Skip to content

comparison.md

Latest commit

 

History

History
83 lines (53 loc) · 5.06 KB

comparison.md

File metadata and controls

83 lines (53 loc) · 5.06 KB

AI Dev OS in 5 Minutes — Framework Comparison

What is AI Dev OS? — AI Coding Rules Framework

A framework for turning tacit developer knowledge into explicit, enforceable rules for AI-assisted coding. Unlike workflow orchestration tools, AI Dev OS treats the problem as knowledge management.

Quick Comparison

AI Dev OS ContextKit (Anthropic) Spec Kit (GitHub) GSD BMAD bkit
Core idea Knowledge layers with lifespans Structured context documents Spec as source of truth Execution orchestration Multi-agent roles PDCA cycles
Rule organization 4 layers (Philosophy → Decision Criteria → Guidelines → AI Frames) Product/Architecture/Style docs Flat spec/plan/task 3 documents 19 agents, 50+ workflows Phase-based
Tool support Claude Code + Kiro + Cursor Claude Code (primary) Cross-tool (shell) Claude Code only Multi-tool Gemini + Claude
Rule evolution Built-in (ai-dev-os-evolve) Manual Manual Manual Manual Manual
Conflict resolution Specificity Cascade Not defined Not defined Not defined Not defined Not defined
Context overload Two-tier (static 3-5 rules + dynamic full check) Document-level scoping Not addressed Fresh sub-agent Agent specialization Hook-based
Tool migration 75% preserved (L1-L3 portable) Partially portable Spec survives State docs survive Docs survive Session-based
Theoretical basis 16 classical SE theories Context engineering SDD methodology Context engineering Agile + agents PDCA
Setup npx ai-dev-os init (CLI) Manual Markdown creation Shell scripts Prompt-based Multi-step wizard Config files

When to Choose AI Dev OS

Choose AI Dev OS if you:

  • Want rules that survive tool changes (Claude Code → Kiro → Cursor)
  • Need a system that improves over time (Rule Harvesting + SECI spiral)
  • Value "why" over "what" — understanding principles, not just following templates
  • Work across multiple projects and want reusable guidelines

Choose something else if you:

  • Want lightweight context documents without a layered model (→ ContextKit)
  • Want a one-session workflow optimizer (→ GSD)
  • Need multi-agent orchestration for large teams (→ BMAD)
  • Want GitHub-native spec management (→ Spec Kit)

Trade-offs

AI Dev OS is not the right choice for every situation:

  • Upfront investment: You won't see value in the first 30 minutes. Rules need to be written and refined over time.
  • Setup complexity: Even with the CLI, the submodule model requires git knowledge. Non-git workflows are not supported.
  • Rule maintenance: Rules need periodic review. Without discipline, they become stale.

Key Differentiators

1. Rules with Expiration Dates

Every rule has a shelf life. L1 (philosophy) lasts 2-5 years. L4 (tool config) lasts 2-4 months. This prevents rule rot.

2. Bottom-Up Rule Discovery

Don't write rules top-down. Let AI generate code, find gaps, extract rules from real failures. Rules grounded in experience, not theory.

3. "Less is More" for Context

Benchmark data proves too many rules degrade AI output. AI Dev OS loads only 3-5 project-specific files in static context (~8K tokens), then verifies all rules post-generation via dynamic check+fix. This scored 96.9/100 vs 79.3/100 for loading all guidelines.

4. The Sentinel Rule

Security rules with zero violations are working — never remove them for being "unused."

5. AI Cost Optimization

AI Dev OS reduces AI-related costs in three ways:

  • Fewer rework cycles: Specific guidelines + post-generation check catches issues before code review, reducing human review time
  • Token efficiency: 3-5 files (~8K tokens) in static context instead of 28 files (~75K tokens) — 90% token reduction with better results
  • Tool migration savings: 75% of rules survive tool changes (L1-L3). Switching from Claude Code to Cursor doesn't mean rewriting all your coding standards

Design Philosophy: Markdown by Design

AI Dev OS is intentionally composed entirely of Markdown files. This is a deliberate design decision, not a limitation:

  • Transparency: Every rule is human-readable. No compilation, no binary formats, no opaque configurations.
  • Forkability: Anyone can fork, modify, and redistribute rules without learning a proprietary tool or DSL.
  • Tool independence: Markdown works with any editor, any AI tool, any CI system. No vendor lock-in.

The CLI tool (npx ai-dev-os init) automates setup, but the underlying model is still git submodules — rules live in your repository as auditable, diffable text files. The CLI is optional; manual setup is always supported.

Languages: English | 日本語 | 简体中文 | 한국어 | Español