Beyond Copilot: Chasing the Dream of AI-Augmented Coding

Can an AI truly understand your entire codebase? Is it possible to have a coding assistant that learns your team’s best practices? What if your AI helper could see beyond the current file?

AI is revolutionizing coding. A recent StackOverflow poll revealed that 44% of software engineers already use AI tools in their development processes, with another 26% planning to join soon. I’ve been riding this AI wave, primarily with GitHub Copilot, but lately, our relationship has been on the rocks. I’ve found myself hitting the off switch more often, frustrated by suggestions that miss the mark or contradict themselves in the next line.

This dissatisfaction has led me down a fascinating path: building a local Copilot alternative using Ollama. The goal? A code assistant that truly “gets” our entire codebase, one that’s tuned to recommend our team’s latest guidelines and patterns instead of mimicking outdated code. Imagine onboarding new developers with an AI that guides them towards best practices, not just the first example they stumble upon. Picture getting suggestions based on the context of your entire project, not just the file you’re currently editing.

Now, I’m diving into the world of fine-tuning, RAG (Retrieval-Augmented Generation), and other cutting-edge AI techniques to build something similar. It’s a journey of constant learning, with new AI concepts popping up faster than I can implement them. The road ahead is long, and we definitely need more GPUs to get there. But the potential is enormous. A truly context-aware coding assistant could revolutionize how we write, understand, and maintain code.

This year, I’ve set myself an ambitious goal: to replace 80% of my work with AI. It’s a bold resolution, but one that I believe will push me to explore the full potential of these technologies. Whether it’s through enhancing existing tools or building new ones, I’m committed to leveraging AI to dramatically boost my productivity and transform my coding practices.

Are we on the brink of a new era in software development? Or are we chasing a silicon dream? As I continue this exploration, one thing is clear: the future of coding is AI-assisted, but the form that assistance will take is still very much up for grabs.

Are you satisfied with current AI coding tools, or are you also dreaming of something more? And how much of your work do you think AI could realistically take over this year?