I LOVE a good technical challenge. There is something satisfying about solving a complex problem, optimizing a data structure, or shaving off a few milliseconds of latency. But this past summer, I found myself back in the job market, and let me tell you… Things have changed.
The industry has split into two distinct realities.
On the one hand, I interviewed with established enterprise giants whose processes were like bread and butter to me. I was solving algorithmic puzzles and showcasing my understanding of the DOM and general web development skills. On the other hand, I interviewed with startups where the interviewer essentially handed me the keys to GitHub Copilot and said, "Build this feature. You have 45 minutes. Go."
It made me feel a bit like an "old man yelling at the sky," but it also raised some serious questions about where our industry is headed. Are we testing for engineering skills, or for subscription tiers?
The Great Bifurcation
If you've been interviewing lately, you've probably felt this whiplash. The data backs it up; we are seeing a dramatic bifurcation in how companies hire.
The Enterprise Players
The big players are committed to the LeetCode style interview. Why? Because they are terrified of AI impostors.
When I sat for these interviews, it was all about "Proof of Work". They wanted to know that I possessed the raw cognitive bandwidth to manipulate logic without a robot whispering the answer in my ear. And honestly? I get it. With tools like ChatGPT, it's easier than ever to fake competence. In fact, 81% of interviewers at Big Tech companies suspect candidates of using AI to cheat during remote sessions.
So, the LeetCode grind isn't going anywhere. It's the safety against the "AI-powered poser."
The Startups
Then there are the startups. These folks have embraced our new machine overlords. They aren't looking for someone who can write a linked list from memory; they want "AI Editors" and "Sense-Makers".
In these interviews, the constraints weren't syntax or memory; they were my ability to prompt, debug, and integrate. It was eventually fun, but also frantic. I had to get over the mindset of “showing" or “explaining" my work. I wasn't there to show I know the native functions and behaviors of the DOM. Instead, there was this massive expectation for speed. Because AI tools are supposed to make us 10x developers, the interview pacing was often set to "warp speed," expecting me to blaze through boilerplate code.
The Hidden Barrier to Entry
Here is the part that really worries me, and it's something we aren't talking about enough: THE COST.
In the traditional interview, all you needed was a brain and a whiteboard (or a laptop). Today, for these startup interviews, you are often expected to bring your own AI tools.
The startups aren't always providing the enterprise seats for these AI tools during the interview process. So, are you on ChatGPT's free tier? Or are you shelling out $20/month for Plus? Do you have a personal GitHub Copilot subscription?
It introduces a subtle but real economic barrier. Is your success in an interview dependent on whether you can afford Claude Code Max 20x ($200/month) vs. Cursor Pro ($20/month)? If my model hallucinates because I'm on a lower tier, and I didn't catch it, did I fail the interview?
Startups are essentially asking candidates to pay for the privilege of being efficient enough to get hired.
The "Home Court Advantage" Problem
Some companies, like Meta, are addressing this inequality by introducing standardized interviews that use the same AI tool for everyone. On paper, this sounds fair. It levels the playing field and removes the cost barrier.
But does it?
We all have our "fine-tuned" setups. You may have a CLAUDE.md or AGENTS.md file. We are "pros" with our tools. Throwing a developer into a standardized AI environment, like Meta's CoderPad setup, is like handing a race car driver a rental sedan and asking them to set a lap record. You might know how to drive, but you don't know this car.
From my own experience working at Meta, I found that even their internal tool, Metamate, struggled to handle complex tasks on the actual Facebook codebase. It often felt like I was correcting everything it output rather than it actually making life better. Anyway, if you've never used their specific flavor of AI before, are you ready to use it like a pro in a high-stakes 60-minute interview? Probably not.
Try Before You Buy
Finally, this brings me to the Paid Work Trial.
Startups are increasingly asking candidates to join the team for a few days to work on actual production tickets. In theory, there is some merit to this. It gives you a really good idea of how well you'll click with a team, and it mitigates the risk of a "false positive" hire.
