The instructions say you may use AI. Now what are they assessing?
It is tempting to treat that sentence as the end of the preparation problem. Get the tool to produce the deliverable, check that it looks finished, and submit.
For an AI-allowed take-home assessment, prepare to explain what you chose, how you checked the result, and what you still do not know. Permission to use a tool does not tell you which parts of your work the employer needs to understand.
A published example, with a clear boundary
In its January 2026 account of redesigning a performance-engineering take-home, Anthropic describes an assessment where AI assistance was explicitly allowed. The team still needed to distinguish candidates' abilities as model capabilities improved.
That is one team's published experience. Anthropic's general candidate guidance, last updated in July 2025, says assessments and live interviews should be completed without AI unless the company indicates otherwise. The particular assessment was an exception, not blanket permission.
The preparation lesson is to read the actual instructions. Do not infer the rules from the employer being an AI company, a social post, or another candidate's experience.
Clarify the task before choosing the tool
Check what the employer has specified: the deliverable, time limit, permitted assistance, submission format, and any explanation of your process. If the instructions leave something important unclear, ask before starting.
“The brief says AI assistance is allowed. Could you confirm whether there are tool or data restrictions and whether you want a description of how it was used?”
The answer may be different across employers and assessments. Follow that answer rather than creating extra documentation the task does not require.
Then write down the question the deliverable needs to answer. A finished-looking report or application is not enough if it solves a different problem.
Keep your decisions understandable
When the rules permit working notes, keep a short record of the choices you may need to explain. What did you ask the tool to help with? Which suggestion did you accept? Which did you reject? What check changed your mind?
This does not require saving every prompt. A transcript can be long while leaving the important decision obscure. Keep whatever the assessment requires, and make sure you understand the decisions behind the submission.
Use only data and tools permitted for the task. If the brief includes restrictions, allowing AI does not remove them.
A fictional follow-up discussion
Imagine a software assessment that explicitly allows AI and asks for an importer that handles repeated records. The candidate uses an allowed assistant to propose an approach, then notices that it treats any matching customer name as a duplicate.
Here is a fictional explanation in the follow-up:
“I used the assistant to sketch an initial approach, but I rejected its rule for identifying duplicates. The brief supplied a record identifier, while a customer name could legitimately appear more than once. I changed the comparison to use the identifier and checked a repeated record, two distinct records with the same name, and an input without the identifier. I documented the remaining missing-identifier behavior because the brief did not define it. I would confirm that requirement before treating the importer as ready for real data.”
The example is illustrative, not an employer's reported question or a complete software design. Its value is the visible decision: the candidate rejected a plausible suggestion against the actual requirement, checked relevant cases, and named an unresolved boundary.
Stop Guessing. See Exactly How You Sound.
Reading about interviews won't help you. Speaking out loud will. Your first few interview practices on us.
Get specific feedback on what's working and what's killing your chances. Know your blind spots before the real interview.
Prepare beyond the finished artifact
For this editorial exercise, a weak explanation is “the AI built it and the tests passed.” It does not tell the listener whether you understand the requirement or the checks.
A credible explanation connects your choices to the brief and shows how you examined the result.
A stronger explanation also addresses an alternative you rejected and a limit you would resolve before extending the work. That is Revarta's preparation rubric, not a claim about the employer's private scoring.
Practice explaining one important part without your notes. If you cannot do that, return to the work rather than generating a more polished explanation of something you do not understand.
Rehearse the review conversation
Use the difficult decision question to explain a choice from work or a practice project you can discuss. Identify the requirement, the options, your check, and the decision. Describe AI assistance accurately when relevant to that experience.
Then use the follow-up question exercise. Ask why you chose that option and what would change under a different requirement. The personal contribution guide helps separate your judgment from the output a tool or teammate produced.
Common questions
How do I prepare for a take-home assessment that allows AI?
Read the specific assessment rules, identify what you must deliver, and plan time to check the work. Use permitted tools within those rules. Prepare to explain your decisions, how you verified the result, and what remains incomplete or uncertain.
Does an employer allowing AI on one assessment mean I can use it in every interview?
No. Permission can differ by role and stage. Follow the instructions for the particular assessment or interview and ask the recruiter when they are unclear. One team's published example does not establish permission elsewhere.
What should I explain in a follow-up about an AI-assisted take-home?
Explain the problem, your important choices, the AI assistance you used within the rules, and how you checked the output. Be ready to discuss a rejected option, a failure case, and a limitation without relying on a generated script.
The useful preparation is being able to explain the work you submitted, including the parts that are unfinished.
