I've been conducting some interviews for a backend engineer/founding engineer role, helping a friend with her new AI startup over the last couple of weeks.
Some of my observations in the post-AI world if you're preparing for an interview 🧵
Mehul Mohan
@mehulmpt
1. More personalised interviews, fewer typical coding questions
I have stopped asking questions that are like "fix this bug" or "what is this code doing wrong" coding snippets.
They are a useful measure, sure, but "fixing" is no longer your job. Figuring out that the thing needs to be fixed from broader environment data is.
Now I start the interview with their background, what tech stack you have used in the past, trying to get out your "strong opinions" and reasoning, and that builds on to the next whiteboarding, brainstorming, and deepdiving into internals.
I have stopped asking questions that are like "fix this bug" or "what is this code doing wrong" coding snippets.
They are a useful measure, sure, but "fixing" is no longer your job. Figuring out that the thing needs to be fixed from broader environment data is.
Now I start the interview with their background, what tech stack you have used in the past, trying to get out your "strong opinions" and reasoning, and that builds on to the next whiteboarding, brainstorming, and deepdiving into internals.
1:24 PM UTC · Aug 27, 2026 · 15.5K Views
1013220
Mehul Mohan
@mehulmpt
After this, based on their background and interests, I'll ask them to whiteboard a solution to a problem that the real company faces or will face in the future.
I ask the candidate to think out loud. I want to see your "reasoning traces" as a human. I am looking for intelligent questions at this point.
That tells me if you're better at *thinking* a problem compared to AI and if you are able to absorb more context than just the problem at hand.
I ask the candidate to think out loud. I want to see your "reasoning traces" as a human. I am looking for intelligent questions at this point.
That tells me if you're better at *thinking* a problem compared to AI and if you are able to absorb more context than just the problem at hand.
1:24 PM UTC · Aug 27, 2026 · 15.6K Views
1110410
Mehul Mohan
@mehulmpt
2. Stop wasting time
One of my complaints has been people are not mindful of time. It's a 30-minute interview slot which I clearly tell upfront is dividied into 5-6 minutes introduction and rest of the 25 minutes deep diving into the problem I have for you.
I want to quickly know your achievements and hardest problems you faced in first 5 minutes, I want to see if you are really a solid deal in the next 25. It's more than enough time.
I have talked to candidates who go on a talking spree and tell me everything about their life for 20 minutes straight. I know your life is interesting now, but it doesn't help me at all in telling if you're technically good.
One of my complaints has been people are not mindful of time. It's a 30-minute interview slot which I clearly tell upfront is dividied into 5-6 minutes introduction and rest of the 25 minutes deep diving into the problem I have for you.
I want to quickly know your achievements and hardest problems you faced in first 5 minutes, I want to see if you are really a solid deal in the next 25. It's more than enough time.
I have talked to candidates who go on a talking spree and tell me everything about their life for 20 minutes straight. I know your life is interesting now, but it doesn't help me at all in telling if you're technically good.
1:24 PM UTC · Aug 27, 2026 · 14.6K Views
3010010
Mehul Mohan
@mehulmpt
3. Please stay up to date with AI for AI roles
Almost every role requires you to now use AI in some capacity to improve your output quality or speed.
If I ask you what AI model the company should use to solve this (hard) problem, given pricing is not a big constraint, and you tell me it is Gemma 3, a very very old AI model, self-hosted because you like privacy, that is not a great answer and shows me you are totally out of touch.
Almost every role requires you to now use AI in some capacity to improve your output quality or speed.
If I ask you what AI model the company should use to solve this (hard) problem, given pricing is not a big constraint, and you tell me it is Gemma 3, a very very old AI model, self-hosted because you like privacy, that is not a great answer and shows me you are totally out of touch.
1:24 PM UTC · Aug 27, 2026 · 12.8K Views
30919