But let's look at the logistics. What if you already have a job?
Can you really take 3 to 5 days off to work somewhere else? That's a massive time commitment that privileges people who are currently unemployed or have incredibly flexible schedules. Hell, does your current employer allow you to moonlight at another job? Most employment contracts have clauses that prevent us from working for another entity, even for a short "trial." The Paid Work Trial requires us to breach our current contract to potentially secure a new one.
tl;dr Everything is Awful
I wish I had a magic answer, but the reality is that for the foreseeable future, we're going to have to be bilingual.
- Keep your fundamentals sharp: The "old way" isn't dead. We still need to prove we understand the code well enough to audit the AI's output.
- Master (Generic) Prompts: Don't just rely on your custom configs. Get good at "Prompt Engineering" in a vanilla environment, because you might not get to bring your own configs to the interview.
- Protect your time: If a company asks for a multiday trial, ensure it is paid at market rates. Don't work for free.
It's a weird time to be interviewing, caught between the LeetCode grinders and the vibe-coding speedrunners. But hey, at least it keeps things interesting, right?
I'd love to hear your horror stories (or success stories!) from the 2025 interview circuit. Are you seeing more work trials? More AI? Let me know!
Top comments (14)
As the owner of a company I've struggled with this too. I ultimately want people to possess both good analytical skills themselves AND the ability to be effective with AI. In terms of cost to barrier to entry, I think of this like any other craft. If you are a carpenter, you should own a drill. A chef should own their own set of knives. I would expect an engineer to have a good machine of their own and preferences in terms of their tooling. If you don't have a subscription to an AI service, I'm going to question your ability to be effective with it. I use AI to code every single day, and I can tell you that I've for sure gotten better at it over time and I'm still learning more and more about how to get the best quality work for the shortest period of time out of my tooling. If you were planning to just hop into an interview and vibe code with whatever tooling they decide to let you use, I'd say that's a recipe for disaster. I WANT your opinions and tooling. Especially this early in the AI game, I am expecting new ideas from engineers on the best way to work. I'm not sure how that happens if you're not curious enough to do some serious tire kicking with these tools on your own.
I agree! But now we must also bring our own refrigerator and pans too? To what end?
I shouldn't need to bring my own sink to show I can cook. What happened to looking at past results and experience?
dev.to/j4s0nc/the-bloated-kitchen-...
Maybe we are heading for a transformation from needing a deep understanding of all the things, to more focused problem solvers. Tooling is just that, tools, pick it up and go to work.
Yes but you do own your own refrigerator and pans too don’t you? How can you be a good chef but only cook when you’ve got a job? Craftsman got to craft! You do need to have tool preferences and you learn that my trying different tools.
Sure, I got a my own residential tools at home and I have better professional grade tools at work, provided by work (their kitchen, their tools).
my point: craftsmanship should be in focus, not just tools.
I can cut some things with a fork, I know that tool. I also know preparing veggies is faster with a knife or food processor (outcome and volume determines the tool).
Isn't the value in my ability and the shape of the veggies when I'm done? Regardless of the tools I had access too or used.
Yes I’m just questioning the abilities of a person that doesn’t have their own tools to use at home. Maybe that’s wrong, but I’ve never interviewed someone that was great that didn’t immediately bust out their own preferred setup and crush with it. I don't want to dictate the tools someone uses in the interview, I want you to show me what you would use. If you're coding with regularity, I expect you to have those tools on hand.
Oh I agree.. if you only eat McDonald's and order door dash, don't tell me you know how to cook - that's knowing how to order and eat.. way different then cooking.. I question if some people know how to turn the stove on these days.
Bringing it back to tech and dev, some people are building our ai future and bringing new techniques to today's kitchen. Some people are making a mess and calling it innovation. Most of us are worried and wondering where we fit.
Keep solving the real problems with technique and tools like ai.
