After 30 years coding and 8 in AI, I learned there’s no “best” coding assistant – just the right match for your task. Context matters more than capabilities, and mastering how you work with AI beats picking the fanciest tool.
The story
After three decades writing code and eight years deep in AI, I just cracked something important: there is no “best” LLM or coding assistant. There’s only the right match for your specific task.
Here’s what I mean.
When I’m wrestling with a massive legacy codebase – the kind with quirky conventions and nonstandard patterns – I need tools like Augment Code that can ingest the entire monorepo, build artifacts, even PR history. Context is everything. The assistant that surfaces suggestions aligned with my actual stack beats the one trained on vanilla best practices every time.
For mainstream languages and rapid prototyping? Copilot or Tabnine work great. But if I’m in a niche stack, I want something I can steer with custom instructions – Cursor with config files.
Security matters too. On sensitive projects, I only use assistants with on-prem deployment or no-train SaaS options. My code doesn’t become training data for someone else’s model.
The real unlock isn’t picking the fanciest tool. It’s mastering how to work with AI.
Prompt engineering changed everything for me. I treat prompts like briefing a teammate: specific, with background and constraints. Not “Write a sorting function,” but “Write a Python quicksort for integers, O[n log n], no recursion, returns new list.”
I break big problems into chunks. First design the data model, then write the API endpoints. This cuts hallucination and makes outputs testable.
And I never trust AI blindly. I review everything, then refine with targeted follow-ups: “Refactor for readability,” “Document public methods,” “Check for vulne
Recent Comments