I might not be a programmer anymore by 2027

Yesterday, Andrej Karpathy tweeted something that hit different. The idea that “programmers” as we know them might not exist in the traditional sense anymore.

Even Boris Cherny, the creator of Claude Code, echoed this sentiment, emphasizing how fundamentally the craft is shifting.

Strongly agree. I built Claude Code, and even I'm surprised by how fast things are moving. The way we write software is fundamentally changing.

— Boris Cherny (@bcherny) December 2025

And honestly? I’ve been feeling this shift for a while now.

This isn’t a doom-and-gloom post. It’s just me, sitting here at the end of 2025, trying to make sense of what the hell just happened in the past two years.

2025 Is a Brutal Year

Before we dive into the personal stuff, let’s talk about the elephant in the room: 2025 has been absolutely brutal for workers.

Here in Indonesia where I live, the creative industry is facing mass layoffs as AI overtakes traditional roles. One DPR member reported that nearly 60% of employees at her friend’s agency company were laid off in 2025, replaced by AI. Sixty percent! That’s not a trim, that’s a massacre.

Some argue companies are using AI as a convenient scapegoat for post-pandemic over-hiring corrections. Maybe. Probably even partly true. But even if that’s the case, the direction is clear: AI is changing how many humans companies need.

And that brings us to why this feels so personal.

AI shifting

Back in May 2023, I wrote a post about Key Points when Learning Large Language Models. At that time, I was genuinely excited and … dumb :)

ChatGPT was still fresh, LangChain was the hot new thing, and we were all trying to figure out how to build chatbots with “prompt engineering.” We felt like wizards. Powerful. In control.

Looking back, that feels like a lifetime ago. Two years. That’s all it took for the entire landscape to flip upside down. Remember when GitHub Copilot first dropped? We were amazed by tab completions. “Wow, it can autocomplete my function!” We thought that was peak AI assistance.

Then came Windsurf, Trae, the damn Cursor and Antigravity. Tools that don’t just complete your code, they write entire features based on a prompt. You paste an image of a UI mockup, and it generates the component. You describe a bug, and it finds and fixes it across multiple files.

And now? We have CLI agents. Claude Code, Codex, OpenCode, tools that can navigate your entire codebase, run tests, commit changes, and even create pull requests. You give them a task, walk away to make coffee, and come back to a working implementation. Holy shit.

We went from “AI helps me code faster” to “AI codes, I review.” That escalated quickly, to be honest.

What started as curiosity about LLMs turned into watching my entire profession get redefined in real-time. The way we write code, the tools we use, even the fundamental question of “what does a programmer actually do?”, everything changed. And I’m still not sure if I should be excited or terrified. Probably both :(

The Ceiling Is Too Damn High

Here’s the uncomfortable truth that nobody wants to talk about: the barrier to entry for programming just got obliterated, but paradoxically, it’s never been harder for newcomers to stay relevant.

Let’s take a look at this chart from @erikbryn’s research paper:

See that blue line at the bottom doing a sad nosedive? That’s Early Career 1 (ages 22-25) and it’s been in freefall since mid-2022. While every other age group has either held steady or grown, the youngest developers are getting squeezed out of the market.

The dashed vertical line marks roughly when ChatGPT launched. Coincidence? I really, really wish it was.

What’s happening is a compression of the entry-level job market. The tasks that used to be training grounds for junior developers like writing boilerplate code, fixing simple bugs, implementing straightforward features, are exactly the tasks AI handles best. Companies that used to hire three juniors to support one senior now hire one senior who can leverage AI to do the work of four. Great for efficiency. Terrible if you’re 23 and just graduated with a CS degree and student loans.

Meanwhile, look at the purple and brown lines: Mid-Career 2 (41-49) and Senior (50+) are actually growing. Experience, context, and architectural knowledge are becoming more valuable, not less. These are the people who know what to build and why — skills that AI can’t easily replicate. Yet.

Think about it. If someone with zero coding experience can now build and ship a functional app in a weekend using AI tools (and they do, every single weekend, my Twitter timeline is proof), what’s the value proposition for a junior developer who spent years learning the fundamentals?

