Rethinking Claude Code
Exploring a better way to collaborate with AI while coding.
A Workflow That Felt Wrong
While vibe coding with Claude Code, I kept opening a second AI chat just to write prompts for the first one.
The irony was that the AI helping me refine prompts knew the least about my project.
It felt like a broken workflow — and became the starting point for this exploration.
Refine
My first idea was intentionally simple.
Instead of opening another AI tool, Claude could help refine the next prompt before it was sent. Using the existing project context, it could clarify goals, surface missing constraints, and turn rough ideas into actionable requests.
Highlights
- No more switching between AI tools.
- Refine prompts using the current project context.
- Surface missing constraints before execution.
But after using this idea for a while, I realized prompt refinement wasn't really about rewriting prompts.
The best prompts aren't written — they're discovered through discussion.
Thinking Space
Instead of improving the prompt itself, I started designing the conversation before the prompt.
Thinking Space became a dedicated workspace where humans and AI could brainstorm, challenge ideas, and converge on a direction before execution.
Highlights
- Shared context, separate history.
- Think while Claude builds.
- Retrieve discussions, not logs.
- From discussion to execution.
Once thinking became a dedicated workspace, another question emerged:
Should every conversation sound the same?
Thinking Lens
Different questions benefit from different perspectives.
Instead of approaching every discussion the same way, Thinking Lens lets you borrow proven ways of thinking — from product leaders, decision frameworks, or your own team's playbooks.
Highlights
- Apply proven thinking models.
- Follow structured decision patterns.
- Challenge assumptions from different lenses.
- Build your own reusable companions.
Final Thoughts
The best prompts aren't written.
They're discovered through discussion.
As AI takes over more execution, designing how humans think with AI may become more important than designing how they talk to it.