After 30+ years coding and 8 years in AI, I’ve learned everyone focuses on which LLM is best, but misses what really matters: knowing how to talk to it properly.
The story
After three decades writing code and eight years building AI solutions, I’ve noticed something: everyone obsesses over which LLM is “best,” but hardly anyone talks about the skills that actually matter.
Here’s the truth – picking the right tool is maybe 30% of the equation. The other 70% is knowing how to talk to it.
The tools are pretty straightforward:
GitHub Copilot and Tabnine crush closed tasks – boilerplate, refactoring, standard docs. Debugging? Give them a stack trace and watch them shine. Open-ended architecture decisions? They’ll struggle unless you break things down first.
But here’s where most people get stuck:
They treat LLMs like magic boxes. One vague prompt, hope for perfection, get disappointed.
The real leverage comes from prompt engineering. I always include detailed context, specific constraints, and clear outcomes. And I never expect a perfect answer on the first try – iterative prompting is everything.
Break big problems into small, well-defined chunks. Ask for a function signature, then implementation, then tests. One piece at a time, with context at each step.
Sometimes I’ll even prompt the LLM to “think out loud” before coding – propose multiple approaches, explain its reasoning. This surfaces better solutions and catches misunderstandings early.
The bottom line:
Modern software craftsmanship isn’t about finding the perfect AI tool. It’s about learning to frame problems so AI can actually help. Chunking. Context. Constraints. And always, always review what it gives you.
Master that, and you’ve got yourself a powerful coding ally.
Recent Comments