What Should You Study If AI Writes the Code?

A group of high school seniors sat Ofer Mendelevitch down at their computer science club. They’d all just been accepted into college CS programs. And they looked at him and asked a question that’s become the defining anxiety of a generation entering tech: “Did we make the most incredibly stupidest mistake ever? What are we gonna do for four years in college?”

It’s the honest version of a question a lot of working engineers are also asking, quietly. If coding agents keep getting better, what exactly are you training for? Ofer is worth listening to on this because he isn’t a pundit floating think-piece takes — he’s the author of O’Reilly’s Hands-On RAG for Production, an independent AI advisor who led developer relations at Vectara, and someone who has personally lived through cycle after cycle, from naive Bayes to XGBoost to GPT-2 to today’s agents. He also uses coding agents daily and mostly guides them. So his answer comes from the seat, not the sidelines.

And he doesn’t wave the fear away. “I recognize it’s a pain that impacts real people with real families.” What he offers instead is a reframe of what the job becomes.

Directors, not actors

Here’s the shift in one line: “We’re gonna end up in engineering and data science being — it’s not a word I invented — directors as opposed to actors.”

The actors — the people typing the code line by line — get replaced by the agents. “The coding agents will write the code,” and they keep improving. But the director’s job survives, and it’s the harder job: “What remains a really critical challenge is to understand what we need to build and guide it.” Deciding what to build, choosing the architecture, catching the places where there are five ways to do something and only one is right, knowing when the agent’s output is subtly wrong — none of that is typing, and none of it is going away soon.

So the value migrates up a layer, from production to judgment. “The best engineer or data scientist of the future would be people who essentially augment themselves effectively with the coding agent.” Not people who resist the tools, and not people who lean on them blindly — people who direct them well.

What that means for a CS degree

This is exactly what he told the high schoolers, and it’s the practical core of his advice. The old equation is breaking: “Companies now equate a junior engineer with somebody who just graduated from college. That’s one-to-one. But it doesn’t have to be.”

His reframe of college turns the four years from a liability into leverage:

• Come out ahead of the old baseline. “You can graduate from college with the capability of a mid-level or senior engineer today. So use the next four years to learn how to be that.” The compressed entry-level rung isn’t a dead end — it’s an invitation to skip it.
• Master the tools deliberately. Learn to wield coding agents as effectively as you can, because directing them well is the new core skill.
• Don’t skip the fundamentals. “It’s still important in universities to learn the fundamentals of computer science, fundamentals of statistics, machine learning, how it works.” Those are what let you catch the agent when it’s wrong. The rote parts are what fade — “write an algorithm to solve Fibonacci in five lines of Python, that’s worthless” as a test now.

The goal, in his words, is to graduate as a hybrid — the computer scientist or data scientist of the future who pairs fundamentals with fluency in the tools.

How hiring will have to change

If the skill is directing agents, the way companies test candidates has to follow. Ofer expects the whiteboard-algorithm interview to give way to something that looks like the actual job: “I wouldn’t be surprised if they’ll say, here’s ChatGPT or an Anthropic coding agent, and we’ll give you some exercise and we see how you do with it together.” The evaluation becomes how well you collaborate with the agent to produce something high-quality and correct — how you make good decisions in the process — not whether you can reproduce an algorithm from memory. Angelina noted she’s already interviewed founders who are folding AI tools into their hiring loops. It’s not a forecast; it’s underway.

Ofer’s own concern isn’t really the technology — it’s what society does with it. He worries about the transition being handled badly: layoffs that overshoot, inequality, “a lot of pain” in the gap. But his advice to individuals, including his own kids, is consistent and unpanicked: adopt it, embrace it, augment yourself, and aim to be the director. The people who do that, he believes, “will succeed more.”

FAQ

Is a computer science degree still worth it if AI writes code?

Yes, but the goal changes. Instead of graduating as an entry-level coder — the rung agents are compressing — you use the four years to reach the capability of a mid-level or senior engineer. That means mastering AI tools and, crucially, the fundamentals of CS, statistics, and machine learning that let you judge and correct what agents produce.

What does “directors, not actors” mean for engineers?

It’s Ofer’s model for the future of engineering. Coding agents become the “actors” that write code; engineers become “directors” who decide what to build, choose architecture, and guide the agents where they go wrong. The value shifts from typing code to judgment, direction, and knowing when the output is subtly incorrect.

What skills should engineers learn to stay valuable?

Two things together. First, the fundamentals — computer science, statistics, and how machine learning actually works — because they let you catch agents when they’re wrong. Second, fluency in directing coding agents effectively. The strongest engineers pair deep fundamentals with the ability to augment themselves with AI tools rather than resisting or blindly trusting them.

Will AI coding agents replace software engineers?

Ofer sees replacement of the “actor” role — line-by-line coding — but not the engineer. The hard parts remain: understanding what to build, architecture decisions, and guiding agents through cases with many possible approaches. He acknowledges real pain in the transition but expects engineers who augment themselves with agents to succeed, not disappear.

How will technical interviews change because of AI?

Ofer expects a move away from memorized-algorithm tests like writing Fibonacci from scratch, which he calls worthless now. Instead, interviews may hand candidates a coding agent and an exercise and evaluate how well they collaborate with it to produce high-quality, correct results — testing judgment and decision-making rather than recall. Some companies are already doing this.

What should CS students do during college now?

Treat the four years as time to exceed the old entry-level baseline. Learn to use coding agents as effectively as possible, but don’t skip fundamentals — they’re what let you direct and correct AI. Aim to graduate as a hybrid engineer who combines strong CS and ML fundamentals with real fluency in AI tooling.

Does this apply to data scientists too?

Yes. Ofer frames the shift across engineering and data science alike — both move toward directing agents rather than doing all the hands-on production work. The same advice holds: keep the fundamentals of statistics and machine learning that let you evaluate outputs, and get good at augmenting yourself with AI tools.

What’s the biggest risk in the AI-and-jobs transition?

Ofer’s larger worry isn’t the technology but how society handles it — overshooting layoffs, widening inequality, and real hardship during the gap. He believes the tools can benefit everyone, but the outcome depends on how companies and governments manage the shift. For individuals, his guidance is to adopt the tools and become the director rather than wait to be replaced.