Adaptive Task & Lifecycle Architecture System
Structured orchestration framework for building companies from zero to scale.
New in v3.0: Clean installation with improved strategy structure (individual step files, high-level tracker)
ATLAS is a three-phase AI orchestration system that transforms strategic vision into systematic execution.
Three AI phases (sequential):
- AI1 Research → Investigates entire substep, produces comprehensive START_HERE
- AI2 Planning → Validates research, creates execution guides for all checkpoints
- AI3 Execution → Implements all checkpoints, produces working deliverables
Not checkpoint-based roles. Linear handoff: AI1 completes → AI2 completes → AI3 executes all checkpoints.
Examples in this framework use: Next.js + Vercel + Supabase
The framework structure (AI1→AI2→AI3 phases, checkpoints, progress tracking) is technology-agnostic. The orchestration system works with any tech stack - Python, Go, mobile, desktop, or any other platform. Examples show web development patterns, but adapt the implementation details to your stack while keeping the framework workflow intact.
See ONBOARDING.md for guidance on adapting to your technology choices.
ATLAS provides orchestration structure, not project-specific guardrails.
You must customize role files to encode YOUR engineering standards as AI instructions:
- Security requirements: HIPAA compliance checks vs reasonable auth vs minimal (internal tools)
- Quality standards: API stability requirements, performance targets, testing depth
- Git workflow: Your specific branching strategy, PR requirements, deployment process
- Documentation needs: API docs, compliance documentation, or minimal internal notes
- Domain-specific validations: What should AI watch for in YOUR project type?
The framework works because you adapt it. Generic guardrails fit no one perfectly. Customized guardrails make AI follow your exact engineering practices.
See ONBOARDING.md - Phase 3 for detailed customization guide with concrete examples.
Perfect for:
- Building companies from 0 to 1
- Multi-month projects with evolving strategy
- Work requiring deep research → detailed planning → precise execution
- Teams needing exact progress tracking and resume capability
- Projects where learning can compound over time (with later efforts becoming more efficient)
Not needed for:
- Weekend projects
- Simple bug fixes
- Well-defined 1-week tasks
- Work with no strategic uncertainty
What ATLAS provides:
- Structured workflow for planning and execution
- Knowledge capture and compounding mechanisms
- Progress tracking and resume capability
- Systematic approach to complex projects
What ATLAS does not guarantee:
- Market fit (you must validate your market)
- Execution quality (requires skilled execution)
- Business outcomes (framework reduces chaos, doesn't ensure success)
- Speed improvements (gains depend on consistent application and learning)
Reality check: Building a successful company is extremely difficult. Most fail due to market factors, not execution frameworks. ATLAS helps with the "how to build" - you're responsible for "what to build" and "who wants it."
New to ATLAS? → QUICK_START.md - 30 minutes to operational
Want complete understanding? → ATLAS_OVERVIEW.md - Complete system explanation
Ready to configure for your company? → ONBOARDING.md - Setup guide
Need technical specification? → ATLAS_SPECIFICATION.md - Complete workflow spec
Validating framework? → ATLAS_VALIDATION.md - Verify correctness
ATLAS3/
├── README.md # You are here
├── ATLAS_OVERVIEW.md # System explanation
├── ATLAS_GLOSSARY.md # Terminology
├── ATLAS_ENTRY_POINT.md # Auto-detection & routing
├── ATLAS_SPECIFICATION.md # Technical spec
├── ATLAS_VALIDATION.md # Validation checklist
├── QUICK_START.md # Fast-track guide
├── ONBOARDING.md # Company setup guide
├── HUMAN_GUIDE.md # Human operational manual
│
├── roles/ # AI role guides
│ ├── AI1_RESEARCHER_GUIDE.md
│ ├── AI2_GUIDE_CREATOR_GUIDE.md
│ ├── AI3_IMPLEMENTER_GUIDE.md
│ ├── AI3_DEVELOPER_BEST_PRACTICES.md