Our tech kitchens are about to get remodeled
Totally like there are things that people are talking about like RAG and multiple agents at once, etc. things that sound good but I know if you’ve been actually using agents and these tools aren’t as great as they sound in marketing material. So when I interview I want to hear some real opinions that show you’re pushing this stuff and finding what works and what doesn’t.
This back-and-forth is exactly what keeps me up at night regarding this shift! 😅
@mrispoli24, I totally get the "craftsman" perspective. I personally pay for my tools because I want that muscle memory and efficiency. But like @j4s0nc hits on, it's the quality of the tools we have at home vs. at work (the equity component). If the "set of knives" now costs $50–$200/month in subscriptions (Cursor Pro+, Claude Pro), we're creating a pay-to-play layer for entry-level or unemployed devs.
We're going to see a clash between "show me your curiosity/investment" and "don't gatekeep based on disposable income" for a while. Love the kitchen analogy you two ran with here; it illustrates the friction perfectly.
So I can see this perspective too but I actually think the cost of the tools might go down relative to the quality. I personally think the Claude code or opencode workflow is king over the new IDE’s. Opencode is actually awesome and free. In terms of models I do pay for the Claude max plan at $200 per month, which I think if you weren’t employed you could just drop that down to the mid tier plan. However, I’ve also been digging into the open source models to run locally, not because of cost, but because of risk. I find Claude is so integral to my margins that I actually need to explore the open landscape should it go down or become prohibitively expensive some day. The open models especially the larger ones are actually quite good and getting better. Now you do need a pretty good setup to run them but it’s never been more cost effective to get tons of ram and harddrive space. A good gaming computer could run these nicely. I recently bought a framework desktop to be able to do this. I think any of these models could help you in an interview demonstrate your dev flow and also how you review and correct LLM code! But I would say we should all get down with ollama and some open source models and check it out. Coupled with opencode you end up with a free local LLM that I think could be a great replacement for the paid tools someday.
This feels so true. The whiplash between LeetCode mode and vibe-coding-with-AI mode is unreal.
At Everdone CodeDoc, we see people struggling not with skill but with context switching — between deep knowledge and fast prompting, between clean thinking and “go fast or die trying.”
Interviews in 2025 basically expect you to be a hybrid human. Wild times.
If u value the startup and its potential as high then taking 3 -5 days off for work trial is ok as you will anyhow find a way to do it. Joining a startup needs that sort of problem solving skills. And if the trial is successful then the candidate must be paid full salary as per offer letter . If the startup is convinced of the skills of the person they can pay up to 50% for that work trial irrespective of outcome.
I have to push back on this. Framing this as a "problem-solving skill" ignores the massive risk asymmetry. You’re asking candidates to burn PTO and return to a backlog of work for something that isn't guaranteed. Unless there is a signed contingent offer and you're the sole candidate, this is likely just a "bake-off" against others. Companies (startups or enterprise) will always do what's best for them; we need to do the same. Risking stable employment for a 50% paid trial is rarely a smart move. And this doesn't even begin to cover the potential legal risks developers may be taking on doing this!
Hi @michaelsolati .. I will make it more clear . Suppose the startup is Space X and there is no Elon musk. You only know they are building reusable rockets , first time in the world. You want to join it for the unique vision and the unique growth and skill trajectory it can give you . So you are selected and there is a paid trial ... I am against free work any day . Space X would want to know how you are going to set in their work setup as you are giving them equity or stock options and even you would want to know what's your role and whether its actually as the hyped and you want to take this opportunity . You wouldn't miss this . Of course this doesn't apply for the 100 plus new AI startups blooming everyday. Only for great startups or vision.
I hear you 100%. If it’s a once-in-a-lifetime "rocket ship" (pun intended), the risk calculation definitely changes for the candidate.
My worry is just that every founder tends to think they are building the next SpaceX, and they use that "vision" to justify asking for the unreasonable during an interview process. It's a slippery slope! But I appreciate the clarification; context definitely matters, and for the right opportunity, breaking the rules can make sense.