Every year GitHub publishes Octoverse, a report on what happened across the platform. The 2026 edition, published on 28 October and covering the 12 months to 30 September, has one headline: AI agents have become ordinary teammates. They now open a large share of the code changes on GitHub, and code review has become the place where people and agents meet. If your analytics team keeps dbt models, SQL or Python in Git, this is your future too.
The numbers
Source: GitHub Octoverse 2026. GitHub reports a sharp acceleration around June, linked to stronger models, better orchestration tools and agents that run in the cloud for days.
| Activity, 12 months to September 2026 | Total |
|---|---|
| Pushes | 17.9 billion, more than 2024 and 2025 combined |
| Commits | 34.3 billion |
| Merged pull requests | 1.3 billion |
| GitHub Actions jobs | 45.4 billion, about 1,440 a second |
| Merged Dependabot pull requests | More than 30 million, about double 2025 |
| Public contributions | 2.5 billion, more than double last year |
One finding is easy to misread. New repositories that include an agent configuration file, such as an AGENTS.md that tells coding agents how the project works, averaged eight times as many commits as those without one. GitHub itself calls this a correlation, not proof: busy projects may simply be more likely to set agents up.
What it means for data teams
Analytics engineering already runs on pull requests: a dbt model change, a new test, a fix to a Power Query script. Agents such as Claude Code, Codex, Copilot, Genie Code on Databricks and CoCo on Snowflake can now open those pull requests for you. Snowflake even announced a CoCo GitHub integration at Expedition 2026 so that agent changes go through normal review.
- 1
Write an agent guide for your repo
Add a short AGENTS.md: how to run tests, naming rules, which folders are off limits, and how to describe changes.
- 2
Make tests the gate
Agents should not merge anything that has not passed dbt tests or unit tests. Our guide to data quality tests shows where to start.
- 3
Keep pull requests small
Ask agents for one model or one fix per pull request. Small changes are reviewed properly; big ones are rubber-stamped.
- 4
Review data, not just code
Check row counts, nulls and key metrics before and after, not only the SQL diff.
- 5
Protect secrets and production
Agents should work on a branch with development credentials, never on production data directly.
Where agents help most
Low risk- Adding tests and documentation to existing models
- Fixing lint errors and small bugs
- Upgrading packages, which Dependabot already does at scale
Good first jobs for a coding agent.
Where people stay essential
High risk- Changing metric definitions everyone relies on
- Anything touching personal or financial data
- Deleting or rewriting history in production tables
Keep a person in charge of these decisions.
Check your understanding
1/3What share of GitHub pull requests did agents start in September 2026?
It rose from 17% in March to over 40% in September.
2/3What does an agent configuration file such as AGENTS.md do?
Repositories with one averaged more commits, though GitHub says that is a correlation, not proof.
3/3According to Octoverse, where do people and agents now meet?
Code review has become the central point where work is divided between people and agents.
You got 0 out of 3.
Read next: a security checklist for AI agents that touch your data, long-running AI agents, data quality tests and the Git cheat sheet.
Sources: GitHub Octoverse 2026 (28 October 2026), as summarised by ToolNavs and Visual Studio Magazine (29 October 2026); Snowflake Expedition 2026 announcements (3 November 2026). Figures cover the 12 months to 30 September 2026.
Alessandro Ecclesie AgazziWritten by Alessandro Ecclesie Agazzi, analytics engineer in London. Updated 3 November 2026. About me