Among GitHub Copilot users, some fraction of their code is written by themselves and some generated by the model. In any population of users, what will be the highest fraction generated by the model as reported by a credible source by 2025?

360Ṁ56

resolved Jan 1

100%32%

50%

41%Other

80%

60%

70%

1.2%

90%

As indicated by at least one study published in a peer-reviewed journal or news article published in a source considered "generally reliable" on Wikipedia.

Technical AI Timelines

GitHub

Ancient Markets

Get

1,000

to start trading!

🏅 Top traders

#	Name	Total profit
1		Ṁ103
2		Ṁ11

People are also trading

Over the next five years, what percent of GitHub code will be authored by an AI agent?

A major tech company, besides Anthropic, reports at least 98% of its code is AI-generated before April 1, 2026

16% chance

Will Microsoft Copilot continue to rely primarily on OpenAI foundation models at the end of 2025?

65% chance

In 2028, will programmers usually be able to use GitHub Copilot and similar tools in interviews?

31% chance

Top score on codeforces by an AI model at the end of 2025

Will AI generated content be ~50% or more of content creation on major social media platforms in 2025?

16% chance

By 2030, will over 50% of software development projects be primarily created by AI, with minimal human coding?

73% chance

Will GitHub release an assistant, similar to Copilot, able to insert comments in code by 2028?

77% chance

Will the GitHub Copilot Litigation Succeed?

Sort by:

50% now: "the same amount of characters in the code are now completed with AI-based assistance as are manually typed by developers" From Google https://research.google/blog/ai-in-software-engineering-at-google-progress-and-the-path-ahead/

How are you resolving inexact answers? e.g. if 50% and 60% are responses and the actual highest reported percentage is 52%, how do you resolve it? (same question for 55% and 58%)

@vluzko To whatever is closest to the actual number. If two are closer in both directions, underestimating and overestimating by the same amount, I resolve for both.