pre.dev offers two specification levels. The right choice depends on your project’s complexity, timeline, and how much guidance you want build agents to have.
Fast Spec (Plan)
Structure: Milestones → User Stories
A high-level spec that gives agents the big picture without granular subtask breakdowns. Agents figure out implementation details autonomously.
Best for:
- MVPs and prototypes
- Personal/side projects
- Rapid iteration workflows
- Solo developers
- Projects where speed matters more than detailed planning
What you get:
- High-level milestones with complexity estimates
- User stories for each feature
- Basic acceptance criteria
- Technical architecture overview
Cost: ~3-8 credits | Speed: ~30-60 seconds
Deep Spec
Structure: Milestones → User Stories → Granular Subtasks
A comprehensive spec with implementation-level detail. Every task is broken down into specific, actionable subtasks with their own acceptance criteria.
Best for:
- Production applications
- Complex architectures
- Team coordination
- Mission-critical systems
- When you want maximum control over what agents build
What you get:
- Detailed milestones with phased delivery
- User stories with comprehensive acceptance criteria
- Granular implementation subtasks
- Task-level complexity estimates
- Detailed technical architecture with schema designs
Cost: ~15-40 credits | Speed: ~2-5 minutes
Comparison
| Feature | Fast Spec | Deep Spec |
|---|
| Milestones | Yes | Yes |
| User Stories | Yes | Yes |
| Acceptance Criteria | Basic | Comprehensive |
| Granular Subtasks | No | Yes |
| Architecture Depth | Good | Comprehensive |
| Effort Estimates | High-level | Per-subtask |
| Documentation Scraping | Yes | Yes |
Decision Framework
Use Fast Spec if:
- You’re validating an idea quickly
- The project is straightforward (CRUD apps, simple tools)
- You trust agents to make implementation decisions
- You want to iterate on the spec after seeing initial output
Use Deep Spec if:
- You’re building something you’ll ship to real users
- The project has complex business logic
- You want precise control over each implementation step
- You’re working in a regulated domain (healthcare, finance)
- Multiple people need to understand the build plan
Combining Both
A common pattern is to start with a Fast Spec to validate the overall direction, then generate a Deep Spec once you’re confident in the approach. The Deep Spec can reference the Fast Spec as context.
You can also generate a Fast Spec and selectively request deep breakdowns for only the complex milestones — keeping simple features at the story level while getting granular subtasks for the tricky parts.