Surviving the AI Revolution as a Junior Developer

Aug 9, 2025 - ⧖ 3 min

As a Jr Developer, I've been navigating this AI revolution, which honestly makes it both exciting and intimidating. The landscape is changing so fast that I feel like I'm always catching up, but I'm doing my best to adapt and learn. Here's what I've learned about surviving in this new era.

A Historical Overview:

  1. The chat era (2022-2023) - ask, copy and paste. If you are lucky, maybe it will work.
  2. Code completion arrives (2023-2024) - the AI understands the code and tries to learn the patterns. Useful.
  3. The agents (2025) - It can write, test and debug. The editor now tries to predict the next codebase edition.

At this time, if you are a developer, you probably heard about or use Agentic Coding: Cursor, Claude Code, Windsurf, Devin and more. These tools are awesome and empower people to do more and quickly. On the other hand, these tools also generate technical debt, overengineering and nonsense. Yet the good news: It's more about you and how you use it.

Here's the thing:

1 - As Fabio Akita says: "don't outsource your decisions"

Coding is not an end in itself but a means to build something, and you are responsible for understanding the best way to achieve the goal. The system's business rules are not something esoteric, and an agent is not a fortune-teller.

2 - We have to learn, we have to study

As a Jr developer, every day is a class and every problem is a lesson. We have to police ourselves. AI cannot be a crutch, but a teacher that can help you when you know which question to ask. And I know it's hard, but there isn't much more that I can say about it.

3 - A little bit of code

If you're comfortable asking AI to do a task, break it into small pieces, so you can read, read, read, review (probably correct something) and commit. We are the ones in charge of the code. Your codebase doesn't need nonsense code.

4 - Be clear

The way you communicate with AI is part of the result you will achieve. We need to be clear and logical, indicating what we expect and the limitations it will face in the process.

My Take

I'm curious and eager to see the next innovations in AI. In the last five months, the agentic tools have achieved another level. Today these tools are really useful, the state of the art. But if you are starting to learn coding today, disable the code completion and think by yourself (sometimes it's hard and frustrating, I know). The only way to learn to code is by coding. Do it: write your bad code, fix it, and do it again until you get it right.

I'm thinking of writing more about practical tips with these tools—things like cursor rules and automations that actually make a difference. The future of AI? That's probably worth a whole separate discussion. But for now, this feels like a good start.