Stack Overflow's 2025 Developer Survey found that 84% of respondents use or plan to use AI tools in their development work, and 51% of professional developers use them daily. Hiring has caught up unevenly. Some companies now expect you to use Copilot, Cursor or Claude in the interview. Others treat it as cheating and end the process the moment they notice.
Find out which kind you're facing before you type a line.
Key Takeaways
- Ask for the rules before the interview, in writing. Never assume.
- When AI is allowed, the real test is your judgment: what you ask for, what you reject, what you check.
- When AI is banned, secret use is the fastest route to a rejection. Interviewers spot it.
- Practice both ways. Many companies, Ruzora included, test with and without AI.
How to tell what's allowed
Ask. Email the recruiter: "Are AI assistants like Copilot or ChatGPT allowed in the coding round? If yes, are there limits?" It's a normal question now, and the answer protects you.
| Format | Common rule | What they're measuring |
|---|---|---|
| Live coding in a shared editor | Often no AI | How you reason from scratch |
| Live coding in your own IDE, screen shared | Sometimes allowed, always watched | How you direct and check AI |
| Take-home project | Varies, read the instructions | Final code quality and your explanations |
| Async or recorded assessment | Usually stated in the rules | Depends on the test |
If the instructions say nothing and you can ask, ask. If you can't ask, disclose what you used in your submission. Our guide to passing a take-home coding challenge goes deeper on that.
When AI is allowed, make your judgment visible
The interviewer can't grade what happens in your head. Say it out loud.
- Plan, then prompt. Say what you're about to ask for and why. "I'll have it scaffold the handler. I'll write the validation myself, because that's where the bugs live."
- Read what you accept. Walk through each suggestion before keeping it, and point out what you'd change.
- Reject bad output on purpose, and say why. Catching a wrong import or an off-by-one tells the interviewer more than ten accepted completions.
- Test it. Ask for edge-case tests, then add one the AI missed.
- Know when to stop prompting. Three failed tries at the same prompt means write it yourself.
The same survey found more developers distrust the accuracy of AI tools (46%) than trust it (33%). Interviewers share that suspicion, so showing that you verify is the signal they're looking for.
When AI is banned
Don't use it. A second monitor, a phone below the camera, a tool reading your screen: interviewers have seen all of them. The tells are familiar. Long pauses followed by perfect code. Eyes drifting off-screen. No ability to explain a line when asked to change it. One follow-up question usually ends it.
Prepare for no-AI rounds by coding without autocomplete for a couple of weeks. Most of us have lost some recall of syntax and standard library names, and that's fine. Say "I don't remember the exact method name. I'd check the docs, it's something like..." Interviewers accept that. Silence worries them.
For how employers think about all this, read rethinking technical interviews in the AI era and how companies vet developers who use AI coding tools.
A Concrete Version
A composite. Lucia is a Node backend engineer in Montevideo with a 60-minute live round where AI is allowed. The task: an endpoint that imports a CSV of orders and rejects bad rows.
- Minutes 0-5. Two questions (how big can the file get, and what counts as a bad row?), then she states her plan.
- Minutes 5-15. She asks the assistant for an Express route and a CSV parser. She keeps the route and rejects the parser because it loads the whole file into memory. Files can reach 200 MB, so she streams it.
- Minutes 15-40. She writes the validation herself and uses AI for a date-parsing helper. It reads "03/04" as March 4, but the sample data is day-first. She catches it, fixes it, and says so.
- Minutes 40-55. She asks for tests and gets eight. Two only test the library, so she deletes them and adds one for an empty file.
- Minutes 55-60. She lists what she'd do with more time.
The interviewer's notes focus on the rejected parser and the date bug. Those two moments earned the next round. The AI saved her maybe 15 minutes of typing.
The Honest Counterpoint
AI-allowed rounds can make average engineers look better than they are, and some companies have gone back to strict no-AI rounds for that reason. You don't control the format, so don't prepare for only one. If you lean on AI for everything, a no-AI round will expose it, and so will the job.
Disclosure isn't always rewarded either. A few interviewers still judge any AI use harshly, even when the rules allowed it. Be honest anyway. A company that punishes allowed, disclosed use would probably be a hard place to work.
Frequently Asked Questions
Is using ChatGPT in a coding interview cheating?
Yes, if the rules ban it or you hide it. If the rules allow it and you use it openly, it's part of the test.
Does Ruzora allow AI tools in its assessment?
We test both ways, with and without AI tools, because the engineers our clients want can work either way. The instructions for each step say which applies. Start from our open roles.
Should I list AI tools on my CV?
Yes, if you use them seriously. Name the tools and one thing you built faster or better with them. Skip "prompt engineering" as a standalone skill unless you're applying for AI roles.
The Bottom Line
Ask the rules, follow them, and when AI is allowed, make your judgment the thing the interviewer notices. Practice both modes. Then apply for a role, for example Node.js backend engineer or fullstack AI engineer.
---
Roberto Espinoza is CEO of Ruzora, which places senior LATAM engineers with US startups. Browse open roles.
