From experienced engineer to AI beginner: My unexpected journey

After 15+ years of writing code, I was comfortable with my tools — IDE shortcuts, terminal commands, debugging techniques. Then AI tools appeared, and I became a beginner again.

Starting simple: ChatGPT replaces Google

My first AI interaction was using ChatGPT instead of Google for programming questions. Rather than sifting through Stack Overflow threads, I could ask natural language questions and get direct, working answers. This saved hours each week.

IDE integration challenges: Why Copilot didn’t click

I tried GitHub Copilot in my IDE next. When it worked, it was impressive — suggesting exact functions I needed, or completing complex regex patterns. But the experience was inconsistent.

The autocomplete suggestions often interrupted my thought process. While solving complex problems, Copilot would suggest basic implementations that missed important nuances. It felt like an eager junior developer interrupting with obvious solutions.

More importantly, Copilot lacked project context. I couldn’t ask “How do these classes fit together?” and get meaningful answers. So for daily coding, I disabled most Copilot features and returned to traditional IDE autocomplete.

The breakthrough: Terminal-native AI integration

Then Claude Code came along and changed everything. As someone who lives in the terminal, it integrated seamlessly into my existing workflow without forcing interface changes.

I could ask complex questions about unfamiliar codebases and get intelligent responses based on actual file contents. For Android SDK development — where visual debugging is difficult — Claude Code analyzed code patterns and suggested solutions that made sense.

Context-aware project understanding

The real breakthrough was Claude Code’s project context through CLAUDE.md files. Instead of starting fresh with every question, I provided comprehensive project documentation that Claude Code referenced for every interaction.

This not only improved my experience; it improved the entire team’s experience. When colleagues used Claude Code on our projects, they immediately accessed the same contextual understanding. The knowledge was version-controlled and automatically shared.

Rubber duck debugging, reimagined

Working remotely sometimes means losing valuable debugging conversations with colleagues, or risking disturbing them at inopportune moments when getting stuck. But Claude Code became my patient debugging partner, helping me think through problems without interrupting anyone’s deep work.

Effortless script writing

Claude Code excels at utility scripts. Tasks that took 30–60 minutes now happen in minutes. Some examples are:

• Converting complex legacy CI scripts into local debugging tools, and then feeding improvements back to CI
• Creating customer communication tools for release management
• Setting up development environments (like migrating from Bash to Zsh)

AI handles tedious syntax and edge cases, letting me focus on the actual problem.

Where AI excels and struggles

After months of daily use, I’ve developed a clearer picture of where AI tools genuinely excel versus where they still fall short. Understanding both strengths and limitations helps me set realistic expectations and use these tools more effectively.

The wins

Specification writing — AI helps draft comprehensive specifications and think through edge cases I might miss.

Test creation — Writing comprehensive tests became much faster. Claude understands codebase structure and generates meaningful test cases.

Codebase navigation — For inherited or poorly documented code, Claude excels at tracing component connections and explaining complex interactions.

The limitations

Android UI debugging — Working on Android SDK development, I found AI struggles with UI rendering issues. It makes confident suggestions but can’t properly iterate on them or test the changes. However, it’s excellent at helping discard potential solutions and narrowing down the problem space, while the visual “spark” of understanding complex UI interactions still requires human insight.

Complex legacy codebases — AI struggles with spaghetti code and large interconnected classes where dependencies span multiple layers. It can lose track of complex relationships and provide incomplete analysis. However, it excels at providing insights to help untangle the complexity — identifying key connection points and suggesting refactoring approaches. I’m sure this will improve over the coming years as context windows also increase.

Key takeaways

The AI landscape is evolving rapidly. Six months ago, I barely used AI tools. Today, they’re integral to my workflow. The pace is unlike anything I’ve seen in engineering.

Despite significant productivity gains, I’m still learning. Next, I want to explore AI agents for complex tasks, better context management for larger projects, and specialized commands for specific domains.

For hesitant engineers

If you’re worried that AI will take your job, rest assured: AI isn’t replacing engineers; it’s augmenting our capabilities. Engineers who learn to work effectively with AI tools gain significant advantages, not because AI does their job, but because it amplifies existing skills.

Start with solid fundamentals before diving deep into AI tools. Use AI to supplement your knowledge, but understand the basics first.

The bottom line

Embracing AI tools required finding the right tool that could seamlessly fit into my workflow, as well as admitting I still have a lot to learn. The payoff: faster research, efficient scripting, better documentation, enhanced problem-solving, and a huge reduction in time wasted on boilerplate and menial tasks, as well as unit testing.

If you’re hesitant about AI tools, start small, be patient with the learning curve, and focus on tools that enhance rather than replace your workflow. The combination of engineering experience and AI augmentation makes us more capable than ever.

And if you’re interested in how AI can transform how you work with documents, explore our AI-powered document processor and AI Assistant to see how we’re integrating intelligent capabilities into our products.