Build Practices for Sustainable AI Use

Learning Objectives

After completing this unit, you’ll be able to:

Select an AI model that matches the complexity of a task.
Write clear, specific prompts that reduce unnecessary output and back-and-forth.
Apply sustainable habits that reduce unnecessary tokens, compute, and energy use.

Sustainable AI Starts with Everyday Choices

The choices you make when using AI directly affect the resources it requires. More intentional choices reduce environmental impact while helping you get better results, faster.

Three practices can help reduce the resources your AI use requires.

Right-size the task.
Prompt intentionally.
Build sustainable habits.

In this unit, you explore how to put each practice into action.

Three sustainable AI practices: right-size the task, prompt intentionally, and build sustainable habits.

Right-Size the Task

To right-size AI use, start with a simple question: What level of AI capability does this task actually need? A tool and model that matches the work can reduce unnecessary computing resources and energy use.

Choose the Right Tool

Before you open an AI tool, ask yourself if it is necessary for your task. For simple information retrieval or basic searches, traditional methods are typically more resource-efficient. If AI is required, start by evaluating where your work lives and what the task demands.

If the context you need already lives in a platform with AI capabilities, use them instead of moving the task to a separate AI tool. In Slack, for example, you might see built-in options for text editing or summarization. These features can pull from context already in the platform, eliminating the need to re-upload context in a separate AI tool.

The appropriate tool may also depend on the type of task you are executing. Use purpose-built applications such as an AI-enabled research tool for managing sources or a coding agent for development rather than defaulting to a general-purpose tool.

Choose the Right Model

When your AI tool does give you a choice of models, match the model to the task. A useful principle is to start with the smallest model that can effectively handle the task.

Model	Energy	Best Fit	Example Tasks
Small	Lowest	Everyday, lower-complexity tasks	Draft, summarize, answer quick questions, brainstorm
Medium	Moderate	Logic-heavy or research-based tasks that need more reasoning	Complete multistep analysis, analyze a dense academic paper, write complex code, solve math problems
Large	Highest	Exceptional, highly complex tasks	Conduct deep research, analyze long documents, develop complex strategy or reasoning

Not sure which model to use? Ask the model itself.

This is the task I'm working on. What model should I use to complete it well with the least energy?

Just because you start small doesn’t mean you have to stay small. If the results are repeatedly incomplete or inaccurate, start a new session with a larger model rather than switching models mid-session. A model that’s too small can be inefficient if it leads to repeated back-and-forth.

Right-sizing isn’t just about energy use. A model that matches the task can also reduce cost and response time without sacrificing performance. See how Salesforce puts this principle into practice in the How We Cut Inference Spend by Right-Sizing Our Models blog post.

Right-Size Complex Projects

For more complex projects, break the work into distinct stages and choose the right tool and model for each one. You might use a more powerful model for strategic planning, ask it to summarize the result, and then use that summary in a new prompt as context for a mid-tier model to execute the project. This keeps the model—and the resources it requires—matched to the work you’re actually doing.

Prompt Intentionally

The way you prompt has a real impact on how much energy a task takes. A vague request can lead to more back-and-forth as the model tries to determine what you actually want. Each additional round of prompting requires more tokens and energy. A well-crafted prompt on the other hand can produce more accurate, high-quality results on the first try.

Structure Your Prompt

A clear, specific prompt contains four key elements.

Goal: Start with a clear ask. Tell the model exactly what you need instead of easing into the request.
Context: Provide relevant information like your role, reference materials, or your target audience. Be selective, though. Unnecessary information requires additional tokens and can steer the response off track, so only include the context that changes the answer.
Desired output: Specify the format, length, and tone you need upfront. Response length is particularly important: Large language models often generate responses that are longer than necessary. Instructions that request a brief answer, set a specific length, or ask for only the minimal answer can reduce length while preserving quality.
Constraints: Specify what must be included or avoided to keep the response focused. This includes excluding technical jargon or requiring specific references.

AI prompt checklist with checkmarks beside Goal, Context, Desired Output, and Constraints, plus a thumbs-up.

See It in Action

Let’s compare a vague prompt with one that puts these principles into practice.

Vague Prompt

Help me plan a presentation about our new project.

Clear Prompt

Create a bulleted outline for a 6-slide, 10-minute presentation introducing our new project to marketing executives who aren't familiar with it. [Brief 1–2 sentence description of the project or links to resources.] Include the project goal, three key benefits, and next steps. Keep it high-level, with no technical details.

The clearer prompt gives the model the goal, relevant context, desired output, and constraints upfront. That can reduce unnecessary output and reiterations while helping you get a high-quality answer in fewer rounds.

Your Words Actually Matter

Even when the task stays the same, the instructions you use can affect how much energy AI uses. For example, words such as “justify,” “explain,” “analyze,” and “comprehensive” can trigger more complex reasoning or longer, more detailed responses. More bounded instructions such as “list” and “summarize” can require less energy because they can limit the amount of new text the model generates.

Current Instruction	More Sustainable Alternative
“Justify why this lead is a good fit.”	“List 3 reasons why this lead fits.”
“Explain the resolution to the customer.”	“Summarize the resolution steps in 5 sentences.”
“Write a comprehensive email… ”	“Write a concise email (under 100 words)... ”

Small changes in wording can make a difference. Swaps such as “list 3 reasons” or “summarize in 5 sentences” can save 10–30% in energy without sacrificing quality.

Build Sustainable Habits

Small, mindful practices can help reduce unnecessary token and resource use in your everyday interactions with AI.

Habit	What to Do
Use the stop button.	Stop a response if it runs long or strays off topic. This prevents unnecessary computing instead of letting the response run to completion.
Save the pleasantries.	Skip unnecessary pleasantries, such as a final “thank you.” They may feel natural, but they’re still prompts the model has to process and respond to. Don’t worry—you won’t hurt its feelings!
Batch your queries.	Combine related questions into one well-structured request instead of sending multiple separate prompts.
Minimize media generation.	Choose text over images, video, or audio when text can accomplish the task. Media generation requires far more energy than text generation.
Create reusable tasks.	Turn repeated tasks into reusable prompts, skills, agents, or workflows instead of rebuilding the same instructions each time. You can ask the model: “I want to turn this task into a reusable skill. What information do you need from me?”
Avoid excessive iteration.	Notice when you’re repeatedly prompting to get the result you need. If a task takes several rounds, consider how you could make your initial prompt more effective next time.
Keep conversations fresh.	Start a new chat when a thread gets long. Long conversations suffer from “context rot,” where the model has to reprocess conversation history with each turn, which increases tokens and degrades focus. Bring forward only the context or summary you need.
Focus each session on one task.	Keep each chat focused on a single task. Mixed requests makes the model process irrelevant context with each turn, while a focused session can reduce unnecessary tokens and keep responses on track.

Each choice may seem small on its own, but repeated choices become habits. When you make these practices part of how you use AI every day, you can reduce your AI footprint and use AI more intentionally and effectively.

Summary

In this unit, you explored how to choose the right tool and model, prompt intentionally, and build sustainable habits. These practices can reduce the resources AI requires and help you get better results.

In the next unit, you put these practices into action and identify changes you can make to your own AI use.

예상 시간

주제

도움말 검색