The mission-control (Flight Control) project is being closed upstream. This embeds the project files directly in the repo under container/mission-control/, bakes them into the Docker image at /opt/mission-control, and copies them into place at container startup instead of git cloning from GitHub. Also adds missing osc52-clipboard, audio-shim, and triple-c-sso-refresh to the programmatic Docker build context in image.rs. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
4.8 KiB
Flight Control Overview
Flight Control is a methodology for AI-first software development that maintains meaningful human oversight while maximizing AI effectiveness.
Philosophy
AI-First, Human-Guided
Traditional development methodologies were designed for human developers. They assume humans will interpret requirements, make design decisions, and adapt to changing circumstances. AI agents work differently—they excel with explicit structure but struggle with ambiguity.
Flight Control inverts the traditional approach:
- Humans define outcomes, not implementation details
- AI executes implementations, not strategic decisions
- The methodology bridges the gap through progressive specification
Why Aviation Works
Aviation provides a proven model for high-stakes operations where planning and execution are separate concerns:
| Aviation | Flight Control |
|---|---|
| Mission objectives | Mission outcomes |
| Flight plan | Flight specification |
| Flight legs | Implementation legs |
| Pilot authority | Human oversight |
| Autopilot execution | AI execution |
The key insight: pilots don't recompute routes in real-time. They follow pre-computed flight plans while retaining authority to adapt when circumstances demand. Similarly, AI agents shouldn't reinvent architecture with each task—they should execute well-specified legs while flagging issues for human review.
Key Principles
1. Outcome-Driven Planning
Missions start with outcomes, not tasks:
Traditional: "Build a user authentication system" Flight Control: "Users can securely access their accounts with minimal friction"
The outcome framing keeps focus on what matters while leaving implementation flexible.
2. Adaptive Specifications
Flights are living documents. Unlike traditional specs that become stale, flight plans explicitly track:
- Open questions requiring resolution
- Design decisions and their rationale
- Prerequisites and dependencies
- Adaptation criteria (when to deviate from plan)
3. Structured Execution
Legs are optimized for AI consumption:
- Explicit acceptance criteria
- Required context clearly stated
- Expected inputs and outputs defined
- No ambiguity in scope
4. Layered Feedback
Information flows both directions:
- Downward: Missions inform flights, flights generate legs
- Upward: Leg completion updates flights, flight outcomes inform mission status
Comparison to Traditional Methodologies
vs. Agile/Scrum
Agile emphasizes iterative human collaboration. Flight Control complements this by structuring how AI fits into iterations:
- Sprints can contain multiple flights
- Stories map roughly to flights
- Tasks map to legs
Flight Control adds the missing layer: how to specify work for AI execution.
vs. Waterfall
Waterfall assumes complete upfront specification. Flight Control embraces uncertainty:
- Missions can spawn new flights as understanding evolves
- Flights can be modified in-flight when circumstances change
- Legs can be aborted and replaced
vs. CRISP-DM / ML Workflows
Data science workflows focus on experimentation. Flight Control adds structure without eliminating iteration:
- Experimental flights can have "explore" legs
- Failed experiments inform mission outcomes
- Reproducibility is built into leg specifications
The Audience Gradient
A core innovation is the audience gradient—documentation shifts style based on who consumes it:
Human Readable ◄─────────────────────────────► AI Optimized
│ │ │
Mission Flight Leg
│ │ │
Narrative prose Technical spec Structured format
Outcome-focused Checklist-driven Explicit criteria
Flexible scope Bounded scope Fixed scope
This gradient acknowledges that humans and AI have different strengths:
- Humans excel at ambiguity, context, and strategic thinking
- AI excels at following explicit instructions consistently
Flight Control puts each audience where they're strongest.
When to Use Flight Control
Flight Control works best when:
- AI agents are part of your development workflow
- Work benefits from clear specification before execution
- You need traceability from outcomes to implementation
- Multiple people (or AI sessions) contribute to a single outcome
It may be overkill for:
- Quick one-off scripts
- Solo exploratory coding
- Highly uncertain R&D with no clear outcomes