The ceiling is so high now. Senior engineers with deep system knowledge? More valuable than ever. But early-career developers doing low-context work? That’s exactly the work AI handles best. It’s a brutal reality, and I genuinely feel, feel sooo bad for CS or related fresh grads who are entering the job market today.

The Dilemma Nobody Talks About

Back in day, my job was pretty straightforward: I spent roughly 80% of my time actually writing code, and the remaining 20% reviewing other people’s work. That was the rhythm. Code, review, repeat. Simple. Predictable. Comfortable.

Now? It’s completely different. I barely write code manually anymore, maybe 10% of my time at most. Instead, I spend about 20% of my day prompting and guiding AI, crafting the right instructions to get the output I need. The bulk of my work, easily atleast 50%, is working alongside AI assistants, reviewing their suggestions, iterating on outputs, course-correcting when they go off track. And that remaining 20% is still code review, but even that’s increasingly assisted by AI tools that catch things before I do.

This is why I said “programmer” in the title. By 2027, I might not “write” code and “program” anymore. But I’ll still be a Software Engineer. Someone who understands engineering principles, system design, scalability, trade-offs. Someone who can architect a project and then hand it off to AI assistants to implement. The job title stays. The actual work changes completely.

Another things that gigled me up is at work, we’re pushed to use AI tools to be more productive. “Adopt or get left behind.”. But here’s the catch: if AI makes each developer 3-5x more productive, does the company still need the same headcount? The math is brutal. Companies might lay off 50-70% of their tech employees and still ship the same amount of work. And it’s not hypothetical anymore, it’s happening.

Meanwhile, non-programmers are now building and launching products faster than ever. Every week on Twitter/X, there’s a new indie hacker who “built this in a weekend with AI” and it’s actually good. The barrier to building software has collapsed, and suddenly everyone can compete. The tech bubble on social media is flooded with launches, side projects, SaaS products, all built at lightning speed by people who couldn’t code two years ago.

It’s overwhelming. And a little bit funny in a sad way.

My Personal Take

I’m not going to pretend this year was easy.

Professionally, 2025 has been full of ups and downs. Job security feels different now. The dilemmas are real: Am I still relevant? How do I stay competitive? What skills actually matter anymore? These questions keep me up at night.

I’ve taken weeks off from social media multiple times this year. Not because I wanted to touch grass (okay, partly), but because watching everyone else move so fast was genuinely stressful.

The constant stream of “I shipped X in 24 hours with AI” posts creates this pressure that’s hard to escape. When you have a family, limited time, and responsibilities outside of work, it’s tough. You feel like you’re falling behind. Behind on news, behind on technology, behind on ideas. Everyone else seems to be building the future while you’re just trying to keep up.

And here’s the thing nobody talks about: to actually be productive with AI, not just vibe coding or generating AI slop, but producing real, production-quality work, you need to pay up. Cursor Pro, Claude Pro, ChatGPT Plus, API credits for your agentic workflows, we’re talking $50+ per month minimum. If you don’t subscribe, you fall behind. If you do subscribe, there’s this pressure to justify the cost by being constantly productive. It is a “Pay to win” in a way, sadly.

So What’s the Plan?

Actually, I don’t really have the answers. I think nobody does. We’re all figuring this out in real-time, together, confused but trying our best.

But here’s my plan: by 2026, I’m shifting focus. Less coding, more architecting. Less implementation, more review. Less typing, more thinking. The goal is to become the kind of engineer who designs systems before AI implements them, someone who breaks down problems, hands off implementation to AI, and focuses on review and validation.

I’ll still be a Software Engineer. Just not a “programmer” in the traditional sense.

And honestly? That’s kind of exciting. Terrifying, but exciting. Like riding a roller coaster that was built by AI and you’re not 100% sure it was tested properly.

By 2027, I might not write a single line of code. But I’ll still be engineering software. I’ll still be solving problems. I’ll still be building things that matter. Just… differently.

See you on the other side. Hopefully we’ll both still have jobs by then. hahaha