Over the last year we have seen massive shifts in how organisations are leveraging AI within development environments which has led to a lot of changes in the way interviews are conducted. Whilst some firms will still heavily leverage traditional coding interviews based on leetcode style problem solving questions, a number of fintechs and start ups have incorporated rounds designed to assess a candidates usage of AI.
The first question with any take home assignment or live coding interview is now “Can I use AI?” – and rightly so, after all it is designed to assess someone’s on the job skills , and if it is now being leveraged across the board, then it is reasonable to suggest it should be leveraged, yet there is still a disparity between those that encourage it vs those that want to maintain the assessment of problem solving capabilities without it.
A few months ago, a hiring manager we work with told us “I don’t care if a candidate uses AI to write the code anymore, I want to ensure they can own and defend it though.”
That’s the shift, in one sentence. We’ve been sitting close to a lot of these interview loops lately, and hope by sharing our findings, it will help development candidates prepare better for these assessments.
Take-home assessments
Take-home assessments had widely been abandoned – as they were time intensive (both for candidate & for marking) and added significant bottlenecks to the interviewing process. On the flip side, they offered a strong opportunity to give a candidate a problem that closely mimicked their day to day work. Now with AI – take home assignments that encourage AI usage are common as they have significantly reduced the time taken.
Take caution however, as these are now almost always followed by some form of pair programming interview – designed to see if the candidate can defend the reason certain decisions were made, and equally for them to code extensions to their assignments that follow the same structure & patterns.
“Prompt the agent, keep the leash”
The other type of live AI interview we are seeing involved candidates interacting directly with an AI agent , with an interviewer acting as an observer to the process. Candidates getting dropped into a half-baked, unfamiliar codebase alongside an AI agent, with a task to solve the problem by directing the agent. What’s actually being assessed has almost nothing to do with prompting skill in the flashy sense. It’s about control. Good candidates use the agent to validate hypotheses they already have, “confirm this function is doing what I think it’s doing”, rather than outsourcing the entire thinking process e.g “figure out what’s wrong with this.” The moment a candidate starts assuming that whatever the agent suggests is correct without pushing back , the critiques will start coming in . It’s also assessing a candidates’ capability to control scope creep. When the agent cheerfully offers to refactor three unrelated files or bolt on a feature nobody asked for, the strongest candidates will be able to say “No, keep to what I asked”. The idea of these rounds is to showcase you are still in control of the problem solving process, leveraging AI as your own personal assistant along the way.
The importance and rise of the “Requirement-gathering” skill
This is probably the biggest mindset shift we’re seeing from managers, and it’s easy to miss if you’re prepping purely on the technical/tools side. The feedback we have been getting is that the bottleneck in AI-assisted development isn’t generating code, it’s translating an ambiguous business need into something specific enough for an AI to act on correctly.
That’s just requirements gathering, but companies now want to assess this in real time, out loud, under interview conditions. Candidates who ask sharp clarifying questions before touching the keyboard, are outperforming candidates who jump straight into “let me just try something and figure it out along the way.” Communication has quietly become one of the highest-leverage skills for a developer on the market.
Our honest take, for candidates reading this:
If you’re prepping for a role where AI tools are part of the process, stop practising purely how to prompt well . Interviewers aren’t trying to catch you using AI. They’re trying to find out what’s left when the AI is wrong, half-right, or quietly doing too much. That’s the actual skill on show. Do you remain in control , and is your work a reflection of your own ideas being put into deliberate actions.
The landscape is moving quickly and we are seeing new formats and trials of these rounds coming in every day. Interview formats from even six months ago are already starting to look dated so keeping an eye out for new trends is also important.
