate has to separate a real cause from three plausible ones in the data, ship the fix in the repo, and present their recommendation.

Final round · 180 min
Find the missing partner quote starts

The partner channel shows zero quote starts on the dashboard, but partners say they send traffic and partner policies exist. Find the tracking bug, fix it, and stop it from happening silently again.

Screen · 60 min
Sample company · AI customer-support agent platform · 2 tasks

Relay

A B2B platform that puts an AI agent in front of customer support teams.

Codebase relay-consoleFiles 41Product data 8 tables
Decide what the agent handles next, then build it

Six in ten conversations still go to a person. The candidate decides which request type to automate next using the data, builds the playbook in the agent runtime, and defends the choice against the loudest customer request.

Final round · 180 min
Find out why refund hand-offs tripled

Since mid October the agent hands almost every refund to a person. The candidate finds the release that did it, proves it in the data, fixes the bug and adds the test that would have caught it.

Screen · 60 min

For hiring teams

Set up in about ten minutes

  1. Pick a task.

    Use it as it is, or change the role, the brief, the stages and the time limit.

  2. Run the setup checks.

    We start a fresh sandbox and confirm the task works: the data answers, the tests run and a reference solution passes.

  3. Send invite links.

    See who has opened theirs, who is working and who has submitted.

  4. Review and decide together.

    Each reviewer rates the stages and leaves notes. Then mark the candidate advance, hold or decline.

For candidates

Nothing to install

  1. Open the invite.

    See the company, the stages and how long it takes before the timer starts.

  2. Dig into the data.

    Query it with SQL or plain English, with or without AI, and write up what you find.

  3. Make the change.

    Edit the code, run the tests and the app, and commit, all in a private sandbox.

  4. Present and submit.

    Slides or a video. One button submits everything.

Try it from both sides