Recently, some of my mentees—freshmen in college—asked me the question everyone is thinking: “If AI is going to disrupt everything, is it even worth pursuing Computer Science? What do we even do? If there are going to be no jobs in the future, what is the point?”
This reminded me of a recent insight from Andrej Karpathy:
"I often say that pre-AGI education is useful. Post-AGI education is fun. In a similar way, people go to the gym today. We don’t need their physical strength to manipulate heavy objects because we have machines that do that. They still go to the gym... Why? Because it’s fun, it’s healthy... Education will play out in the same way. You’ll go to school like you go to the gym."
I was kind of stumped because I've had the same existential question. Dario Amodei, the CEO of Anthropic, recently said "AI Could Replace Software Engineers in 6 to 12 Months." We see legal stocks falling due to AI, and we know that knowledge work—editors, translators, accountants—is shrinking because AI can function as a "good enough" tool for these tasks.
But simply spiraling into fear isn't the answer. Here is how I am navigating this, and what I told my mentees.
The Reality Check (Don't Be a Denialist)
First up: Don't be an AI denialist.
I talk to people across the software industry who still think AI models "just predict the next token" and won't get better. That is a dangerously simplistic view. Yes, they predict tokens, but they also reason, check their answers, and perform deep research. The progress in just three years—from ChatGPT's hallucination-prone launch to today's reasoning models—is undeniable.
The restructuring is real. In 2025 alone, AI impacted over 54,000 jobs. Companies are getting leaner, investing in "AI efficiency" rather than headcount. If you are not one of the few researchers at the forefront of building these models, you are not untouched.
Why This Time Is Different (The Velocity Problem)
People often counter this doom with: "When the Industrial Revolution happened, we didn't run out of jobs; we just found better ones. Same with the Internet."
While historically true, there is a flaw in that logic: Velocity.
When the Industrial Revolution happened, it took decades to spread across the globe. You had to physically manufacture machines, ship them, and install them. There was a natural scaling limit. Even when the internet arrived, it took years to penetrate businesses that were stuck on manual bookkeeping.
AI is different because it is software. It helps its own adoption. The distribution is nearly instant. My fear is that the displacement of current jobs will happen much faster than the creation of new ones. We might face a painful gap where unemployment rises before the new economy stabilizes.
So, What Do You Actually Do?
You cannot wait for the dust to settle. You have to pivot now.
1. Embrace the Tool. Stop fighting it. Use AI for everything. If you are a student or a junior engineer, you need to be faster and better than the person refusing to use it.
2. Understand "Context" and "Grounding." This is where the new jobs are. Stop worrying about syntax and start worrying about system architecture.
- Context Rot: Learn how to manage the information you feed the AI.
- MCP (Model Context Protocol): Look up how to create MCP servers, agents, and sub-agents.
- Grounding: How do you ensure the LLM's data is your data? How do you prevent hallucinations in a business context?
There will be a massive demand for engineers who can architect these pipelines—people who can empower LLMs to do the work reliably.
3. Move from "Artisan" to "Architect." You might not be writing "artisanal" code line-by-line anymore. But you still need to know good design from bad design.
If a client gives you a vague requirement, an LLM will give you a vague (and probably bad) output. You still need the expertise to:
- Fill in the blanks of user requirements.
- Evaluate the AI's output for security and efficiency.
- Judge the architectural decisions.
If you are a Subject Matter Expert (SME), your value shifts from doing the work to evaluating the work.
4. Don't Forget the Joy of Building. There is a lot of difference in reading the solution to a math problem v/s the struggle of solving the problem yourself. The struggle is where the learning happens; it’s where you train your brain to recognize patterns and figure out what works and what doesn’t work.
If you stop building, you lose that intuition. When an LLM inevitably hallucinates or generates a subtle bug, you won't have the muscle memory to spot it. You cannot effectively debug or architect a system if you haven't experienced the friction of building one from scratch. Be a builder.
The Verdict
Your skepticism is valid. Why lift heavy weights if a machine can do it? Because, like the gym, you need the foundational strength to handle the machine.
You gotta do what is best for you right now. Whatever I am saying might be outdated in six months, given the pace of this industry, but one thing is certain: ignoring it is the only wrong move. Get the basics right, learn to architect the agents, and ride the wave instead of drowning in it.