AI Assisted Building
Praxis was built in collaboration with Claude Code—demonstrating human-AI partnership in modern software development.
Project Overview
A real-world example of human-AI collaboration
Effective human-AI collaboration requires clear role definition, where humans provide judgment and creativity while AI handles execution and pattern recognition. Praxis is a living demonstration of this principle in action.
A site teaching AI prompting techniques should demonstrate those techniques in its own creation. Praxis isn't just theory—it's proof that effective human-AI collaboration produces real, working software that meets professional standards.
The Story Behind Praxis
Practicing what we preach
Solo Development
Traditional- Developer handles all tasks alone
- Repetitive work drains creative energy
- Cross-file changes prone to errors
- Documentation often neglected
- Slower iteration on complex features
Human-AI Collaboration
Praxis Method- Human directs, AI executes efficiently
- AI handles repetitive implementation
- Consistent cross-file updates
- Documentation maintained continuously
- Rapid iteration with human oversight
Human Vision, AI Execution
Every feature, design decision, and architectural choice originated from human creativity and judgment. AI helped implement these visions efficiently, but the direction always came from human intent.
Claude Code
Praxis was built primarily using Claude Code—Anthropic's AI coding assistant. It handles file operations, code generation, refactoring, and complex multi-file changes while the human maintains creative control.
Full Transparency
We don't hide AI involvement. Every git commit includes "Co-Authored-By: Claude" attribution. The collaboration is documented, traceable, and open for examination.
Quality Standards
AI assistance doesn't mean lower quality. Praxis achieves 100% Lighthouse scores, A+ security ratings, and WCAG AA compliance. The results speak for themselves.
What Claude Code Does
AI capabilities in the development workflow
Claude Code is a command-line AI assistant that can read, write, and modify code across your entire project. Here's how it contributed to Praxis:
Code Generation
Writing HTML pages, CSS styles, JavaScript functions. Claude generates code that follows established patterns and conventions in the codebase.
Refactoring
Restructuring code for better organization, updating patterns across multiple files, consolidating duplicate logic into reusable functions.
Cross-File Changes
Adding navigation to all 25+ HTML pages, updating consistent patterns, ensuring changes propagate correctly throughout the project.
Bug Fixing
Identifying issues, understanding root causes, implementing fixes while considering edge cases and maintaining existing functionality.
Documentation
Writing comprehensive content, creating glossary entries, documenting code patterns, and maintaining handoff documents for session continuity.
Security Compliance
Ensuring CSP compliance, avoiding inline scripts/styles, checking for security vulnerabilities, maintaining A+ security rating.
The Collaboration Model
How human and AI work together effectively
Human: Strategy & Vision
Deciding what to build, why it matters, how it should feel. Setting priorities, making trade-offs, defining success criteria. The creative and strategic direction is entirely human.
AI: Implementation & Execution
Writing code that matches the vision, handling repetitive tasks, maintaining consistency across files. Executing the plan efficiently and accurately.
Human: Quality Judgment
Reviewing AI output, accepting or rejecting changes, catching errors, ensuring the result matches intent. Final approval always rests with the human.
AI: Pattern Recognition
Understanding codebase conventions, maintaining consistency, applying established patterns to new features. Learning from context to produce coherent code.
Human: User Empathy
Understanding user needs, accessibility requirements, real-world constraints. Making decisions that serve actual people, not just technical elegance.
AI: Comprehensive Coverage
Remembering to update all related files, checking edge cases, maintaining completeness. Reducing human cognitive load on details.
What AI Doesn't Do
Understanding the limits of AI assistance
AI is a powerful tool, but it's not magic. Understanding what AI can't do is as important as knowing what it can. Here's where human judgment remains essential:
Original Ideas
AI doesn't conceive new features or decide what should exist. The vision for Praxis—making AI accessible to everyone—is entirely human. AI implements; humans imagine.
Aesthetic Judgment
AI can generate code for a design, but it doesn't know what "looks good" or "feels right." Color choices, spacing, visual hierarchy—these require human aesthetic sense.
User Understanding
AI doesn't know your users. It can't feel the frustration of a confused beginner or the impatience of an expert. Human empathy shapes user experience.
Business Decisions
What to prioritize, when to ship, what trade-offs to accept—these are human calls. AI provides options; humans choose directions.
AI-generated code can have bugs, introduce security issues, or miss requirements. AI code generation requires human oversight for quality assurance. Every change is reviewed before acceptance. Trust but verify—the same principle that applies to any collaborator.
