Phase 1: The Planning Phase
This is the highest-leverage thing you can do. Most people open Claude Code and immediately start prompting for code. That's the wrong move.
The planning phase is where you — the human — bring domain expertise, creative direction, and strategic thinking. Claude Code amplifies what you feed it. If you feed it a vague idea, you get vague code. If you feed it deep research and clear thinking, you get production-ready software.
Don't write code first
Before writing a single line, fire off research subagents. Claude Code can run multiple research tasks in parallel using the Agent tool:
Research the current state of [your domain]. I need to understand: 1. What open-source solutions already exist? 2. What APIs are available for [specific data source]? 3. What are the common pain points users face with existing tools? 4. What would a 10x better solution look like? Search GitHub, npm, and relevant documentation. Give me a comprehensive landscape analysis.
Why this phase matters
When I built the turbulence dashboard for Moomoo's Skills API, I didn't start by coding chart components. I started by understanding:
- What data the Skills API actually provides
- How financial professionals interpret turbulence metrics
- What existing visualization patterns work for time-series financial data
- Where the gaps are in current tooling
This research phase took about 20 minutes with Claude Code's parallel subagents. It would have taken me half a day manually.
The research subagent pattern
Domain deep dive. Have Claude research your problem space. What solutions exist? What are their limitations?
Technical landscape. What libraries, APIs, and data sources are available? What are the integration patterns?
Prior art audit. Search GitHub for similar projects. What worked? What didn't? What can you learn from their issues and PRs?
Tip
The key insight: during planning, you are the subject matter expert. Claude is the research assistant. You direct the investigation based on what you know about the domain. This is where human expertise is irreplaceable.
What "zero code" means
In the planning phase, the only outputs are:
- Research summaries — landscape analysis of existing solutions
- Data source documentation — what's available and how to access it
- Architectural sketches — high-level decisions about stack and approach
- Risk register — what could go wrong and how to mitigate it
No React components. No API routes. No database schemas. Those come later, informed by everything you've learned here.
Based on this research, what are the top 3 architectural approaches for building [project]? For each approach, outline: - Pros and cons - Key libraries/services needed - Estimated complexity - Known risks I'm leaning toward [your intuition]. Challenge that assumption.