We're searching for the best way to get working code from LLMs.
To avoid falling into the vibecoding trap, some developers are turning into structured approaches.
The other day, while catching up with some friends, one of them preached Spec Kit and its Specify/Plan/Tasks/Implement process like the holy Gospel. That's GitHub's approach to coding with AI.
More recently, I found this great read on the real purpose of AI:
It includes these four questions to regain control after drifting from too much prompting:
Do I understand exactly the specifications I'm trying to implement, or the bug I'm trying to solve?
Do I have an exact plan for implementing my changes?
What is the current abstraction level to which I should be prompting now?
Which other information am I lacking?
Spec Kit and those four questions capture the essence of coding.
So to get better result from AI and other tools, we have to do what we're supposed to do as coders in the first place:
- Understand the problem to solve
- Decompose that problem into smaller ones
- Think at the right level of abstraction
- Ask enough clarifying questions
Ironically, AI is making us go back to the mindset we should have never left behind. AI just has revealed who has good coding hygiene.
Context comes before coding—with or without AI.
That's why one of the lessons I included in Street-Smart Coding is "Don't rush to code." (That's on Chapter #3.) That's even more helpful with AI as your coding assistant.
Top comments (0)