Hiring entry-level software developers in the era of AI

What is advice on entry-level positions evaluation in spite of these challenges?

If you don't want candidates to use AI to formulate their response, you need to ask questions that must be answered live.

On a video call or in person, for example, show them a problem and ask them to verbalize how they would approach a result.

Take home coding tests have always had the opportunity for cheating. Or they were so complex people who were getting interviews just skipped them. AI makes it even easier to cheat.

You need to look for other ways to evaluate entry level candidates. Coding tests filter the candidates, but there is no guarantee that they filtered them in a way to select the best candidates.

If this is an entry level position realize that few of the people who are hired at this entry point will be there in 5 years. If you do pick the best (and you will never know if that is true) they may stay forever or leave after 6 months. The same is true if you pick anybody else.

Set your goal to avoid picking people that have little chance of working out. If you are only selecting 10 people for final interviews, your interview process should be able to find 4 or 5 that would be good enough to be hired.

So how do you get to 10 people for the final interview. If there are less than 10 candidates. interview everybody that is a maybe: they have a decent resume, and passed the phone screen.

If there are a lot more than 10 take the first 10 that are at least maybes. Then turn off the stream of applications. Getting hundreds of applications, and then taking a month to move to the next stage doesn't help the candidates, and doesn't increase the quality of the hires.

You now are in a position to reexamine what skills are needed in junior programmers. AI has altered things so that coding skills are not that important. Far more important are debugging skills, understanding requirements, being able to write valid documentation, and more.

Instead of offering a "take home" test, offer an online, timed test. Such a test can include ways to identify whether a person pasted something in. Such a test can also ask for skills other than raw coding such as identifying bugs in code, simple design flaws, and "what is missing" type questions which are far more difficult for AI to answer. You can be looking for comprehension of code blocks instead of writing them.

If the test no longer works, you need a new test. Like Joe Strazzere says, you can do live questions/exercises instead which would be the best solution without changing your criteria. However, consider changing your criteria. If an entry level dev can use AI to solve the simple interview test, then who cares if they can do it without AI? They'll have it available to them when they come to work for you, so perhaps you should try to evaluate how good they will be in that scenario.

Instead, ask a harder question with the knowledge that they will need to know enough to get the right answer even with AI assistance. Or start prioritizing candidates that have built projects they can show or have meaningful coding experiences to talk about. Ask them something subjective, like their favorite piece of knowledge they've learned about programming or something similar that could reveal a deeper understanding.