# Hiring's changed a lot in the AI era on the management side, but I'm still finding great people. Want to know how to find people and avoid scammers? Want to know what people (should) look for? Here's what I ask. — retained record

> Retained reddit record from reddit:cscareerquestions.

Canonical: https://highsignal.app/data/records/c02c6c979e36f712

[← sources](https://highsignal.app/data)


source: [reddit](https://highsignal.app/data/reddit) (reddit:cscareerquestions)  published 2026-10-20 01:27 ingested 2026-10-21 02:36

[original source ↗](https://www.reddit.com/r/cscareerquestions/comments/1xaf2x9/hirings_changed_a_lot_in_the_ai_era_on_the/)

## retained text



```
I've posted pretty openly on my career progression on this forum, from maintaining a decades-old PHP-based web platform to getting hired by a customer of said platform (most flattering hire ever) to turning into a manager for said customer and developing my own development team!  VP now, yippee-yippee-creature!

Been hiring for like 7-8 years now, and wow, things have changed.  [I talked a lot a few years back](https://reddit.com/r/cscareerquestions/comments/zg5wtb/ive_hired_people_who_failed_coding_questions_let/) about hiring people who failed coding interviews because of having great organizational and logistical skills (namely, problem categorization, the ability to communicate your understanding of the problem and your proposed solutions, a willingness to ask questions and not just silently struggle and get stuck, tenacity, and research capacity), and I figured, hey, it's been a while, I should update everyone with my new heuristics now that i've spent a couple years really ironing out my hiring process.

So what's changed, and what's the same?

AI.  AI has changed.  I was asked, "What will you do if they use GPT?", and I jokingly said, "I'd love to see them try!".  

And that's, surprisingly, still true!  I answered facetiously previously as it was GPT-3 around then and it was pretty poor, but now AI functions as the greatest lossy compression-and-retrieval scheme for boilerplate code and common design patterns ever invented, with all the world's code stored in it, and utilizing it as said reference tool is incredibly good for the 90% of code people make (aka boring code).

First, I've expanded my "ask me anything" policy.  As part of the interview process, I tell people straight up, "Congratulations, this is an open-book, open-note, open-AI (hah, get it?) test!.  There is no* cheating on this test.  You can use literally anything, and even have AI print all the solutions for you, and can, and should, ask me anything!  ASK QUESTIONS!"

And then I go into questions with solutions I know.  But the solutions to what, you may ask?  Well, I've abandoned leetcode entirely.  I've got a couple basic question patterns I do now.  Leetcode is just too well ingrained in AI, and I want to see if they can learn domain-specific terminology over the course of 90 minutes.

First, I do fizzbuzz.  If they take more than 15 minutes once the coding challenge starts to fizzbuzz with the entire world and AI at their hands, I don't know what to tell you.  I so wish I didn't have to fizzbuzz people, but [when  this piece of 2007 wisdom remains so true](https://blog.codinghorror.com/why-cant-programmers-program/), alas.  Stop giving me people with a decade of experience who can't fizzbuzz, and I'll stop fizzbuzzing.

Fizzbuzz aside, First category of problems:  I randomly generate fake data sets and have them do extremely basic manipulations on the data set using preferably Python, as we're a Python shop.  Stuff as simple as, "Hey, this some fake raw appointment data shaped like some database, how many appointments did each doctor do?".  Do have to, in some form, make a program that does this evaluation though - can't just toss it into AI, but you can have AI generate 100% of the code.  

Ah, but that's where I get tricky, because the AI can generate all the code it wants, but half the appointments don't have doctors!  What does this mean?  Did they ask me questions about that, or did they just blind fire the code into their IDE?  If I say, "hey, what do you think the 21731 appointments without doctors mean?", do they panic?  Do they ask the AI to fix the code after explaining the problem?  Or do they stop, think of a second, and realize that hey, not all appointments at clinics are with doctors?

I leave several pitfalls and strange situations in the code.  Things like a summary row at the end of the file that puts strings into every column, including integer ones, things like a fake doctor name that's in a language most people don't have installed on their computer, things like having a really, really long and incredibly obnoxious file name just to see if they rename it to simplify their lives.  I throw four files at them and have them only join two and leave the others completely alone, to see if they actually take stock of what they have.  I'm looking, specifically, to see that they regard and then disregard the extra files, rather than simply ignoring them or overly fixating on them.

And all of this is right in line with my original interviewing philosophy - **I want to see research capability, communication capability, on-the-fly learning, and ability to adapt.**

Second category of problems:  Install this library, use it to solve this specialty problem.   My favorite is to have them do a small Selenium browser automation that loads up Google, grabs their logo, and renders it in Python.  It's maybe 15 lines of very simple, straightforward code that AI can generate in a heartbeat if you're using it right.  But you have to install some libraries!  Not a lot, no more than 4, but can you find them, install them, understand what needs to happen to make it happen?  And most people use, funny enough, Google or AI.  Google research is good if you can find the libraries, but AI will do this instantly.  But are they using AI right?  Do they give it a good description of the problem, evaluate what it says and make a reasoned decision on what to use or not use?  Or are they blind-copying what AI gave them?  Very few people will ask AI, for example, "Hey, interviewer asked me about what to do when Google rate limits or detects automated traffic" but *that*, sadly, is considered above-average on-the-spot thinking in my personal interviewing experience.

Third category of problems:  Someone wrote this trash code and it needs to be fixed.  It's not doing what the user expects.    Please evaluate it and figure out what's wrong and fix it.  Biggest thing:  I never, ever tell people what "the user expects" initially.  They need to *ask*.  I've had people spend half an hour solving a problem they didn't even think to clarify, and that worries me!  Often, the code will have one or two small syntax errors, but once running, will have a logical "bug" (usually something stupid and simple like, "I wanted this to give out one value per month, not one value per week", quick filter tweaks fix it) that needs to be located and edited.  Not something AI will obviously identify as a bug, and people will be absolutely baffled that the fully working code that provides results got them a 0 grade and then understanding once I explain the gimmick.  (And let me remind you, I really, *REALLY* emphasize "ASK LOTS OF QUESTIONS" at the start!)

And, ultimately, that's what I'm looking for - people who can understand requirements and, using any tools at their disposal, meet people's needs.  And it's very specialty, so they need to either know, or demonstrate a capacity to learn, said specialty!

I've done over a hundred candidate tests at this point, and it's surprising how many will fail to get past question Fizzbuzz or question 1, even with the entire world at your fingertips.  I've had some people seem technically strong but were unable to have real conversations, and I've taken a chance on some people who got correct answers but struggled with conversations due to nervousness.  I've had some bad hires and had to really adjust my hiring methods to fit the AI age, but a true open-anything test makes it impossible to cheat.

Feel free to ask me anything regarding hiring in these wild times!

* One note about not cheating - despite telling people "it's an open everything" test, I did have one poor person silently type into AI and then copy-paste code into the IDE for 20 minutes on fizzbuzz and then 40 on the first data manipulation question, and then lied to me about using AI despite me saying it's open-everything!  Very, very bizarre behavior.  Maybe don't do that.
```



document fetched 2026-10-21 00:23 · canonical https://reddit.com/r/cscareerquestions/comments/1xaf2x9/hirings_changed_a_lot_in_the_ai_era_on_the
