Your take-home now takes four seconds. Here is what to test instead

A take-home assignment used to answer one question: can this person do the work? Today it answers a different one: can this person paste a brief into an AI tool? Almost everyone can. So the fix is not to ban the tools or to drop the take-home. It is to change the task so that a finished answer is only worth something when the person behind it understood it.

What went wrong with the take-home?

Nothing went wrong with the idea. Asking someone to do a small piece of the real job is still one of the better ways to judge them. What changed is the cost of producing a good-looking answer. A brief like "build a small assistant that answers questions from these documents" now gets a working result, with a tidy README, in the time it takes to press Enter.

Hiring teams are feeling it. In a Robert Half survey of more than 2,000 US hiring managers published in March 2026, 65% said AI-enhanced applications make it harder to verify a candidate's skills, and 67% of HR leaders said reviewing them has slowed hiring. Greenhouse's 2025 AI in Hiring report found that 65% of hiring managers had caught applicants using AI deceptively.

The trouble is easy to state. If every hand-in is polished, polish stops separating people. A take-home that everyone passes looks, from the outside, exactly like a take-home that works.

Should you ban AI on the assignment?

No. There are two reasons.

First, you cannot enforce it. Nobody is watching the candidate at home, and tools that claim to detect AI-written work are wrong often enough that acting on them means turning away honest people.

Second, it tests the wrong job. Your new hire will use these tools from their first morning. A test that forbids them measures how someone works in a world your team no longer works in. The useful question is whether they can use the tools well and still know what they have made.

So allow AI, say so plainly in the brief, and change what you look at.

What should you test instead?

Three changes do most of the work.

1. Give every applicant their own material

When forty applicants get the same brief and the same files, one good answer can be shared forty times, and an AI tool gives all forty much the same result. Give each person a different version of the material (different documents, different numbers, different catches hidden in them) and an answer only fits the person it was made for. Copying stops being useful, and comparing people stays fair because the task and the marking are the same.

2. Mark the decisions, not the finish

A finished product is cheap now. Judgement is not. Build the marking around the points where a careful person and a careless one part ways:

• Did they notice that two documents disagree, and say which one wins and why?
• Did they check the result against the material, or trust the first output?
• Did they say what their work does badly?

A short written note from the candidate ("what I did, what I checked, what I would do with another day") often tells you more than the work itself.

3. Ask them about their own work

This is the step that makes the rest hold. Take fifteen minutes in the interview and ask about the hand-in they sent: why this approach, what happens if this input changes, where it breaks. Someone who did the thinking answers easily, with or without AI. Someone who forwarded a result cannot. We wrote a separate guide to the five follow-up questions that show whether a candidate understood their own work.

What does a good task look like now?

A task that survives AI has four properties:

1. It is a real piece of the job, not a puzzle. Real work has messy inputs, and messy inputs are where understanding shows.
2. The material is the candidate's own. Their answer cannot be borrowed.
3. It is short. Thirty to sixty minutes for an early step. Long unpaid tasks lose your best applicants first, because they have other offers.
4. The marking is written before anyone applies, and everyone is marked the same way.

Notice what is missing from that list: surveillance. No locked browser, no webcam, no detector. You do not need to catch anyone when the task itself only rewards understanding.

Where does this leave the live interview?

In a better place. A live coding round or a case interview asks a nervous person to perform on something they have never seen. An interview about work they have already done is a conversation about something concrete, which is kinder to the candidate and more useful to you. The take-home and the interview stop being two separate tests and become one: do the work, then explain it.

The short version

• Polished take-homes no longer separate candidates, because polish is free.
• Do not ban AI. You cannot enforce it, and your team uses it.
• Give each applicant their own material, mark their decisions, and ask them about their own hand-in.

This is how Wooble works: every applicant does a real piece of the job on material only they have, you get a ranked list with the reason for every place, and the interview questions come from each person's own work. You can see how it works or read our guide to work sample tests.