TL;DR: I used GitHub Copilot for most of 2023 and switched to JetBrains AI Assistant. Copilot’s completions looked impressive but required constant correction. AI Assistant completes less code – but its deep IDE integration delivers more value. Commit messages, code explanations, refactoring, and contextual chat changed how I work more than autocomplete ever did.
The Copilot Honeymoon
Insight: In my experience, first impressions with most AI code tools are misleading. The real test is whether you still accept suggestions after three months.
I bought a Copilot subscription in early 2023. The hype was hard to ignore, and I wanted to see what AI code completion could do inside PhpStorm.
The first impression was genuinely good. Copilot suggested code as I typed, recognized other functions in my open file, and provided completions that felt context-aware. I was hooked.
That honeymoon lasted about nine months.
Code Completion Hits a Wall
By late 2023, the shine had worn off. I spent more time correcting Copilot’s suggestions than I saved by accepting them. The completions were close enough to look right, but wrong enough to need manual fixes on almost every suggestion.
I started ignoring most completions. Typing the correct code myself was faster than fixing generated code.
Copilot’s marketing showed it turning comments into working code. In practice, it usually “completed” my comments into longer descriptions instead of generating code. The feature existed, but it didn’t match the demo.
Meanwhile, I spent more time in ChatGPT, copy-pasting code back and forth for architectural questions Copilot couldn’t handle. That’s a sign. When your coding assistant pushes you toward a separate chat window, it’s not assisting enough.
I Almost Uninstalled JetBrains AI Assistant
Gotcha: If you evaluate AI Assistant only on code completion quality, you’ll miss the features that actually matter.
I got access to JetBrains AI Assistant and it underwhelmed me at first. Where Copilot flooded me with suggestions, AI Assistant held back. Fewer completions, less spectacle.
I opened the Settings > Plugins dialog to uninstall it. That’s when I saw the full feature list below the uninstall button.
Code completion wasn’t the main feature – it was one part of a much deeper IDE integration.
Deep IDE Integration Changes the Workflow
Context: The assistant sees your open file, your project, and your IDE state. No copy-pasting required.
AI Assistant doesn’t just complete code. It understands PhpStorm’s context – your open files, your project structure, your staged changes.
I mapped Cmd+^ (= key above the tab-key on my German keyboard) to open the AI chat sidebar. One keystroke, start typing. To understand an unfamiliar Python script, I opened the .py file, hit the shortcut, and typed: “Explain briefly what the currently opened script is doing. Max 100 words.“
No copy-pasting. No switching to a browser tab. The assistant already knows what file you’re looking at.
After moving a PHP project to a new directory, I had index issues in PhpStorm. Instead of searching through docs, I asked the assistant: “I’ve moved the current project to a new directory. What steps do I need to take?“
What Actually Helps Day-to-Day
Insight: The real value isn’t code generation. It’s having a context-aware assistant that knows your IDE, your project, and your current file
Here’s what I use AI Assistant for on a typical workday:
- Function documentation. It generates complete docblocks including parameter descriptions. This alone saves minutes per file.
- Refactoring suggestions. Select code, ask for improvements. It catches repetitive logic and suggests extraction into functions.
- IDE usage questions. “How do I connect to a local Postgres DB in PhpStorm?” Context-aware answers, no browser needed.
- Code generation. “Write a shell script for X” or “How do I hook into the login workflow in WordPress?” – answered inline.
- Commit messages. I use these less for the message itself and more as a quick summary of my changes for time tracking.
I stopped opening ChatGPT for code questions. AI Assistant gives better answers because it has the context ChatGPT lacks.
The workflow I’ve settled into: discuss a function with the AI in an ongoing chat. Ask it to extend or modify its suggestion. Challenge its choices – “Why did you use X? Can you make this shorter?” Once the code looks right, hit “Insert Snippet at Caret” and do final checks.
Code Completion Is the Wrong Race
If you’re evaluating AI coding tools on completion quality alone, you’re measuring the wrong thing. Copilot wins on volume of suggestions. AI Assistant wins on depth of integration.
I think deep IDE integration matters more than code completion, because completions save you keystrokes while integration changes your workflow. The difference is typing speed versus thinking speed.
If you’re already in a JetBrains IDE, start with AI Assistant’s chat and refactoring features. Ignore the completions for a week. See if the assistant features alone justify the subscription. They did for me.