The Role of AI in Development
2024 has been a transformative year for coding productivity, largely thanks to the rise of AI tools. This post reflects on my experiences with AI helpers and integrated development environments (IDEs) and how they’ve shaped my workflow.
My Journey with IDEs: From JetBrains to VS Code (and Back Again)
I’ve been a loyal user of JetBrains IDEs, particularly PHPStorm, for over seven years. JetBrains tools are productivity powerhouses, and while they come at a cost, their features—like built-in support for WordPress hooks—have been invaluable for my work.
In contrast, I’ve never warmed to VS Code. Despite its popularity as a free tool, it’s not a full IDE out of the box. Setting it up requires finding and installing third-party extensions, which has consistently left me frustrated.
This background on IDE preferences sets the stage for my exploration of AI tools.
Experimenting with AI Tools
Early Days: Copilot in PHPStorm
In 2023, I integrated GitHub Copilot into PHPStorm. It provided a solid productivity boost, allowing for:
- Auto-completing code snippets.
- Generating long arrays without typing every element.
- Reducing the need to look up every function.
While helpful, it was far from writing autonomous code. Copilot acted more as an assistant than a creator.
Switching to JetBrains AI
At the start of 2024, JetBrains launched their AI offering. It worked similarly to Copilot but included extra features, such as generating meaningful commit messages based on revision context. This small improvement alone made my commit messages far better than the generic “work in progress” placeholders I’d used before.
Satisfied, I dropped Copilot and committed to JetBrains AI.
The Game-Changer: Sonnet 3.5
Everything changed in mid-2024 with the release of Sonnet 3.5. Its capabilities sparked discussions in the “Build in Public” community, where people claimed they were building entire apps without writing a single line of code. Having used AI assistants for two years, I was sceptical but intrigued enough to explore further.
A Shift Away from WordPress
September 2024 brought unexpected drama in the WordPress ecosystem. Actions by its leader, Matt Mullenweg, destabilised the community, prompting many developers—myself included—to consider alternative tech stacks. This shift coincided with my deep dive into Sonnet 3.5 and its potential to streamline coding outside the WordPress world.
The combination of new tech stacks and Claude’s Sonnet 3.5 proved to be a far more capable coding environment. By this point, I had subscribed to Claude Pro and was ready to experiment further.
Exploring AI Tools Beyond the Basics
Discovering CodeBuddy
I wanted to integrate Claude’s Sonnet 3.5 directly into PHPStorm and found a solution with CodeBuddy.ca. Initially, it was impressive but expensive, as using it all day required purchasing substantial credits.
However, I discovered the risks of auto-generating code:
- It’s easy to go in circles with vague prompts.
- Short prompts waste credits without delivering useful results.
- AI can accidentally break working code, creating more problems than it solves.
Codebuddy is powerful – but I’m not 100% convinced.
Returning to Copilot (Briefly)
When Copilot announced support for Claude Sonnet 3.5, I resubscribed. Unfortunately, while Claude could be selected in GitHub, it wasn’t available in PHPStorm. This limitation made Copilot impractical for me, and I moved on.
Building My Workflow
By late 2024, I had developed a hybrid workflow using Claude Desktop and PHPStorm:
- Screen Setup: On one side of my 49-inch screen, I have Claude Desktop open. On the other, PHPStorm. (Insert Screenshot: Claude Desktop and PHPStorm side-by-side on a 49-inch monitor)
- Claude Projects: I upload project documents and guidance prompts to a chat thread, using up to five files per chat. This ensures clear context for each step, resulting in more accurate code generation.
- Checking Code: For existing files, I use PHPStorm’s scratch file feature to check AI-generated code. I then perform a line-by-line comparison to ensure no errors are introduced.
Revisiting Cursor
Despite my dislike for VS Code, I gave Cursor (a fork of VS Code with built-in AI) another shot. After about three hours of configuration, I got Cursor to function similarly to a JetBrains IDE. While it couldn’t match JetBrains’ code refactoring capabilities, it was a close contender. However, like CodeBuddy, Cursor struggled with auto-patching code—great when it worked but disastrous when it didn’t.
Lessons Learned and Final Thoughts
As 2024 comes to an end, I’ve settled into a workflow that combines:
- Claude Desktop Projects for context-rich AI assistance.
- JetBrains IDEs for precise, manual control over my code.
This balance of automation and manual input has proven to be the most reliable and productive approach for me. AI is a powerful tool, but its limitations require careful oversight and clear guidance from the developer.
Reflecting on AI’s Future
Looking ahead, I’m curious to see how AI tools evolve. Will they move closer to autonomous coding, or will their role remain that of a highly capable assistant? For now, the human touch is still essential for quality and accuracy in software development.
What’s your experience with AI tools in coding? Share your thoughts in the comments below!
Leave a Reply