agent-estimate-marketplace

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