Here’s something that happened in early 2025: a large number of companies quietly updated their interview policies to explicitly address AI tool use during take-home assessments. Some banned it. Some allowed it with disclosure. Some said nothing and then flagged candidates who used it, after the fact. The norms around AI in coding interviews are still genuinely unsettled, and that matters for how you prepare.
I want to go through some of the ways AI has changed the coding interview process, and also some things that haven’t changed at all.
how LeetCode-style prep has changed
The basic workflow of problem-solving practice looks different now. A lot of people are using ChatGPT or Claude to explain why their solution is wrong, instead of staring at the solution tab on LeetCode for 20 minutes. This is genuinely faster. You can ask “why does this dynamic programming approach fail on this input” and get an explanation at whatever level of depth you need.
The problem is that explanations aren’t the same as understanding. I’ve seen this in how people describe their solutions during interviews, they can get to the right answer on medium problems, but when an interviewer asks “why not a BFS here?” they hesitate in a way that suggests they know what the answer is, not why. That distinction comes up a lot.
There’s also the issue of problem supply. AI can generate infinite new LeetCode-style problems, which sounds useful, but the difficulty calibration is inconsistent. GPT-4 will sometimes give you a problem it describes as “medium” that’s actually quite hard, and vice versa.
live coding interviews and AI assistance
Most companies doing live coding interviews (CoderPad, HackerRank proctored, Karat, etc.) explicitly prohibit AI assistance. This isn’t controversial, it’s the same as saying you can’t have a friend whispering answers over Discord. So the live interview itself hasn’t changed much from a rules standpoint.
What has changed is the spread of ambient AI tools. Some candidates run AI assistants on a second screen. Some don’t. Whether companies can detect this varies by platform, HackerRank and Codility have some detection capabilities, but they’re not foolproof.
I’d say most people who use AI assistance during live interviews are making a mistake, not an ethical one, but a practical one. If you can’t explain your code, the technical screen is probably not the last technical question you’ll face before an offer. Getting caught out later is worse than a slower performance now.
what the data says about interview formats in 2026
The Stack Overflow 2024 Developer Survey found that 62% of developers are now using AI tools as part of their daily workflow. Companies have updated their technical assessments partly in response, there’s been a visible shift toward take-home projects, pair programming formats, and system design discussions, and away from pure algorithmic challenges. The Bureau of Labor Statistics projects software developer employment growing 25% through 2032, which means companies are still hiring, just with different expectations around what candidates can do on day one.
This trend has some real momentum. Companies like Shopify, Linear, and a growing number of mid-size startups now weight system design and “show us something you built” more heavily than LeetCode problems. That’s not universal, FAANG-tier companies still lean algorithmic, but the distribution is shifting.
the parts that still require real practice
Algorithmic thinking on time pressure. Graph traversal, dynamic programming, binary search variations, AI can help you understand these, but you need to be able to write the code under 30-minute time constraints without AI. That requires actual repetition.
Communication while coding. This is probably the skill most candidates underinvest in. Talking through your approach before writing any code, narrating tradeoffs, flagging assumptions out loud, none of this gets practiced by solving problems alone. Pair programming practice or mock interviews are the only real substitute.
Debugging under pressure. A lot of candidates can write a solution but fall apart when the interviewer introduces an edge case that breaks it. Debugging in front of someone while explaining your reasoning is a specific skill. It doesn’t come from LeetCode grind.
AI interview preparation tools: where Craqly fits
There are a few categories of AI tools now aimed at interview prep. Flashcard-style systems (Anki with generated decks), code explainers (ChatGPT, Copilot), mock interviewers (several startups including Craqly), and real-time copilots designed to sit alongside you during an actual interview.
Craqly is specifically built for live interview support, it runs during the interview itself and can surface hints, relevant documentation, or clarifying context without taking over the solution. Whether you use a tool like this depends on the company’s stated policy, which you should check before using anything.
a note on the anxiety problem
This is something I don’t see written about much. A lot of engineers are genuinely anxious during coding interviews in a way that doesn’t reflect their actual ability. The whiteboard context, the time pressure, the interviewer watching, it produces performance that’s much worse than the same person writing code at their desk.
AI-powered mock interviews help here, not because AI is better at mimicking a human interviewer, but because practice volume matters. Doing 47 mock problems in interview conditions is better preparation than doing 200 problems casually. Context matters as much as content.
The gap between what people can do at their desk and what they produce under interview conditions is real, and it’s probably the main thing worth working on in 2026, more than knowing one more algorithm.