In partnership with

This week, we’re digging into the energy use of AI agents. how designing AI agents—like the ones you’re building or dreaming up—ties into energy use and sustainability. AI’s growing fast, and its thirst for power is real. But don’t worry, it’s not all bad news—there are emerging tools to make affordances to be efficient.

AI Agents And The Energy Crunch

Those clever AI models you’re working with? They’re power hogs. Training something like GPT-4 can burn through as much electricity as 5,000 homes in a year. That’s wild! All that power draw by server farms causes grid instability. This instability is causing strain on our home appliances which causes them to break sooner. As designers, it’s worth thinking about how our agents add to this load.

Grids Under Pressure From AI Infrastructure

Here’s the catch: our electrical grid is old-school. It wasn’t built for the explosion of data centers powering AI. These hubs are popping up faster than ever, and the grid’s struggling to keep up. Some companies are even pitching “AI economic zones” with beefed-up power networks. It’s a challenge—and an opportunity—for us to design smarter systems.

The Big Picture Impacting Health and Environment

There’s a flip side. Data centers pumping out AI magic can also churn out air pollution. Operation and cooling data centers can contribute 300 metric tons of CO2e, which is 0.6% of global greenhouse gas emissions. That leads to stuff like asthma or worse, especially for folks—often low-income communities—living nearby.

A study by UC Riverside and Caltech estimates that by 2030, AI-related air pollution could lead to 1,300 premature deaths annually in the U.S., with health costs approaching $20 billion per year, including cancers and asthma Study on Air Pollution and Public Health Costs of AI.

The applications we design and develop appear clean as there is no tailpipe on our computers to see the emissions when we run complex AI agent computing tasks like on our cars when we accelerate, but these studies punctuate that IT infrastructure is having a large negative real-world effect. We need the tools to measure and reduce these emissions.

Flipping The Script To AI for Climate

Now, the cool part: AI agents can help solve this mess! They can optimize energy use, predict solar or wind power output, as designers we can design affordances for how much compute to solve a problem, or choose where we compute a problem locally or in the cloud. Google’s already using AI to make its wind energy 20% more valuable. So, while AI’s a power hog, it can also be a sustainability champ— if we make these fundamental design and architectural choices.

How Do We Build AI Agents With Energy Responsibly?

AI agents are game-changers, but their energy appetite needs some taming. As designers and technologists, we can balance the awesome with the responsible—building tools that push boundaries and protect the planet.

It isn’t just about optimizing server infrastructure it is also about how designers can create resource transparent features. Here are a few examples of tools and examples that can help you design your AI agents to be more energy conscious:

Anthropic token budgeting

Claude’s new tool that lets the user choose how many tokens to use on a task. This is akin to how humans work — some tasks you need to quickly accomplish and others you have more time to be thoughtful. This capability is both a productivity benefit allowing user control of deep versus quick work, but has the benefit of controlling energy used on tasks. This trait of energy or cost control could further be detailed for user awareness and control.

Carbon-Aware SDK

This toolkit helps developers build AI agents that schedule tasks based on the grid’s carbon intensity, like running heavy jobs when renewables are dominant.

The Carbon-Aware SDK is a tool that enables developers to create applications, including AI agents, that are carbon-aware Carbon-Aware SDK by Green Software Foundation. AI agents built using this SDK can schedule their tasks based on real-time grid carbon intensity data, such as using Electricity Maps to delay computations to greener grid periods, reducing emissions by up to 15% for machine learning workloads.

OptiCloud Emissions Reduction Startup

OptiCloud helps organizations achieve substantial reductions in both energy consumption and carbon emissions. One notable example is a healthcare company that successfully reduced their carbon emissions by over 30% and saved $1 million in annual cloud bills by implementing Optic Cloud's solutions.  

Practical Steps To Build Carbon Aware AI Agents

1. Integrate Energy Maps into AI Agent Design:
   Use tools like Electricity Maps to monitor grid carbon intensity and schedule AI tasks during low-carbon periods. For example, an AI agent managing a smart home can schedule energy-intensive tasks like charging EVs during greener grid times, reducing emissions.
   

2. Leverage Carbon-Aware SDKs:
   Adopt the Carbon-Aware SDK by the Green Software Foundation to build AI agents that automatically adjust their computation based on carbon intensity data. This SDK provides APIs and libraries in 40 programming languages, making it accessible for developers to create carbon-aware applications GitHub - Green-Software-Foundation/carbon-aware-sdk.

3. Scope and Scale Control of Energy and Token Usage:
   Build features that allow users to scope the depth and breadth of AI agents may leverage in completing user requests. These tools should provide users transparency in how much effort in the form of tokens and energy are to be expelled in service of a task or series of tasks. Claude’s token budgeting to break thinking into blocks that can support token budgeting.
   

4. Optimize AI Models for Efficiency:
   Use energy-efficient AI models and techniques, such as deploying models on edge devices with TensorFlow Lite, which reduces the need for energy-intensive data center operations Machine Learning CO2 Impact Calculator. This can lower the computational load and carbon footprint of AI agents.

Your turn: 

How can we design AI agents to be more sustainable? Hit reply—I’d love to hear your ideas!

There’s a reason 400,000 professionals read this daily.

Join The AI Report, trusted by 400,000+ professionals at Google, Microsoft, and OpenAI. Get daily insights, tools, and strategies to master practical AI skills that drive results.

Sign up now for free and work smarter, not harder.

🧲 AI Agent Magnet

This isn inspiring framework for how teams can design startups using the emerging capabilities of AI agents. Greg’s concept seeds fall into themes related to Dharmesh Shah’s inspiration that SMB will be redefined to mean small mighty business inspiring the next generation of small business owner leveraging AI agents.

An open-source interface toolkit that can serve as a foundation to build agentic systems. This interface toolkit will make it easy to pair this interface layer with agent builders, LLMs, knowledge hubs and tools making it easy to execute highly polished agent features.

This course delves into the inner workings of a new class of agentic coding, exploring how context awareness, tool integration, and intent tracking power platforms like Windsurf. Understand the challenges of code search, learn advanced retrieval techniques, and build a mental model of how these powerful agents operate. Windsurf has one of the most powerful, capable IDEs capable of helping you go from idea to web app or mobile app in hours.

💬 Suggestion box

A newsletter exploring the principles that will help us design AI agent experiences and startups that amplify human creativity.

Subscribe to join a community of designers and developers shaping purposeful AI agents.

New Referral Rewards

Until next time, keep innovating and stay curious!

Jason