│ └── GIT_WORKFLOW.md
│
├── templates/ # File templates
│ ├── START_HERE_TEMPLATE.md
│ ├── OUTLINE_TEMPLATE.md
│ ├── OVERVIEW_GUIDE_TEMPLATE.md
│ ├── TODO_TRACKER_TEMPLATE.md
│ ├── COMPLETION_SUMMARY_TEMPLATE.md
│ ├── CP_GUIDE_TEMPLATE.md
│ └── EFFORT_TEMPLATE.md
│
├── strategy/ # Strategic layer
│ ├── ATLAS_STEPS_TRACKER.md # High-level status
│ ├── STEP_DEFINITION_GUIDE.md # How to write steps
│ ├── steps/ # Individual step files
│ │ ├── STEP_TEMPLATE.md # Template
│ │ └── (STEP_1.md, STEP_2.md created during setup)
│ ├── ATLAS_ASSUMPTIONS.md # Template
│ ├── LEARNINGS_LOG.md # Template
│ └── DECISION_LOG.md # Template
│
├── context/
│ └── CONTEXT_INDEX.md # Context loading rules
│
└── migrations/
└── MIGRATIONS_LOG.md # Database tracking
Substep Defined (steps/STEP_X.md)
↓
AI1 Research Phase (6-12 hours)
- Researches ENTIRE substep scope
- Produces START_HERE.md
→ Marker: START_HERE.md exists
↓
AI2 Planning Phase (8-15 hours)
- Reads & validates START_HERE
- Creates OUTLINE (3-10 CPs)
- Researches EACH CP deeply
- Creates CP guides (one per CP)
- Creates OVERVIEW_GUIDE, TODO_TRACKER
→ Marker: All guides + TODO_TRACKER exist
↓
AI3 Execution Phase (1-4 weeks)
- Pre-flight checks (git sync)
- Executes CP0 (setup)
- Updates TODO_TRACKER
- Executes CP1 (implementation)
- Updates TODO_TRACKER
- ... continues for all CPs
- Final CP: Testing, PR creation
- Creates COMPLETION_SUMMARY
- Updates ATLAS_STEPS_TRACKER.md + steps/STEP_X.md
→ Marker: COMPLETION_SUMMARY exists, PR created
↓
Human Approves PR
- Merges to main
- Substep complete
↓
Next Substep AI1 Reads COMPLETION_SUMMARY
- Cycle continues
1. Linear AI Handoff
- AI1 → AI2 → AI3 (sequential phases, not role-per-checkpoint)
2. Research Tapers
- AI1: 80% research (deep investigation)
- AI2: 50% research (validation + CP-specific)
- AI3: 20% research (spot-checks + sanity validation)
3. Progress is Always Known
- File existence shows AI phase
- TODO_TRACKER shows CP progress
- ATLAS_STEPS.md shows Step/Substep status
4. Auto-Detection
- "Continue ATLAS on X" → Framework determines phase and CP automatically
- No manual state tracking required
5. Learning Compounds
- COMPLETION_SUMMARY → Next AI1 (efficient handoff)
- LEARNINGS_LOG → Future planning (patterns applied)
- Velocity can improve (Effort 20 may be 2-5x more efficient than Effort 1, depending on execution)
Current: v3.0 Status: Structured framework (clean installation) Designed for: Complex software projects requiring systematic execution discipline (any scale, commercial or non-commercial)
ATLAS v3.0 embodies:
Measure thrice, cut once:
- AI1 researches deeply (measure)
- AI2 plans thoroughly (measure again)
- AI3 executes precisely (cut)
Exact progress tracking:
- Always know: Which Step? Which Substep? Which CP?
- Resume from exact point (never lose state)
Knowledge compounds:
- Each Effort teaches the next
- Templates evolve based on learnings
- Velocity can improve with consistent application
Quality is non-negotiable:
- Security gates (AI3 checks vulnerabilities)
- Testing after each CP (not deferred)
- Human approval for production (final quality gate)
First time:
- Read ATLAS_OVERVIEW.md - 20 minutes
- Read ONBOARDING.md - 30 minutes
- Create your company vision doc - 1 hour
- Create ATLAS_STEPS_TRACKER.md and steps/STEP_X.md files (your strategy) - 1-2 hours
- Start executing - "Run ATLAS to research STEP_1_1"
Quick start:
- Read QUICK_START.md - 10 minutes
- Minimal config - 1 hour
- Start immediately
Companies that benefit from systematic planning and execution discipline, particularly those building complex software products with evolving requirements.
Questions?
- Read ATLAS_SPECIFICATION.md - Complete technical spec
- Check ATLAS_VALIDATION.md - Verify setup
- Review HUMAN_GUIDE.md - Operational procedures
Issues?
- Validate setup (ATLAS_VALIDATION.md Level 1-6)
- Check workflow (specification vs implementation)
- Review examples (see how it should work)
Welcome to ATLAS v3.0. A framework for disciplined company building.