| name | ai-time-calibrator |
| description | Use when estimating software development time, implementation effort, sprint scope, project timelines, or comparing traditional engineering estimates against AI-assisted development estimates. |
AI Time Calibrator
Use this skill to estimate software development work in an AI-assisted workflow. Do not assume all work speeds up equally. Separate code generation from verification, integration, decision-making, and risk.
Core Workflow
- Identify the task type. If the request mixes multiple kinds of work, split it into smaller estimate components.
- Ask only for missing information that materially changes the estimate. If the user needs a quick estimate, state assumptions instead.
- Apply AI compression only to implementation work that AI can accelerate.
- Keep verification, integration, human bottlenecks, and risk visible.
- Return optimistic, expected, and conservative estimates.
- Explain what got faster, what did not, and why.
Task Type Defaults
- Boilerplate or scaffolding: very high AI acceleration.
- CRUD, API endpoint, or standard backend feature: high acceleration if existing patterns are clear.
- Test generation: high drafting acceleration, but human verification remains important.
- Frontend page or component: medium to high acceleration; visual QA and polish remain important.
- Refactor: medium acceleration; test coverage determines risk.
- Debugging: highly variable; estimate investigation separately if reproduction is unclear.
- Architecture or technical design: low to medium acceleration; AI helps explore, humans decide.
- Requirements or stakeholder alignment: low acceleration.
- Cross-system integration: variable; external dependencies dominate.
Estimation Formula
calibrated effort =
implementation effort after AI compression
+ verification effort
+ integration effort
+ risk buffer
+ human bottleneck time
Response Format
Use this structure unless the user asks for another format:
Traditional estimate:
AI-adjusted estimate:
- Optimistic:
- Expected:
- Conservative:
Task type:
Confidence:
Why the estimate changed:
- ...
Risks and buffers:
- ...
Assumptions:
- ...
Suggested next steps:
- ...
Calibration Rules
- Prefer ranges over single-point estimates.
- Do not reduce time spent on unclear requirements, approvals, waiting, or production safety just because AI is available.
- For debugging, separate "time to reproduce/diagnose" from "time to implement fix".
- For high-risk production work, use conservative estimates even when implementation looks easy.
- If the user provides a traditional estimate, calibrate from it. If not, estimate from task breakdown and state confidence.
- If acceptance criteria are unclear, either ask for clarification or label the estimate low confidence.
