Skip to content

navveenb/lean-agentic-ai / lean-agentic-ai-claude-implementation/.claude/skills/carbon-estimator

carbon-estimator

Verified

Estimate the carbon footprint of an agentic AI system based on its LLM usage

Carbon Estimator

Estimate the CO₂ emissions of an agentic system based on model usage, call volume, and deployment region.

Lean Principle

  • #6 Emissions don't show up in your logs

Instructions

Step 1: Gather inputs

From the user or from context, determine:

  • Daily LLM call volume (or estimate from code scan)
  • Model distribution (% small, % medium, % frontier)
  • Deployment region (or assume US Average if unknown)

Step 2: Calculate energy

Use these energy-per-call estimates:

Model SizeEnergy per Call (kWh)
Small (Haiku, GPT-4o-mini, Flash)0.001
Medium (Sonnet, GPT-4o, Pro)0.005
Frontier (Opus, GPT-4, o1, Ultra)0.03
Total daily energy (kWh) =
  (small_calls × 0.001) + (medium_calls × 0.005) + (frontier_calls × 0.03)

Step 3: Calculate carbon

Use these regional carbon intensities (gCO₂/kWh):

RegiongCO₂/kWh
Quebec / Norway / Iceland25
France / Sweden70
US West (Oregon)100
Netherlands / UK200
US Average380
Germany350
India700
Poland / Coal regions800
Daily CO₂ (grams) = Total daily energy × Regional carbon intensity
Annual CO₂ (kg) = Daily CO₂ × 365 / 1000

Step 4: Calculate optimized scenario

Assume lean optimizations:

  • Model routing: shift 60% of frontier calls to medium, 30% of medium calls to small
  • Caching: 30% cache hit rate (reduces total calls by 30%)
  • Green region: use the cleanest available region (25 gCO₂/kWh)

Recalculate with these optimizations applied.

Step 5: Write report

Write to reports/carbon-estimate.md with:

  • Current estimated annual CO₂ (kg)
  • Optimized estimated annual CO₂ (kg)
  • Reduction percentage
  • Human-relatable equivalents:
    • Cars driven for a year (÷ 4,600 kg)
    • Flights (÷ 255 kg per domestic flight)
    • Phone charges (÷ 0.008 kg per charge)
  • Top recommendation

Example

Input: 50,000 calls/day, 70% medium, 30% frontier, deployed in US Average.

Output:

Current: 50K calls/day → 212.5 kWh/day → 80.75 kg CO₂/day → 29,474 kg/year
Equivalent: ~6.4 cars driven for a year

Optimized: 35K calls/day (caching) → 28 kWh/day (routing) → 0.7 kg CO₂/day (green region) → 256 kg/year
Equivalent: ~1 domestic flight

Reduction: 99.1%
Top recommendation: Move batch workloads to Quebec (25 gCO₂/kWh)
reverse-engineer-anythingmorluto/reaReverse engineer native, managed, Electron/JavaScript, packaged, firmware, and browser targets with REA, and build agent environments (AgentEnv bundles) of web applications from a live instance or a capture, and of Linux desktop apps with a task from a request. Use shipped-artifact or requested runtime evidence to explain features, compare versions, decompile code, or guide a reconstruction. Skip REA for ordinary source-repository architecture analysis.Stars: 471kUpdated Claude CodeCodexVerifiedInstall
diagnosing-superpowersobra/superpowersUse when a superpowers session went wrong and your human partner wants to know why — repeated work, ignored plans, stumbles, poor results, a skill that didn't fire, "it took too long", "why is it so expensive", "what is it doing" — or wants to build a bug report for the superpowers maintainers, for the current session or a past one identified by id or path, on any harness.Stars: 322kUpdated Claude CodeCodexVerifiedInstall
dispatching-parallel-agentsobra/superpowersUse when facing 2+ independent tasks that can be worked on without shared state or sequential dependenciesStars: 322kUpdated Claude CodeCodexVerifiedInstall
executing-plansobra/superpowersUse when executing an implementation plan in the current session as the implementer yourself — your human partner chose inline execution, or no subagent tool is availableStars: 322kUpdated Claude CodeCodexVerifiedInstall