Octoverse: Commits, pushes, reach new highs in 2026 as developers build with agents

A commit is a recorded change to files in a repository. It preserves what changed and helps track a project’s history. A code push sends one or more commits from a developer’s computer to GitHub. A merged pull request is a proposed set of changes that has been accepted and combined with the project’s code.

These actions happen at different points. A developer can create several commits locally, then push them to GitHub. Those commits can support a pull request, where others review the proposed work. If accepted, the pull request is merged. GitHub counted these activities separately: 17.9 billion code pushes, 34.3 billion commits, and 1.3 billion merged pull requests.

The distinction matters because the numbers measure different kinds of activity. Commits count recorded changes, pushes count transfers to GitHub, and merged pull requests count accepted collaborative changes. They cannot be treated as interchangeable measures of developer output.

GitHub experienced unusually rapid growth in 2026. Developers pushed code 17.9 billion times during the past 12 months. Commits across public and private repositories reached 34.3 billion, more than three times the level two years earlier. Merged pull requests reached 1.3 billion, averaging 3.5 million each day.

The growth accelerated around June. From then onward, about 55% of all commits came from brand-new repositories. Developers who pushed code had been on GitHub for an average of 36 months, while the average developer worked across 10% more repositories than a year earlier.

The article does not provide a total number of GitHub users or developers. It does report 2.5 billion public contributions in 2026, more than double 2025’s 1.12 billion. It also says every projected country closed at least a fifth of its distance toward its 2030 growth projection.

An AI coding agent is software that helps with development work. The article describes agents moving beyond single-session assistance toward persistent systems. These agents can track progress across days, orchestrate specialized agents, and use cloud resources as well as local machines. They became part of developers’ daily work during the year.

An agent configuration file is a repository file that sets up or describes agent use for that project. The article does not specify its format or exact instructions. It does show why the file matters as a signal: among repositories created during the year, those with one recorded eight times as many commits as new repositories without one.

This makes the file a useful marker for agent-enabled development from the start. It does not prove that the file itself caused more commits. The wider pattern connects agent adoption with a markedly different development pace, as developers used more models, tools, automations, and cloud execution.

Repositories built with agents from day one showed a striking activity gap. The average new repository with an agent configuration file recorded eight times as many commits as the average new repository without one. A commit is only a recorded change, so the comparison signals a much faster pace of visible development.

The key mechanism is early integration. Agents were no longer occasional, single-session tools. They could track work over several days, use cloud computing, and coordinate specialized agents. A repository configured for that workflow may support repeated, automated, or agent-assisted changes while developers continue directing the project.

The article does not establish a single cause. The configuration file may identify teams already committed to intensive agent use, rather than create the activity by itself. Still, the result points to a broader shift: agents were becoming part of everyday development, and repositories designed around them moved at a markedly different pace.

Agents became more useful because several parts of the technology improved together. The article points to better underlying AI models, new harnesses and tools around those models, and agents running on top of them. Developers also had more models to choose from, expanding the ways they could build.

The biggest workflow change was persistence. Agents could track progress across days instead of stopping after one session. They could also orchestrate other specialized agents. At the same time, agents made extensive use of the cloud, executing work remotely as well as on local machines. These capabilities supported longer, more connected tasks.

The shift helped explain GitHub’s sharp activity surge around June. Agents moved from something developers were trying out to something they worked alongside. The article links this period with rising activity, but it does not predict a specific endpoint. It does say forecasting models no longer agree on where developer growth will go next.

GitHub Actions automates jobs in a software project. Those jobs can keep development work moving without requiring a person to start every task manually. The article reports 1,440 Actions jobs every second, equal to 124 million per day and more than three times the previous year’s total.

Dependabot automates dependency updates through pull requests. Dependencies are external software components that projects rely on. Teams can review and merge these proposed updates, helping keep dependencies current and reducing use of known vulnerabilities. Developers merged more than 30 million Dependabot pull requests during the year, about twice 2025’s pace.

Together, these systems show automation operating around the code itself. Actions handles recurring project jobs, while Dependabot focuses on dependency maintenance. Their growth means teams can spend less effort on repetitive upkeep. It also shows that software security and maintenance became part of the everyday development workflow at large scale.

Version control gives a software project a shared history. Developers can record changes, send them to a repository, and review proposed work before it becomes part of the project. This creates a common place to build, check, and improve code. Pull requests provide a pathway for discussing and accepting changes.

Open-source collaboration extends that process across communities. Developers can contribute to public repositories and maintain shared tools. The article highlights mathlib, a community library of computer-checked mathematics, and evcc, an open-source electric-vehicle charging controller and home energy management system. Their maintainers continued contributing at high volume.

This model supported 2.5 billion public contributions in 2026, more than double the 1.12 billion recorded the previous year. Developers also reused GitHub Pages and Actions to build Gati, a Nepali-first relay of official disaster alerts. The scale suggests collaboration will keep connecting individual work with community needs.