agent-estimate

Know what an AI task will cost before you run it.

agent-estimate tells you how long an AI agent will take — and how that compares to doing it yourself.

$ agent-estimate estimate "Implement OAuth 2.0 flow (Google + GitHub)"

Metric	Value
Expected case	75.4m
Human-speed equivalent	190.9m
Compression ratio	2.53x

One command. Three numbers. Now you know whether to dispatch an agent or do it yourself.

See it in action

Single task — coding

$ agent-estimate estimate "Implement OAuth 2.0 flow (Google + GitHub)"

Metric	Value
Task	Implement OAuth 2.0 flow (Google + GitHub)
Tier / Agent	M / Claude
Base PERT (O/M/P)	25m / 50m / 90m (E=52.5m)
Best case	44.7m
Expected case	75.4m
Worst case	117.2m
Human-speed equivalent	190.9m
Compression ratio	2.53x
Review overhead	+15m (standard)
Estimated cost	$1.45

A medium coding task. Your agent finishes in ~75 minutes. Doing it yourself? ~3 hours. (Full output)

Single task — research

$ agent-estimate estimate "Audit dependencies for known CVEs" --type research

Metric	Value
Task	Audit dependencies for known CVEs
Tier / Agent	S / Claude
Base PERT (O/M/P)	10m / 20m / 30m (E=20m)
Expected case	38m
Human-speed equivalent	99m
Compression ratio	2.61x
Estimated cost	$0.55

Research tasks have high human-multipliers — pattern matching across hundreds of dependencies is exactly where agents shine. (Full output)

Multi-agent session — 3 tasks in parallel

$ agent-estimate estimate --file tasks.txt

Where tasks.txt contains:

Implement OAuth 2.0 flow (Google + GitHub)
Write unit tests for OAuth flow
Write API reference for auth module

Task	Tier	Agent	Expected	Human Equiv
Implement OAuth 2.0 flow	M	Codex	52.5m	190.9m
Write unit tests for OAuth flow	M	Gemini	52.5m	190.9m
Write API reference for auth module	L	Claude	100.8m	327.6m

Metric	Value
Wave 0	All 3 tasks in parallel (Claude + Codex + Gemini)
Expected case	131m
Human-speed equivalent	709.5m
Compression ratio	5.42x
Estimated cost	$4.84

Three agents working in parallel. ~2 hours wall clock vs ~12 hours sequential human work. That's the power of fleet estimation — you see the compression before you commit the compute. (Full output)

More examples in examples/ — coding S/M, research, documentation, and multi-agent sessions.

Try it now

pip install agent-estimate

agent-estimate estimate "your task description here"

That's it. No config needed — sensible defaults for a 3-agent fleet (Claude, Codex, Gemini).

What it does

agent-estimate produces three-point PERT estimates calibrated for AI agents, not humans:

• Tier classification — auto-sizes tasks as XS/S/M/L/XL based on complexity signals
• PERT math — optimistic, most-likely, pessimistic with weighted expected value
• Human comparison — multiplier per task type (coding, research, docs) so you see the compression
• METR thresholds — warns when estimates exceed model reliability limits (METR p80 benchmarks)
• Wave planning — dependency-aware scheduling across multiple agents with parallelism
• Review overhead — models review cycles as flat additive cost, amortized per agent per wave
• Modifiers — tune estimates with --spec-clarity, --warm-context, --agent-fit

Supported task types

Type	Flag	What it models
Coding	(default)	Feature work, bug fixes, refactors — tier-based PERT
Research	--type research	Audits, investigations, analysis — flat PERT with depth scaling
Documentation	--type documentation	API docs, guides, changelogs
Brainstorm	--type brainstorm	Ideation, spikes, design exploration
Config/SRE	--type config	Deploys, infra changes, CI/CD work

Integrations

Claude Code Plugin

/plugin marketplace add kiloloop/agent-estimate
/plugin install agent-estimate@agent-estimate-marketplace

