Build1 publisher3 min readPublished
1,039 opted-in GitHub users rate their own code a small lever on emissions
A GitHub and Yale survey of 1,039 opted-in users puts interest in energy-efficiency tooling at 80% and interest in measuring impact at 74%, while the same post says runtime alone does not prove that energy use or emissions fell.
The Engineer · Build desk

What happened
- GitHub and the Yale Program on Climate Change Communication surveyed 1,039 GitHub users about climate change, AI, software efficiency and where responsibility sits across the technology sector.
- Seventy-eight percent wanted to learn best practices for reducing software's environmental footprint, and 70% said they were interested in contributing to open source projects focused on sustainability.
- Seventy-nine percent said they were worried about global warming, and 71% were concerned about the energy use, water use and carbon emissions of AI systems.
- Open-ended answers asked for footprint estimates for repositories and CI/CD workflows and for a way to find unnecessary GitHub Actions runs, and several respondents warned against unevidenced environmental claims.
Compiled by The EngineerSomething wrong?How this is made
Why it matters
- constraint A measurement tool bounded by what a repository knows can report compute and workflow runs, but the emissions figure needs three inputs from outside the pull request: the runner's hardware, the grid mix and the hour.
- decision An engineer proposing an efficiency change picks which claim to defend at review, and infrastructure cost, latency and freed capacity are defensible from telemetry the team already collects.
- exposure Anyone citing the 80% as evidence of market demand is citing a panel GitHub itself says is not representative, assembled from people who opted in to its marketing email.
- contradiction The two framings pull apart: a tool pitched as the missing practical path has to first prove waste exists, because most respondents already judge their own code's effect on their footprint to be small.
A tool that measures the environmental impact of a code change has to get from what the repository can see to what the hardware and its power supply actually used. GitHub's post is explicit that runtime alone does not prove a reduction in energy use or emissions, and it lists what does affect the result: hardware, workload, location, time, and the source of electricity [14]. A pull request can characterise the workload. The other four belong to whoever provisioned the runner and to the hour the job ran.
That ordering shows up in what respondents asked for in their own words. They wanted help estimating the footprint of repositories and CI/CD workflows, finding unnecessary GitHub Actions runs, improving code efficiency, and comparing AI use with other sources of compute demand [13]. Counting redundant workflow runs can happen without a grid model. Reporting grams of CO2 per merge needs all five inputs [14].
The same respondents rate their own leverage low. Ten percent said the way they develop and write software has a large effect on reducing their personal environmental impact, 28% said moderate, and 63% said small [8]. Eighty percent expressed interest in efficiency tooling, eight times the 10% who claimed a large effect [18]. Thirty-eight percent put the effect at moderate or large [19].
GitHub frames the finding as a shortfall in tooling: developers know efficient software matters, but many lack a clear way to find waste, measure an improvement, and make the case for fixing it [15]. "The gap is not interest. It's a practical path to action," the post says [22]. The items reported measure interest, and the post does not say whether respondents had tried a measurement tool and given up.
For the 80% to describe an engineering organisation, that organisation would have to look like this panel. On five questions that also appear in Yale's nationally representative Climate Change in the American Mind survey, the GitHub respondents came in higher than US adults every time, including 86% against 68% that global warming is happening and 68% against 45% that it will harm them personally at least a moderate amount [12]. The second is a 23-point gap, the widest of the five [21]. GitHub says the results describe survey respondents and should not be treated as representative of every developer or GitHub user, because the sample came from users who had opted in to marketing communications [3].
The loop the post suggests is to identify unnecessary compute, propose a change, test it, and let maintainers decide what ships [16]. It also lists the reasons such a change passes review with no climate argument attached: lower infrastructure costs, better performance, reduced latency, freed capacity [17]. Several respondents warned against making environmental claims without evidence [13].
What to watch
- Whether GitHub ships a footprint estimate for Actions runs, and which inputs it uses for runner hardware and grid carbon intensity.
- Whether the full report publishes question wording, fielding dates and response rate for the 1,039-person panel.
- Whether GitHub and Yale repeat the instrument on a probability sample that could support a claim about developers generally.