Estimate AI-assisted Development Work Accurately

This skill estimates AI-assisted and hybrid human+agent development work using research-backed formulas with PERT statistics, confidence bands, and calibration feedback loops, producing statistical estimates rather than gut feelings. It follows a seven-step process: mode detection (human-only, hybrid, or agent-first team), task classification by size (XS-XL), complexity, and risk, formula application using research-backed multipliers, PERT calculation for expected values via three-point estimation, confidence band generation (P50, P75, P90), output formatting for Linear, JIRA, ClickUp, GitHub Issues, Monday, or GitLab, and calibration that feeds actual completion times back to improve future estimates.

Use it when estimating development tasks where AI agents handle part of the work, for sprint planning with hybrid human+agent teams, for batch sizing a backlog of 5 or 500 issues, for staffing and capacity planning with agent multipliers, or for release date forecasting with confidence intervals.

For commitments, use the P75 or P90 confidence band rather than P50 - relying on P50 for a hard deadline is a known pitfall that produces overconfident estimates. The skill asks clarifying questions when team composition or agent-usage percentage is missing, so be explicit about those inputs rather than letting it guess. Re-calibrate when team composition or tooling changes significantly, since stale calibration data degrades estimate accuracy. Treat its output as a statistical estimate to validate against your team's actual delivery patterns, not a guaranteed deadline.

Input: a single task, a batch of backlog tickets, or a feature description, ideally with team composition and AI-agent usage percentage specified. Output: PERT-based expected-value estimates with P50/P75/P90 confidence bands, formatted for the target project-tracking tool.

"We have 3 developers using AI agents for ~60% of implementation. Estimate this feature."

Other example requests include a single task ("Estimate building a REST API with authentication using Claude Code") and a batch of tickets ("Estimate these 12 JIRA tickets for our next sprint"). Best-practice guidance recommends starting with a single task to calibrate before moving to batch mode, and using "instant mode" for quick T-shirt sizing when a full PERT analysis isn't needed.

Formats output for Linear, JIRA, ClickUp, GitHub Issues, Monday, and GitLab. Chains with @sprint-planning, @project-management, and @capacity-planning skills.

Engineering managers and team leads doing sprint planning, backlog sizing, or release forecasting for teams that mix human and AI-agent development work, who want statistically grounded estimates instead of ad hoc gut-feel sizing.