Software engineering

The familiar sound of the keyboard, the pleasant sight of VS Code on the screen, and a collection of carefully selected extensions - this was my programming sanctuary just a year ago. Like many developers, I had my workflow perfected down to the smallest detail. The thought of letting AI into this carefully crafted space seemed not only unnecessary but also caused some strange anxiety. After all, why fix something that isn’t broken? Why let a bot into my world?

The problem is that in our semi-world built around software development, you can rarely stay in your comfort zone for long. I got a challenging project that would question everything I thought I knew about my coding process and problem-solving. It required quick switching between different technologies, navigating many unknowns, and constantly learning new things.

My trusted tools suddenly seemed insufficient, like trying to climb a major mountain peak in flip-flops. That’s when I first considered using an AI code completion tool. Initially skeptical, I quickly realized the benefits these tools offered in increasing coding productivity, especially through intelligent code suggestions and support for various programming languages, frameworks, and libraries I was using for the first time.

---

First Lesson: Understanding AI

My adventure with AI development tools began with GitHub Copilot. I remember looking at the first AI code completion suggestion with a mixture of fascination and curiosity. As an AI coding assistant, it offered intelligent code suggestions, task automation, and even error identification. The suggestions were surprisingly good - sometimes unnervingly so - but I was haunted by questions:

• Can I trust these suggestions without checking every line?
• Will relying on AI make me a worse programmer?
• What about code security and quality?
• Am I risking becoming dependent on these tools?

These weren’t just technical concerns; they touched something deeper - my identity as a programmer and the pride I took in manually creating code. Thinking about all this, I tried to understand AI and how it could help me in my daily work.

---

First Dance: AI and Me

The turning point came during a particularly demanding sprint. I was juggling multiple tasks, each from a different scope. I started using AI more boldly and using it for routine activities so I could focus on the broader context.

GitHub Copilot: Became my partner in daily coding

I remember how I approached it with reservation at first - “what’s this bot going to suggest to me?” Now I can’t imagine a day without its support. I especially appreciate it when I need to write another similar piece of code - Copilot seems to read my mind and suggests exactly what I intended to write.

It’s incredibly useful for writing tests too. It can suggest test cases I wouldn’t have thought of, or that would simply have taken me longer to devise. And documentation? Well, I admit - it’s not my favorite part of programming. Fortunately, Copilot helps me with this too, suggesting sensible descriptions for functions and methods.

What amazes me most is how quickly it can generate implementations of popular functionalities. I no longer need to google “how to sort an array in JavaScript” or “how to validate an email”. Copilot just knows and suggests a solution that I can adapt to my needs.

ChatGPT: Evolved into my debugging partner and learning accelerator

Often when I’m working on something, my brain produces hundreds of ideas and questions. ChatGPT is something I bounce all this off of, and just writing with it helps me organize the mess in my head. I’ve also learned to use it to guide me to my goal when I start wandering and going off on tangents.

I especially appreciate its help when diving into new technologies. Instead of trudging through dense documentation, I can simply ask “explain this to me like I’m a child” and get an accessible explanation. And when I’m trying to understand code written by someone else? ChatGPT can “translate” it into simpler language and explain what the author meant.

Interestingly, I’ve also started using it as my rubber duck when making architectural decisions. I can discuss different approaches with it, and it asks relevant questions that help me look at the problem from a different perspective. Sure, I always make the final decision myself, but this ability to analyze different options with AI is awesome.

Claude: Became my specialized research assistant

Since discovering Claude, my project work has become not only more efficient but also more interesting. Claude not only understands project specifics but can apply them practically. And all this because you can easily create a project there and feed it with knowledge that Claude will use during interactions.

I most appreciate its ability to analyze documentation, guidelines, and requirements. When I give it specifications or architectural guidelines to “study,” it can extract the essence and propose solutions that actually fit the context. These aren’t just generic answers from the internet, but thoughtful suggestions considering the project’s specifics.

When I need an implementation example, Claude doesn’t just throw a ready-made code snippet. Instead, it creates examples that perfectly fit into the existing architecture and project conventions. It’s like having an assistant who knows exactly what code style we write in and what patterns we use. Of course, you have to feed it code examples first.

Perplexity: Research and learning assistant

In my AI toolkit, Perplexity holds a special place. It’s my assistant that not only finds information but also shows where it comes from - something I often missed in other AI tools.

I especially appreciate it during research on new technologies. Instead of digging through dozens of Google pages, I can simply ask Perplexity about a specific technology and get condensed knowledge from reliable sources. Moreover, I can always check the linked materials to ensure the information is current and reliable.

Perplexity has also become my inseparable companion in tracking technology trends. Sometimes I ask it for opinions about a new framework or library and always get different perspectives, backed by specific sources.

There’s also one thing I really, really like: the voice function. I often use car time for learning - I ask Perplexity questions and listen to answers as if I had a private tech podcast. It’s a great way to productively use time in traffic or during longer trips.

---

First Breath: Finding Balance

Today, my work style looks completely different than a year ago, but the basic principles remain the same. I’ve learned that there’s no point in blindly accepting or rejecting AI. You need to find a balance that enhances capabilities while maintaining expert knowledge. Choosing the right AI tool can significantly help developers and provide extended coding and problem-solving capabilities.

Here’s what I’ve learned over the last year of experimenting with AI:

AI can handle:

• Routine code generation
• Initial versions of test cases
• Documentation templates
• Basic error analysis
• Quick research tasks
• Providing learning content
• Supporting me in daily work and giving hints

I want to maintain full control over:

• System architecture decisions
• Security-critical code
• Business logic acquisition and implementation
• Code quality standards and code review
• Final decisions on how something should be implemented

---

Practical Tips on Getting Started with Code Generation

If you’re considering incorporating AI into your programming workflow, here’s what I recommend:

Start Small

• Start with one tool (I recommend GitHub Copilot)
• Use it first for non-critical tasks
• Gradually expand as you build confidence in using AI coding assistant to increase productivity and code quality

Verify Everything

• Always review AI-generated code
• Test thoroughly
• Understand what AI wrote and what you’re implementing

Set Boundaries

• Define clear rules for when to use AI
• Maintain confidentiality of sensitive information
• Maintain your coding standards

Never Stop Learning

• Use AI as a learning tool
• Question suggestions
• Stay up to date with tool capabilities

And most importantly: find someone to experiment with so you can exchange opinions and experiences.

---

The Road Ahead

The integration of AI in developers’ work isn’t just a trend - it’s a transformation we can’t avoid.

AI can help with routine tasks and speed up work, allowing us to focus more on solving complex problems.

For me, this journey has led to a more efficient and enjoyable development experience. I no longer fight against AI tools but work alongside them, using their strengths to complement my own expertise.

Whether you’re a skeptic like I once was, or you’re already experimenting with AI tools, I encourage you to find your own balance. Start small, stay curious, and remember that these tools are meant to enhance, not replace, your programming skills.

P.S. If you found this article helpful, consider subscribing to my newsletter, where I share more insights about balancing technical excellence with sustainable development practices.