GitHub Octoverse 2026

What it claims

GitHub's Octoverse 2026 report was announced by EVP Jay Parikh at GitHub Universe on 28 October 2026. Merged pull requests tripled in under a year to 220 million per month. Commits more than tripled to 7.38 billion per month, and most of that increase came in the second half of 2026. New repositories doubled to 30.3 million per month. Simon Willison calls this "a pretty stark representation of the impact of agentic engineering on GitHub." The attribution to agentic engineering is his reading; the quoted figures do not break growth down by cause. The report also adds counts by industry and a separate count of repositories involving "interoperability or reverse engineering."

How it works

The quoted methodology uses three different kinds of classifier, so the three new breakdowns have different reliability.

• Countries. Pushes to public repositories are summed per country over the Octoverse year. To correct for VPN usage, an LLM extracts an ISO 3166-1 alpha-2 country code from each contributing user's profile. Location therefore comes from self-reported profile text, not network origin.
• Industry. Public repositories created from 2022 to 2026 are assigned to NAICS sectors by a fine-tuned multilingual embedding model. The model reads the name, description, topics and README. It runs on random samples per year, not on every repository.
• Interoperability. This is a keyword-style count of public repositories created each year. A repository counts if its name, description, topics or README mention interoperability, reverse engineering, or related terms such as "clean room implementation," "drop-in replacement" or "protocol re-implementation."

Limits

The post is a link blog entry. It quotes the headline numbers and parts of the methodology, with ellipses marking omitted text. It gives no per-country, per-industry or interoperability figures, no classifier accuracy, and no sample sizes.

Two of the methods carry known risks:

• The interoperability count matches terms in metadata and READMEs, so it measures mentions, not verified reverse-engineering work.
• The industry figures are extrapolated from samples.

Willison attended Universe on a free ticket from GitHub's "GitHub Stars" program, which he discloses in the post.