The Development Workflow
A typical session building Praxis
1. Context Loading
Claude reads the HANDOFF.md file to understand current state, recent changes, and established patterns. This ensures continuity across sessions.
2. Task Definition
Human describes what needs to be done: "Add a performance analysis page explaining our optimization approach." Clear intent, flexible execution.
3. Exploration
Claude reads relevant files to understand existing patterns, finds similar implementations, and gathers context for the task.
4. Implementation
Claude generates code, creates files, makes edits. Progress is visible in real-time. Human can intervene or redirect at any point.
5. Review
Human reviews changes, tests functionality, checks visual appearance. Adjustments requested as needed. Nothing merges without human approval.
6. Commit
Changes committed with descriptive messages and Claude attribution. Clear history of what changed and why.
Code Quality with AI
How we maintain professional standards
AI-assisted development can produce excellent code when the collaboration is structured well. Here's how Praxis maintains quality:
Established Patterns
CLAUDE.md defines coding standards, file organization, and conventions. AI follows these rules consistently. New code matches existing style automatically.
Security Constraints
CSP compliance rules are non-negotiable. Claude knows not to use inline styles or scripts. Security is built into the workflow, not bolted on.
Incremental Changes
Small, focused commits rather than massive rewrites. Each change is reviewable. Problems are caught early and fixed easily.
Testing in Browser
All changes are verified in actual browsers. AI doesn't test—humans do. Visual review catches issues that code review might miss.
Performance Monitoring
Lighthouse audits run regularly. Performance regressions are caught and fixed. The 100% score is maintained through continuous attention.
Documentation
Code is self-documenting through clear naming and structure. HANDOFF.md captures decisions and context for future sessions.
Praxis by the Numbers
Current project metrics (updated February 2026)
HTML Files
30+ pages — Learn pages, tool pages, resource pages, all with consistent navigation, accessibility, and structure.
CSS Lines
12,000+ lines — Complete design system covering all components, responsive layouts, animations, visualization charts, and accessibility features.
JavaScript Lines
9,000+ lines — Interactive tools, search functionality, quiz engine, accessibility panel, and all site behavior.
Glossary Terms
5,324+ terms — Comprehensive AI/ML glossary covering concepts from A to Z, all searchable with internal cross-references.
ChatGPT Guide
1,700+ lines — Comprehensive guide with 9 major sections, citations from Stanford HAI, MIT CSAIL, and NIST.
Citations
40+ sources — Verified references exclusively from .gov and .edu domains including NIST, NSF, Census Bureau, Stanford, MIT, and CMU.
Lessons Learned
Insights from building with AI
Clear Context Matters
AI works best with clear constraints and established patterns. The CLAUDE.md file and HANDOFF.md system dramatically improved consistency and reduced errors.
Review Everything
AI makes confident mistakes. Always verify output against intent. Catch problems early when they're easy to fix. Trust the process, not the output blindly.
Humans Stay Essential
AI accelerates implementation but doesn't replace judgment. The hardest problems—what to build, why, for whom—remain deeply human questions.
Iteration Works
Perfect first drafts are rare. Ask for adjustments, refinements, alternatives. The collaboration improves through multiple passes, just like human teamwork.
Building Praxis taught us exactly what we now teach: effective AI collaboration requires clear communication, structured methodology, and iterative refinement. Explicit communication patterns improve collaboration outcomes. The same CRISP principles that make prompts better also make development better.
Full Transparency
Open about AI involvement
We believe transparency about AI involvement builds trust and advances the conversation about human-AI collaboration. Here's how we maintain openness:
Git Attribution
Every commit message includes "Co-Authored-By: Claude" when AI assisted. The git history is a complete record of collaboration.
Open Source
Complete source code on GitHub. Anyone can examine how the project was built, what patterns were used, how problems were solved.
This Page
We don't hide AI involvement—we document it. This page explains exactly how Praxis was built and what role AI played.
Honest Limitations
We acknowledge what AI can't do well. Transparency includes being honest about limitations, not just capabilities.
See the Code
Examine the collaboration yourself. The entire Praxis codebase is open source and available for inspection.
Sources
- Generative AI in Software Engineering - Carnegie Mellon SEI
- AI-Augmented Software Engineering - Carnegie Mellon SEI
Every aspect of Praxis's development is documented in the GitHub repository, including the HANDOFF.md files that maintain session continuity, the CLAUDE.md instructions that guide AI behavior, and the complete git history with Co-Authored-By attribution on AI-assisted commits.