If AI is Writing the Code, What’s Left for Us?
A real look at where developers, CS students, and the job market are heading
Let me say the thing everyone is thinking but nobody wants to say: it feels scary right now. If you are a CS student, a junior developer, or someone who just started learning to code, the world is moving faster than you can keep up with. One day you are learning React, the next an AI is building full React apps in 30 seconds. So what are we even doing?
The answer is not simple. It is not “AI will take all jobs” and it is not “don’t worry, everything is fine.” Both are lies people tell to get clicks. The truth is in the middle, and that middle is where you need to stand to survive the next 5 years.
The weird contradiction nobody is talking about
Two things are happening at the same time and both are true:
Thing one: junior hiring is actually down. Harvard found companies using AI cut junior developer hiring by around 9–10% within six quarters. Big Tech has hired about 50% fewer freshers over the last three years. A hiring head at SignalFire told the New York Times that “nobody has patience or time for hand-holding” anymore, because AI does a lot of that work.
Read that again. Nobody has patience for hand-holding. That is the new reality for freshers.
Thing two: overall developer jobs are still growing. The US Bureau of Labor Statistics says software developer jobs will grow 15% by 2034, 5x the average. Indeed postings are up 11% year on year. IBM just announced they are tripling entry-level hiring in the US.
So which is it? Both. The jobs are there. The door to get in is smaller and weirder than before.
What is actually changing in the work
The old junior developer path is dead. Join a company, get easy tickets, fix small bugs, slowly learn the codebase. Gone. Those easy tickets are exactly what AI is best at. Boilerplate, CRUD APIs, unit tests, small bug fixes. All of it.
What is replacing the junior role looks more like what a senior used to do:
- Understanding the problem properly before writing any code
- Reviewing what AI writes and catching the bugs it confidently put there
- Joining things together — APIs, databases, models, tools — under real world messiness
- Debugging things that AI got wrong (this is a real skill, trust me)
- Owning the outcome, not just finishing the ticket
Intuit’s CTO said they are hiring more early-career developers because they grew up with AI and understand it better than mid-career engineers. Freshers are not finished. But what is expected from them is very different now.
The skills that actually matter now
Here is where I want to get specific, because “learn AI tools” is the kind of advice that sounds good and helps nobody. Let me tell you what I think you should actually go deep on.
1. System design — the WHY, not just the WHAT. AI can write a Redis caching layer in 10 seconds. What it cannot do is decide whether you need Kafka or RabbitMQ for your throughput, whether your system needs strong or eventual consistency, or where to place your cache. These are judgment calls that need business context and failure-mode understanding. System design is not a “senior interview topic” anymore — it is the core skill that decides whether you are an engineer or a code typist.
2. Debugging and code review — the new core skill. Prediction: within 2 years, most production code at top companies will be AI-generated and human-reviewed, not human-written. The bottleneck is not writing code, it is reading and validating code you did not write. AI-generated code has specific failure patterns — subtle logic errors in edge cases, race conditions that only show under load, hallucinated library functions. The engineer who can read 500 lines of AI code, spot the three subtle bugs, and fix them in 20 minutes is 10x more valuable than one who would have written those 500 lines from scratch in 4 hours.
3. Model training and AI integration basics. You do not need a PhD in ML. But if you cannot work with AI models at a basic level today, you are like an engineer in 2010 who could not work with databases. Learn how RAG pipelines work, what LoRA fine-tuning is, how to run models locally with Ollama, how embeddings and vector databases fit together.
4. Infrastructure security — the one most engineers are sleeping on. AI agents are no longer just generating text. They are getting direct access to your servers, databases, and cloud accounts through MCP and function calling. If you do not understand security fundamentals, you are not shipping bugs, you are shipping vulnerabilities at AI speed. Container isolation, network segmentation, zero-trust, and IAM least-privilege are core engineering skills now. OWASP has a Top 10 for LLM Applications — go read it.
And this is not some distant problem. Anthropic just announced Project Glasswing — a coalition with AWS, Apple, Google, Microsoft, Cisco, CrowdStrike, JPMorganChase and others — because their unreleased Claude Mythos Preview model has already found thousands of zero-day vulnerabilities in every major OS and browser. One was a 27-year-old flaw in OpenBSD. Another was a 16-year-old FFmpeg bug that automated tools had hit 5 million times and missed. The model found most of these autonomously. Anthropic is not releasing it publicly because the same capability that helps defenders helps attackers.
The point: attackers are about to get a massive boost, which means engineers who understand security fundamentals are about to get very valuable. This is a window, not a distant trend.
5. Domain knowledge — the unfair advantage. Here is something AI will never have: years of context about YOUR specific business domain. The engineer who understands payments (PCI compliance, settlement flows, chargebacks) will always outperform someone who prompts an AI with “build me a payment system.” Pick a domain and go deep. This creates a moat AI cannot replicate.
6. Communication and influence. The best engineers share one trait that has nothing to do with code: they communicate complex ideas clearly. In an AI-assisted world this matters MORE, not less. You need to write specs and design docs that humans AND AI can work from. Writing a good prompt is just technical communication. Read The Staff Engineer’s Path by Tanya Reilly.
Should you still do a CS degree?
Yes. But not for the reason you think. A CS degree is not a job ticket anymore. What a good one still gives you:
Fundamentals that do not expire. DSA, OS, networks, databases, compilers. AI is amazing at writing code and terrible at explaining why your system is slow or why your database is deadlocking. The people who can answer these are the ones who understood the basics. Do not skip these.
Math and systems thinking. Linear algebra, probability, discrete math. If you want to do anything with ML, search, or ranking — you need this.
People and projects. The four years of building stuff with other people is maybe more valuable than the lectures.
What a CS degree will NOT do anymore is get you a job automatically. The NACE Job Outlook 2026 survey says employers are the most pessimistic about graduate hiring since 2020. They want proof — projects you built, a GitHub with real work. Around 45% of companies now say they do not even need a bachelor’s. The degree helps. The portfolio decides.
So what should you actually do?
Get good at AI tools, but do not become lazy. Use Claude, Copilot, Cursor. But after AI writes something, make sure you can explain every single line. If you cannot, you are training yourself to be replaced by the same thing you are using.
Build real things and put them online. Not tutorials. Not another to-do app. Pick a real problem and ship something that solves it. One real project beats ten course certificates.
Learn to handle mess. Earlier, juniors got clean tickets. Now they get problems. “Our onboarding drop-off is bad, figure it out.” Turning a vague problem into a shipped feature is the skill that separates juniors who survive from ones who do not.
There is no safe area anymore, only areas where the human who understands the problem deeply is still the one directing the AI. Be that human. Whatever you know today will be half-useless in 3 years — Gartner says 80% of engineers will need to upskill in AI tools by 2027.
The honest truth
The market in 2026 is not the 2021 hiring boom. But it is also not the apocalypse. It is a harder, more selective version of what software engineering always was — a field that pays people who keep learning and can actually think.
If you love building things, none of this should stop you. The ladder got pulled up a few steps. So start climbing higher.
That is it. That is all I have to say. If you are a student reading this, do not panic. Just get to work. The people who do the work will be fine.
If AI is Writing the Code, What’s Left for Us? was originally published in Towards AI on Medium, where people are continuing the conversation by highlighting and responding to this story.
