Building a Carbon Impact Agent: Boosting ROI with Sustainable AI in...
summary
Shows a working prototype Carbon Impact Agent built as a Langraph agent that ingests per-request telemetry (tokens, duration, region, task) and uses a carbon calculator plus a cost estimator and optimizer to report per-task CO₂ and water use and suggest fixes. Cites concrete numbers: GPT‑3 training ~552 tons CO₂, BloombergGPT training $2.7M over 53 days on 64 servers with 8 A100 GPUs, and an estimated ~47 million tons CO₂ per year from GPT deployment; prototype surfaces savings like caching, region routing, prompt and model choices.
AI-written summary · about
chapters (19)
- 0:00Introduction
- 1:55Why sustainable AI matters
- 3:41AI’s role in energy and sustainability
- 7:16Macro trends in AI and energy
- 10:52Data center power and supply chains
- 14:04AI growth and energy use
- 16:51AI applications and carbon impact
- 18:16Sustainable AI design
- 23:14Fine-tuning and model efficiency
- 24:40Data workflow optimization
- 26:50Agent architectures and MCP servers
- 28:15Model evaluation metrics
- 33:14Tools for carbon impact
- 35:45Developer influence on sustainability
- 37:56Quantifying AI’s carbon footprint
- 40:23Building a carbon impact agent with Langraph
- 45:19Demo: Carbon impact agent
- 48:12Future improvements and personas
- 50:42Summary
related talks
47mThe Rise of AI: A Reality Check on Generative AI
06mThe Dark Side of AI - Environmental & Social Costs Developers Must Know
01h 12mChatGPT and OpenAI API solutions: 👑 My Favourite Implementations
30mBuilding Reliable AI Agents: An Open Source Approach To Evaluation and Observability
40mMaximizing impact: Measuring the ROI of GenAI in the SDLC
16mPath to a More Artificially Intelligent Future
27mSoftware engineering with GenAI
01h 23mBuild your own AI Agent