Then use directly in Claude Code sessions:

/estimate Add a login page with OAuth
/estimate --file spec.md
/estimate --issues 1,2,3 --repo myorg/myrepo
/validate-estimate observation.yaml
/calibrate

GitHub Action

- uses: kiloloop/agent-estimate@v0
  with:
    issues: '11,12,14'

Full workflow example

name: Estimate
on:
  pull_request:
    types: [opened, synchronize]

permissions:
  contents: read
  pull-requests: write

jobs:
  estimate:
    runs-on: ubuntu-latest
    steps:
      - uses: actions/checkout@v4
      - uses: kiloloop/agent-estimate@v0
        with:
          issues: '11,12,14'
          output-mode: summary+pr-comment

Action inputs and outputs

Input	Required	Default	Description
issues	yes	—	GitHub issue numbers (comma-separated)
repo	no	current repo	GitHub repo (owner/name)
format	no	markdown	Output format: markdown or json
output-mode	no	summary	summary, pr-comment, step-output, summary+pr-comment
config	no	—	Path to agent config YAML
title	no	Agent Estimate Report	Report title
review-mode	no	standard	Review tier: none, standard, complex
spec-clarity	no	1.0	Spec clarity modifier (0.3-1.3)
warm-context	no	1.0	Warm context modifier (0.3-1.15)
agent-fit	no	1.0	Agent fit modifier (0.9-1.2)
task-type	no	—	Task category: coding, brainstorm, research, config, documentation
python-version	no	3.12	Python version to use
version	no	latest	agent-estimate version to install
token	no	${{ github.token }}	GitHub token

Output	Description
report	Full estimation report content
expected-minutes	Expected minutes (when format: json)

Codex Skill

Codex-specific skill at .agent/skills/estimate/SKILL.md. Claude plugin skill at skills/estimate/SKILL.md.

Configuration

Agent fleet

Pass a custom config to model your own agent fleet:

agents:
  - name: Claude
    capabilities: [planning, implementation, review]
    parallelism: 2
    cost_per_turn: 0.12
    model_tier: frontier
  - name: Codex
    capabilities: [implementation, debugging, testing]
    parallelism: 3
    cost_per_turn: 0.08
    model_tier: production
settings:
  friction_multiplier: 1.15
  inter_wave_overhead: 0.25
  review_overhead: 0.2
  metr_fallback_threshold: 45.0

agent-estimate estimate "Ship packaging flow" --config ./my_agents.yaml

Default METR thresholds

Model	p80 threshold
Opus 4.6	90 minutes
GPT-5.4	60 minutes
Gemini 3.1 Pro	45 minutes
Sonnet 4.6	30 minutes
Haiku 4.5	15 minutes

Legacy model keys (Opus, GPT-5/5.2/5.3, Gemini 3 Pro, Sonnet) are still supported for backward compatibility.

Note: Estimates are calibrated against Claude Code (Opus 4.6, High thinking) and Codex (GPT-5.4, Extra High thinking). Different thinking levels or model versions may shift estimates — adjust with --spec-clarity and --warm-context modifiers as needed.

Output formats

agent-estimate estimate "Refactor auth pipeline" --format json    # machine-readable
agent-estimate estimate --repo myorg/myrepo --issues 11,12,14     # from GitHub issues
agent-estimate estimate --file tasks.txt                           # from file

Calibration

Validate estimates against observed outcomes and build a calibration database:

agent-estimate validate observation.yaml --db ~/.agent-estimate/calibration.db

Related

OACP — Coordinate the agents you just estimated. Open Agent Coordination Protocol for multi-agent async workflows.

Contributing

See CONTRIBUTING.md for full workflow and expectations.

pip install -e '.[dev]'
ruff check .
pytest -q

Community

• Code of Conduct
• Security Policy
• Support
• Changelog

License

Apache License 2.0