Integrating AI into my workflow
In the past year, I’ve been blending AI into my workflow bit by bit. Starting with ChatGPT, I explored options that could be directly integrated into VSCode, and finally found my match in Raycast AI—powered by GPT-4.
The game-changer came earlier this year when I joined Sana and embarked on a Kotlin adventure. Instead of the typical web search for documentation and tutorials, I turned to GPT. I pasted code snippets from our codebase—making sure to exclude anything sensitive or critical to the business—and asked it to decipher parts of the code and its syntax. While working, I asked it to review my code, provide feedback, and suggest potential refactoring to craft idiomatic Kotlin code. This feedback loop turned out to be quite engaging.
From my perspective, AI shines brightest as a virtual ‘rubber duck’. It listens, decodes, and helps refine ideas, even when they seem abstract. It’s an outstanding refactoring tool, a task that frequently pops up when working on an existing codebase. It enhances my coding efficiency in a similar way to how Grammarly enhanced my English writing. (Although, I’ve since stopped using Grammarly, and now rely solely on Raycast AI commands to correct spelling and grammar errors.)
AI also stands out as a decoder of errors. Ever squinted at TypeScript error types? They can appear as cryptic puzzles. But with AI, they morph into clear-cut instructions.
Possessing coding knowledge is crucial to discern when AI begins to ‘hallucinate’. A quick web search can verify its suggestions or highlight inaccuracies. However, when equipped with clear tasks, AI proves to be a smart assistant, offering solid suggestions.