GitHub AI agent activity exploded to 275 million weekly commits

GitHub reports explosive growth in AI agent activity with 275 million commits weekly and 14 billion projected for 2026. The surge is causing visible performance issues and outages as infrastructure built for human developers struggles with agent-scale usage. GitHub Actions compute minutes jumped to 2.1 billion per week, requiring urgent capacity planning and potential new pricing models for AI agents.

GitHub paused new Copilot Pro signups and tightened usage limits as agentic workflows consume far more compute than pricing models anticipated. Multi-step agent sessions are 10-20x heavier than autocomplete features, forcing GitHub to push users toward Pro+ plans with 5x token limits. This marks the first public acknowledgment that agentic coding fundamentally changes AI economics.

GitHub introduces AI-powered security detections in Code Security to complement CodeQL, expanding coverage across languages like Shell, Dockerfile, Terraform, and PHP. The hybrid approach processed over 170,000 findings in testing with 80% positive developer feedback. Public preview planned for Q2 2026, integrating directly into pull request workflows.

GitLab released 18.11 with agentic SAST vulnerability resolution now generally available. The feature analyzes security findings, generates code fixes, and opens ready-to-merge requests with confidence scores. New CI Expert Agent helps teams generate first pipelines, while Data Analyst Agent answers natural language questions about development metrics.

GitLab released critical security updates addressing high-severity GraphQL CSRF (CVE-2026-4922) and Web IDE vulnerabilities in versions 18.11.1, 18.10.4, and 18.9.6. The GraphQL flaw could allow session hijacking, while other fixes address access control bypasses and denial of service vulnerabilities. All self-managed installations should upgrade immediately.

Microsoft Digital unveils DevOps Assistant and AI Work Item Assistant, bringing natural language AI directly into Azure DevOps workflows. The chat-based assistant handles project queries and common actions without context switching, while the work item assistant helps create and refine requirements using business context. Both leverage Microsoft Foundry for real-time embedded experiences.

Atlassian releases Rovo Dev, intelligent AI for software teams that accelerates development from planning to deployment. The tool enhances code planning, generation, and pull request reviews by leveraging organizational knowledge from Jira, Confluence, and codebases. Works in terminal for code generation and in Bitbucket/GitHub for PR reviews and requirement matching.

Bitbucket Cloud now supports the Atlassian Rovo MCP Server, enabling AI clients like Claude, ChatGPT, Cursor, and VS Code to browse repositories, create commits, open pull requests, and check pipeline results. This extends the secure MCP connection already available for Jira and Confluence. Currently limited to API token authentication with OAuth support planned.