Last month, in a fresher interview, a candidate pasted a clean, working solution to my problem. Then I asked why his approach handled one edge case badly. He couldn’t tell me. He hadn’t written the code. An AI had, and he’d never understood it well enough to see where it broke.
That interview is the whole story of what’s changed, and what hasn’t.
I’ve run hiring loops for engineering roles for years. At Google, at HackerRank, and now at Kalvium. The tools candidates walk in with have changed completely in two years. What actually gets someone hired has barely moved. If anything, it’s gotten sharper.
The thing everyone is getting half-right
Yes, AI can write code now. It can pass a lot of exams, summarise a textbook, and produce a working function from a one-line prompt. That part is real, and pretending otherwise is silly.
The wrong conclusion is that this makes engineers less valuable.
On the ground, in the interviews I run, it’s the opposite. AI did the easy 60%. That made the remaining 40% the entire job. And that 40% is built from skills most degrees never taught in the first place.
Here are the ones that keep separating the candidates who get hired from the ones who don’t.
1. Framing the problem worth solving
AI answers the question you ask it. It has no idea whether that’s the right question.
The candidates who stand out aren’t the fastest typists. They’re the ones who, handed a vague problem, ask three sharp questions before writing a single line. What are we actually optimising for? Who uses this? What breaks if it’s slow?
A degree tests you on problems that are already fully specified. Someone else decided what to build and handed it to you as a question paper. Real work is mostly the opposite: figuring out what to build. AI-assisted work even more so, because the machine will happily build the wrong thing, fast. No exam ever made a student sit with an unclear problem and find the real one inside it.
2. Judgment about whether the answer is any good
AI is confidently wrong all the time. So the skill that’s suddenly worth a lot is knowing when it’s wrong.
I sometimes put an AI-written function in front of a candidate and ask a simple thing. Would you ship this? The gap in the room is enormous. Some spot the security hole and the unhandled case in thirty seconds. Others just trust it, because it looks right and it ran.
That judgment, a real sense of what “good” looks like, comes from having built and broken things yourself. You can’t shortcut it. The syllabus grades you on producing the right answer. It never trains the harder muscle: judging someone else’s answer, including a machine’s.
3. Owning a thing when it breaks
AI generates. It doesn’t take responsibility.
When a system is down at 2am, someone accountable has to hold the whole picture in their head and make a call. That’s one of the most human parts of engineering, and it’s the least teachable from a lecture.
In interviews I ask a standard question. Tell me about something you built that broke, and what you did about it. Candidates who’ve never owned anything end to end have nothing to say. Candidates who have owned something light up, because they’ve lived it. Ownership isn’t on any marks card. It only comes from having your name on a thing that could actually fail.
4. Explaining a decision to a human
The higher AI raises the coding floor, the more the job becomes communication.
Convincing a teammate. Writing a proposal a busy person can follow. Telling a non-engineer why the timeline is what it is, without hand-waving. I’ve passed on strong coders who couldn’t explain their own work, and hired weaker coders who could reason clearly out loud. It isn’t close.
And this is the single most predictable thing Indian degrees skip. Four years, and almost nobody is graded on explaining anything to anyone. Then day one at a job is nothing but explaining.
5. Learning the next thing, fast
This is the big one.
The specific tool you learn today has a short shelf life. Two years ago, nobody was hiring for LLM integration. Now it’s on half the job descriptions I see. The durable skill was never any single framework. It’s the ability to pick up the next one quickly, on your own, without a course holding your hand.
What I actually look for is evidence a candidate taught themselves something hard, recently, by choice. That signal predicts performance better than almost anything on a resume. It’s the skill that outlasts every tool, including the AI ones, because the one certainty in this field is that the tools will change again.
Notice what’s not on this list
None of these five is “coding.”
AI made the rote content of a degree nearly free. What it can’t do is precisely the list above: frame the real problem, judge the answer, own the outcome, explain it, and learn the next thing. And that list is exactly what a syllabus-and-exam system was never built to produce.
That’s the gap. It was always there. AI just made it impossible to ignore.
So is a B.Tech worth it in 2026?
My honest answer, from the hiring side, is this. The degree is worth exactly as much as the durable skills it builds in you.
A programme that’s four years of lectures and exams is worth less every month. The thing it trains, recall of content, is the thing AI made cheap. A programme that puts you in front of real problems, real teams, and real failure is worth more than ever, because that’s the only place the durable skills come from. Don’t ask whether a B.Tech is worth it in the abstract. Ask what this particular programme will actually make you able to do.
What building for this looks like
This is the part I care about, and it’s why I helped build what we did at Kalvium.
Students ship working software from Semester 1, not Semester 7. The DOJO runs daily coding practice six days a week, which is really a machine for building learning velocity: you get used to being uncomfortable and working it out anyway. There’s an actual Learning How to Learn course in the first semester, and an Autonomous Learning principle that runs through all four years, learn past the marks and the syllabus. From Semester 3, students build alongside industry partners, where code gets reviewed by someone who isn’t there to award marks.
And yes, they learn to build with AI. LLM integration and AI agents are in the curriculum. But the point is the judgment around the machine, knowing when it’s wrong, not just how to prompt it.
You can’t build problem-framing, taste, ownership, or learning velocity in a lecture. You build them by doing real work, early, with people who push back on you. The 2026 batch was 82.40% placed as of March 2026, with a median of 16.5 LPA and companies like Morgan Stanley, PhonePe, and Thoughtworks hiring from it. That outcome isn’t about the specific tools those students learned. It’s the durable skills underneath.
What I’m not saying
Let me be clear, because it’s easy to misread this.
I’m not saying AI is hype, or that coding skills don’t matter. They’re table stakes. You still have to clear the technical bar, and the companion piece on the six things a hiring loop tests in CSE freshers covers exactly what that bar is.
What I’m saying is that AI raised the floor, and in doing so it raised the ceiling. The rote part of the job is going away. The judgment part isn’t. It just got more valuable. The engineers who’ll do well aren’t the ones who fear AI, and they aren’t the ones who blindly trust it. They’re the ones who can do the things it can’t.
The one thing to take from the hiring side
Don’t spend your four years optimising for what an AI can now do in seconds.
Optimise for the one thing it can’t: becoming someone who can frame the real problem, judge the answer, own the result, and learn the next thing without being told. That was always the actual job. AI just made it obvious.
This piece sits under a bigger argument: that a college should build an adult, not just award a degree. For the evidence base on how capability is actually built, see what the learning science says works. For families turning this into a decision, the framework for choosing a B.Tech CSE programme and the complete guide to what Kalvium involves both apply the same filter: does this programme build the durable skills, or does it optimise for the exam?
Anil is a co-founder of Kalvium and previously led engineering teams at Google and HackerRank. He writes about what the Indian tech market actually rewards, from the interview side of the table. Read more from Anil or browse the Careers category.