Advice for a Sophomore CS Student Whose Lab Partner Is a Language Model

by educationLeave a comment 6 min read

We asked Hal9000 (Claude): “What recommendations would you make to a 2nd year CS student for his career? How should they prepare for AI?”

If you’re a second-year computer science student in late 2026, you’ve probably noticed something awkward. The tool that helps you finish your homework is the same tool being blamed for the shrinking junior job market. It’s a bit like your study buddy also interviewing for your future job.

The junior market really is the tightest it has been in a decade. Here’s the more useful framing, though: it’s a squeeze, not a wall. Software engineering overall isn’t shrinking. Senior roles are holding steady, and demand for AI, security and infrastructure work keeps growing. What has collapsed is the old entry route of “graduate, get hired as a junior, learn on the job by writing the boilerplate.” The boilerplate now writes itself.

So here’s your goal for the next two and a half years. Don’t graduate looking like a 2019 junior developer. Graduate looking like someone who already has a year or two of real work behind them. Here’s how.

1. Use AI constantly, but don’t let it do your learning

This is the trap that worries me most, because it doesn’t feel like a trap. Let a model write your data structures assignments and you’ll leave with a good GPA and almost no judgment. Your transcript will say you can code. Your first production incident will say otherwise.

The value of a junior engineer is moving from writing code to judging it. That means reading unfamiliar code, spotting the bug that compiles cleanly, and explaining what “correct” means before anyone writes a line. You can’t build that judgment if you skip the struggle that produces it.

A workable rule:

  • Coursework: do it the hard way first. Struggle, get it wrong, debug it yourself. Then ask the model to critique your solution, not write it.
  • Side projects: let AI rip. Use agentic coding tools to build things that would otherwise take three times as long. Learn to write specs, review diffs and catch the model being confidently wrong.

By graduation you should be fluent in AI-assisted development. Employers will assume you are. Your edge is the part they can’t assume: that you can tell when the output is garbage.

2. Go deeper on fundamentals, not shallower

The tempting conclusion is that operating systems, networking, databases and algorithms matter less now. After all, the model already knows them. The opposite is true.

When code is cheap to produce, the scarce skill is understanding why a system is slow, leaking memory, deadlocking or quietly exposing its database to the internet. The model will happily write your caching layer. It won’t notice the cache is serving one user’s data to another unless someone who understands the system asks the right question.

Those “boring” core courses are what let you supervise AI output instead of rubber-stamping it. Take the hard electives. Take compilers if you can stand it.

3. Aim where the demand is going

The numbers point in a clear direction. SignalFire found that new-grad hiring at large tech companies fell 65% between 2019 and 2026, and that generalist, stack-specific roles like front-end are shrinking fastest. Meanwhile Indeed Hiring Lab data, as reported here, shows security engineering postings up 124% year over year. ML engineer openings are up 59% over their 2020 level, while general software engineering openings are down 49%.

Directions worth a serious look:

  • Security. Every AI-generated line of code is a line nobody fully wrote, and attackers have noticed. Security rewards exactly the skepticism models lack. It’s also a field where “it compiles” has never been good enough.
  • AI and ML infrastructure, data engineering. Someone has to build the pipelines, evaluation harnesses and serving systems that make AI products actually work.
  • Systems, embedded, infrastructure. Domains where mistakes are expensive and the context is hard to fit in a prompt.
  • CS plus something else. Biology, finance, law, energy, manufacturing. A specialist who can code tends to outlast a generalist who can only code.

You don’t need to choose forever. You need to choose something you can go deep on by your senior year.

4. Build things real people use, and chase internships early

To-do apps and class projects no longer stand out on a résumé, because a model can produce one in an afternoon. What stands out is evidence of judgment under real conditions:

  • A tool your club, lab or a local business actually depends on.
  • An open-source contribution a maintainer reviewed and merged.
  • Research with a professor, even if it’s unglamorous.
  • A small product with real users and real bug reports.

Internships have gone from “nice to have” to close to essential, and the same Handshake data puts tech internship postings down 30% since 2023. Start applying this year, and cast a wide net: startups, mid-size companies, and non-tech companies with tech teams. Big-name internships are a lottery. Smaller shops often give you real responsibility in week two.

5. Build the skills a model can’t do for you

The skills that will set you apart in five years aren’t on any syllabus:

  • Writing clearly. Specs, design docs, bug reports, and prompts are all writing now. Engineers who write well get to direct the work.
  • Translating fuzzy requests. “Make it faster” and “users are confused” aren’t requirements yet. Turning them into requirements is a human job, at least for now.
  • Ownership. Seeing a problem through to the end, including the parts nobody assigned to you.
  • Explaining tradeoffs to people who don’t care about your architecture but do care about cost, risk and deadlines.

6. Network like it matters, because it does more now

Hiring pipelines are flooded with AI-polished résumés that all sound alike. That makes a human vouching for you worth far more than it used to. Go to meetups and hackathons. Talk to your professors. Email alumni. Become a familiar name to the maintainers of a project you use. One warm referral beats a hundred cold applications.

7. Get good at learning

Nobody knows what this field looks like in 2030, including the people confidently predicting it on LinkedIn. The safest bet is to get fast at learning new tools and domains, and not to tie your identity to one language or framework. Your first job title may not exist yet. That’s fine. Plenty of today’s titles didn’t exist when their holders were sophomores either.

The short version

Master the fundamentals. Use AI as a power tool without letting it think for you. Go deep somewhere that’s growing. Graduate with proof that you’ve done real work for real people.

The junior developer isn’t extinct. The junior developer who can only do what a model does for $20 a month is. Be the one who checks the model’s work.

Sources