AI in the Developer's Toolkit

I love AI, and I have plenty of opinions about it. Those opinions recently led me to help develop an initiative within my organization: empower our engineering team to adopt AI in a responsible and productive way.

It’s been an interesting journey—one that mirrors the broader shift I’ve noticed among engineers everywhere.

From Fear to Empowerment

Like many developers, my first reaction to AI was a quiet fear. Would tools like ChatGPT, Copilot, or Claude eventually make engineers like me obsolete? After all, if a machine can write code faster, cleaner, and with fewer mistakes, what’s left for us?

But over time, my perspective has changed. I now see AI not as a threat, but as a powerful accelerator of software development. When used responsibly, AI has the potential to increase the need for engineers rather than replace them. It can free up time, create new opportunities for experimentation, and even reshape the role of a developer into something closer to a hybrid of engineering and product ownership.

That shift will primarily benefit product owners with technical chops and software engineers with a product-focused mindset. In other words, engineers who can think beyond syntax and algorithms, and connect the dots between business goals, user experience, and technical execution.

The Tools I’ve Tried (and Why I Landed on Claude)

If AI is now a requirement in a modern dev’s toolkit, then naturally I’ve tried almost everything out there.

• Cursor was my first real AI development assistant. I loved it—until my IDE started hoarding all my memory and slowing me to a crawl. That was a deal breaker.
• Claude Code has become my favorite. It integrates seamlessly into the command line while still playing nicely with my IDE. I especially appreciate how easy it is to add and maintain rule sets, and the ability to initialize a CLAUDE.md file for an existing codebase. Sure, it’s just a fine-tuned prompt under the hood, but it’s an elegant feature that makes my workflow smoother.
• Gemini CLI and Copilot didn’t blow me away compared to Claude. That said, Copilot’s integration with GitHub is naturally strong, and I’m exploring ways to use it in automation. For example, I’m experimenting with setups where AI can automatically submit pull requests for unit tests if it detects missing coverage in recently changed lines.

Beyond Code Agents: The Rise of MCPs

AI isn’t just about coding assistants—it’s also about how developers interact with systems. That’s where Model Context Protocol (MCP) servers come in.

I’ve used MCPs extensively, especially product-specific ones like GitHub and Figma. These provide structured, context-rich interfaces for AI to interact with services directly. My organization even shipped an MCP server for our own API, and I wouldn’t be surprised if this becomes the norm across the SaaS space.

Why? Because AI is quickly becoming the default interface for users. MCP servers bridge the gap, letting AI agents connect with tools and APIs in a safe, consistent, and useful way.

Building an AI-Friendly Codebase

Of course, none of this works well if your codebase isn’t set up to play nicely with AI. The most important habit for any engineer is to treat AI as a tool, not a crutch. “Vibe coding” with AI is fun—but it doesn’t belong in production.

Pragmatic, targeted use of AI is the way forward. And that requires a codebase optimized for AI assistance:

• Good hygiene and best practices. Clean, well-structured code makes it easier for AI to understand intent.
• Unit tests. These provide a critical feedback loop, letting AI validate its own output and giving developers confidence in the results.
• Clear documentation. Good docs provide context—one of the most important signals an AI system can use to generate relevant, reliable code.

In short: context and feedback are the beating heart of an AI-integrated codebase. Without them, you’re asking AI to guess. With them, you’re setting up your team (and your AI tools) for success.

Looking Ahead

Whether AI leads to more jobs, fewer jobs, or just different jobs, one thing is undeniable: it’s here to stay. For engineers, it’s no longer an option to ignore it. The real question is how to adopt it in a way that makes you more effective—not just as a coder, but as a product-minded problem solver.

For me, that means experimenting with tools, sharing what works, and helping my team find the balance between speed and responsibility. AI isn’t going to replace us—but it is going to change us. And honestly? I’m excited about that.