In short
Use the assistant where it clearly pays back, give it focused context rather than whole repositories, pick a smaller or faster model for routine edits, and break hard tasks into simple steps you review. One study found AI-assisted coding had around 33 times the carbon footprint of the manual approach on the tasks tested, so habits matter.
Steps
- Decide whether the task needs the assistant or is quicker to type yourself.
- Choose a smaller or faster model for routine completions and edits.
- Point the assistant at the relevant files rather than the whole repository.
- Break a hard task into small steps and review each before moving on.
- Set step or spend limits on agent modes and stop runs that loop.
- Track your team's assistant usage alongside cost.
Use the assistant deliberately, not by reflex, and keep what it reads small. Coding assistants are useful, but their energy is invisible to the person using them.
Why does it matter?
Cheung and colleagues compared manual and AI-assisted development on a set of programming tasks and found the AI-assisted approach had around 33 times the carbon footprint of the manual one, with the gap widening as tasks got harder. That does not mean never use assistants; it means use them where they save real time.
Agent modes cost more again. Anthropic reported that its agents typically used about four times the tokens of a chat, and multi-agent systems about fifteen times.
What makes the biggest difference?
- Context. Every file the assistant reads is processed again on each turn. Open or reference only the files that matter, and start a new session for a new task.
- Model choice. Many tools let you pick a model. A small, fast one is usually enough for completions, renames and boilerplate; keep the large one for design and debugging.
- Agent limits. Give agents a clear goal, a stop condition and a step limit. See how to stop AI agents wasting tokens.
- Small steps with review. A person checking each step avoids long runs heading the wrong way.
Common mistakes
- Letting an agent "fix the tests" in an open-ended loop.
- Regenerating a whole file to change one line.
- Accepting code without checking its efficiency: see how to get AI to write energy-efficient code.
For the bigger picture, read agentic AI and token waste.
Sources
- Comparative analysis of carbon footprint in manual vs. LLM-assisted code development, Cheung, Kaul, Jahangirova, Mousavi and Zie, ResponsibleSE 2025 (ACM) (2025)
- How we built our multi-agent research system, Anthropic (2025)
- GreenCode Constitution: agent principles for optimizing code and LLM inference energy use, Kapenekakis, Kim, Ionescu, Thomsen and Albano, GreenAI workshop, ECML PKDD 2026 (2026)
Related
How do I get AI to write energy-efficient code?
How to prompt AI coding tools for efficient code, why efficient-but-wrong is worthless, and how to test and measure generated code before you accept it.
How do I stop AI agents wasting tokens?
Practical guardrails for AI agents: stop conditions, step and token limits, loop detection, context trimming and small models for simple steps.
Agentic AI and token waste
Why AI agents can use many times the tokens and energy of a chat, how loops waste them, and the guardrails that keep agents right-sized.
Use AI efficiently: everyday habits
Everyday habits that cut the energy of using AI: clear prompts, fewer regeneration loops, care with images and video, caching and turning off defaults.