agent-estimate 0.0.1

Effort estimation for AI coding agents — PERT + METR + wave planning

agent-estimate

The first open-source effort estimation tool built for AI coding agents.

Not story points. Not human-hours. Wall-clock time for multi-agent parallel execution.

Why

Every team using Claude Code, Codex, Cursor, or similar AI coding agents needs a way to estimate how long agent-driven work will take. The current answer is "guess." We replace that with a calibrated, methodology-backed tool.

What It Does

• PERT for AI agents — three-point estimation (Optimistic / Most Likely / Pessimistic) adapted for AI agent task types
• METR-aware — flags tasks exceeding the reliable-completion threshold for the target model
• Wave planning — dependency-aware parallel scheduling across multi-agent fleets
• Self-calibrating — tracks estimated vs actual completion times, auto-adjusts defaults

Quick Start

pip install agent-estimate

# Estimate from a description
agent-estimate "Build a REST API with auth and role-based access control"

# Estimate from GitHub issues
agent-estimate --issues 119,122,125 --repo myorg/myrepo

# Estimate from a spec file
agent-estimate --file docs/mvp-spec.md

# Custom agent fleet
agent-estimate --config agents.yaml "Build a REST API"

Agent Config

agents:
  - name: claude
    capabilities: [reasoning, orchestration, refactoring, architecture]
    parallelism: 1
    model_tier: heavy
  - name: codex
    capabilities: [shell, patching, config, scripts, crud]
    parallelism: 1
    model_tier: standard

settings:
  friction_multiplier: 1.15
  inter_wave_overhead: 15  # minutes between waves
  review_overhead: 0.10    # 10% for self-review

Status

Pre-release — architecture designed, implementation starting. See project plan for details.

License

MIT