A global survey of computer science educators has found that many instructors have already changed both their teaching and assessment methods in response to generative AI coding tools. The survey, conducted by an ACM task force, included more than 700 educators from 49 countries and found that 69 percent believe AI has changed the skills required for software development.

The findings show a broad shift away from traditional programming assignments focused on writing code from scratch. Instead, educators are placing greater emphasis on code comprehension, debugging, problem-solving, oral examinations, code defense sessions, and project-based work. At the same time, nearly half of respondents said they lack proven examples for incorporating AI into their courses.

AI coding assistants such as ChatGPT and GitHub Copilot can complete many typical programming assignments used in introductory courses, often performing at the level of an average student. That creates a practical challenge for courses that have traditionally used those assignments to test whether students can independently produce and explain working code. The new study offers a broader view of how computer science educators worldwide are responding to that challenge. It was conducted by the "ACM Task Force on Generative AI and Programming Assessment," an expert group established by the education committee of the international computing society ACM.

The task force surveyed 763 educators from 49 countries between May and October 2025. About 500 nearly complete questionnaires were included in the analysis. According to the results, 69 percent of respondents said AI has changed the skills needed for software development. The leading concern among educators was increasing student dependency on technology, cited by 87 percent, followed by cheating and plagiarism, cited by 72 percent.

Teaching moves from writing code toward understanding it

Sixty-four percent of respondents said they have already changed the way they teach. A thematic analysis of open-ended responses showed a consistent direction: less focus on writing code from scratch and more focus on understanding code, finding and fixing errors, and solving problems. Thirty-nine responses also identified AI use and prompt engineering as explicit teaching topics. Other instructors reported demonstrating AI tools in class to illustrate both what they can do and where their limitations remain.

"We know that students will likely be developing with AI tools when they enter the workforce," said Steven Gordon, a professor at Ohio State University and lead author of the report. "The key is to produce graduates who are competent in programming and who understand both the capabilities and limitations of AI."

Oral exams and project-based assessments gain ground

Assessment practices are changing even more quickly. Sixty-eight percent of respondents said they have modified how they test students. The most frequently cited changes included more proctored in-person exams, with 56 mentions; reduced weighting for homework, with 38 mentions; oral exams and code defense sessions, with 36 mentions; and paper-based tests, with 35 mentions. Project-based assessments were also cited 34 times.

Some educators now require students to disclose whether and how they used AI tools, and in some cases to submit logs of their interactions with those tools. These changes reflect a growing effort to distinguish between a student's own understanding and work produced or heavily assisted by software.

Institutional policies remain uneven. Forty-five percent of respondents said their institution has guidelines on AI use, while 39 percent said their institution does not. Existing policies range from complete bans to explicit permission or encouragement, provided that students document their use of AI. For instructors, that uneven policy landscape means course-level assessment changes are often being made while institutions are still deciding how AI use should be governed.

Educators cite a lack of best-practice examples

The most common barrier to AI integration, cited by 48 percent of respondents, was the absence of proven best practices. Twenty-eight percent said they do not have sufficient expertise with AI, while 20 percent said they see no need to integrate AI into their teaching.

Respondents also identified a need for more professional development. Seventy-four percent said they want training on effective teaching methods, and 66 percent said they want support in redesigning assessments. The task force's resource site addresses that gap by collecting practical examples, but the survey results suggest many educators are still looking for tested models they can adapt to their own courses.

The study's authors noted several limitations. Participants were recruited mainly through ACM and SIGCSE mailing lists and selected national educator directories. The respondent pool was weighted heavily toward North America, with 227 respondents, and Europe, with 106. Asia accounted for 57 respondents, South America for 10, Oceania for 9, and Africa for 3. Among respondents who reported their institution type, 77 percent worked at universities, while K-12 schools and vocational programs were only lightly represented. The findings are also based on self-reported data.

The ACM study is freely available online. The task force also created a website with practical examples and resources for educators.

Recent studies point to higher grades but weaker learning outcomes

Several recent studies align with the concerns raised by respondents. An Anthropic study found that software developers who used AI assistance to learn a new Python library scored 17 percent lower on a follow-up knowledge test than a control group that did not have AI access. A small subset of participants who fully delegated coding to AI completed tasks the fastest but averaged only 39 percent on the knowledge test.

Similar findings have appeared outside computer science. A Chinese longitudinal study of more than 26,000 students found that AI use increased homework grades by 18 percent and reduced completion time by 30 percent. However, after six months, scores on closed-book exams fell by 20 percent. On later entrance exams, scores declined by 18 percent and 24 percent, depending on the test.

A UC Berkeley analysis covering more than 500,000 grades found a sharp increase in top marks after the release of ChatGPT, particularly in courses with substantial writing and programming assignments. The increase was most pronounced in courses where unproctored homework carried significant weight in the final grade. According to the ACM report, that is the same category of assessment that many educators are now reducing in importance.