Claude Code Sessions
Jimmy's usage patterns and preferences for Claude Code, Anthropic's AI coding assistant.
Overview
| Attribute |
Description |
| Tool |
Claude Code (Anthropic) |
| Usage |
Daily coding and research |
| Model Preference |
Opus for complex, Sonnet for routine |
| Context |
Terminal-based AI assistant |
Usage Patterns
Model Selection
| Task Type |
Preferred Model |
Rationale |
| Deep research |
Opus 4.5 / Opus 4 |
Best reasoning, longest context |
| Daily coding |
Sonnet 4 |
Fast, capable, cost-effective |
| Quick tasks |
Sonnet 3.5 |
Speed priority |
Session Characteristics
- Extended sessions: Multi-hour deep dives
- Research-heavy: Company analysis, market research
- Code generation: Full-stack applications
- Documentation: Creating structured reports
Session Summaries
Typical Research Session
- Initial query: Broad topic exploration
- Deep dive: Follow-up questions, data gathering
- Synthesis: Compile findings into structured format
- Output: Markdown reports, wiki entries
Coding Session Flow
- Requirements: Describe project goals
- Architecture: Design system structure
- Implementation: Generate code iteratively
- Testing: Verify functionality
- Documentation: Add comments and README
Key Preferences
Communication Style
- Direct: Prefers concise, actionable responses
- Technical: Comfortable with detailed technical depth
- Structured: Likes organized, hierarchical information
Output Formats
- Markdown for documentation
- Tables for comparisons
- Code blocks with syntax highlighting
- Diagrams (ASCII or Mermaid)
Notable Sessions
2026-04-02: Agentic Commerce Deep Dive
- Duration: Multi-hour
- Topics: Tempo, MPP, AI search, company research
- Output: Research diary with multiple company deep-dives
- Models used: Opus 4.6, MiniMax M2.7, Kimi K2.5
2026-04-04: Open Interest & Leverage
- Topic: Crypto derivatives mechanics
- Output: Detailed explanation of OI, liquidation, PFOF
- Use case: Understanding market structure
Tool Integrations
Preferred Stack
| Category |
Tools |
| Search |
Perplexity via Tempo MPP |
| Coding |
Claude Code |
| Memory |
OpenClaw wiki system |
| Communication |
WhatsApp, Telegram |
Workflow Integration
Research need → Claude Code session → Synthesize →
Save to memory → Compile to wiki → Share insights
Performance Observations
Model Benchmarks (Jimmy's Testing)
| Model |
Simple Query |
Complex Reasoning |
| MiniMax M2.7 |
1.8s |
1.8s |
| Kimi K2.5 |
1.8s |
1.7s |
| Gemini 3.1 Pro |
1.1s |
1.8s |
Preferences Emerging
- Speed matters for routine tasks
- Quality matters for research/analysis
- Willing to switch models for optimal cost/performance
Related
- claude-code — Tool overview
- anthropic — Company behind Claude
- ai-coding-tools — Comparison with Cursor, Copilot, etc.