Your Thinking Is Only as Good as Your Instructions
- 3 minsWhen Everything Lives in Your Head
When I coded myself everyday, I moved fast. When I implemented my own ideas, I didn’t need to make them fully explicit. The logic lived in my head. Gaps filled themselves. Decisions felt obvious in the moment.
Documentation can be sloppy. Planning was minimal. It worked—because I was both the thinker and the executor.
Then I started vibe coding.
The Hidden Tax of Ambiguity When Execution Is No Longer Yours
Because AI doesn’t share your intuition. It doesn’t “just get it.” When instructions are vague, AI doesn’t “fill in the gaps” the way you intend—it fills them in its own way.
That leads to:
- Subtle misalignments in logic
- Wrong assumptions baked into the implementation
- Rework that feels confusing
And suddenly, things break in ways that are hard to explain.
I found myself reacting the same way over and over: “That’s not what I meant.”
But the reality was simpler: I hadn’t said clearly what I meant.
Lesson 1: Planning Became Non-Negotiable
Before working with AI agents, I could get away with partial clarity. I would figure things out while coding. Now, that approach breaks down.
Because the AI executes immediately, any ambiguity in my thinking gets translated into concrete (and often incorrect) output.
So I had to change my process.
Instead of jumping in, I started:
- Writing structured instructions (and now there are skills!)
- Defining inputs and outputs clearly
- Specifying constraints and edge cases upfront
In other words, I started planning more thoroughly.
Lesson 2: Documentation Became Part of the Interface
Documentation is for human, and the AI too.
What I document:
- The purpose of a function
- The assumptions behind a model
- The structure of a pipeline
….
And turn it into reusable context:
- I can paste it into prompts
- I can refine it incrementally
- I can maintain consistency across iterations
It reduces blurriness incredibly.
From “Doing” to “Thinking Clearly”
Working with AI agents forced me to confront a hard truth: I used to confuse speed of execution with clarity of thought.
But now, execution is cheap. The AI can generate code instantly.
So the real differentiator becomes:
- How clearly I can define a problem
- How precisely I can communicate intent
- How well I can anticipate edge cases
This is less about coding and more about structured thinking.
A New Kind of Discipline
I’ve developed a simple rule for myself: If I can’t explain it clearly, I’m not ready to prompt it.
That means:
- Writing a short plan before asking the AI to implement
- Breaking problems into smaller, well-defined steps
- Keeping a running document of decisions and changes
It feels slower at first. But it eliminates the back-and-forth, the confusion, the subtle bugs that come from misalignment, and of course, it leads to less frustration.
The Irony of Vibe Coding with AI
Vibe coding suggests looseness, intuition, flow.
But when you introduce AI into that process, the opposite becomes true:
The more powerful the tool, the more precise you need to be.
AI doesn’t replace discipline—it punishes the lack of it.
Closing Thought
In the age of AI, the hardest and most valuable skill is thinking in a way that can be understood by something that doesn’t share your intuition.
And that’s what vibe coding with AI ultimately taught me:
Clarity is no longer optional. It’s the interface.
Anqi Zhu
A curious scientist with a prior knowledge in Biostatistics.