Cursor, GitLab, Zed split on fixing GitHub's agent crisis

Key Takeaways

• GitHub availability dropped to 88.4% in June 2026 as AI coding agents flood the platform with automated commits
• Cursor's new Origin feature rethinks code synchronization for agent-driven workflows
• Annual commits projected to hit 14 billion in 2026, up from 1 billion in 2025

GitHub's infrastructure is buckling under the weight of AI coding agents, and the industry's major players agree on the diagnosis but not the cure. Platform availability dropped to 88.4% in June 2026. Cursor, GitLab, and Zed are each proposing radically different architectures to handle what GitHub's own COO calls "an era where agents drive the repository lifecycle."

The numbers tell the story. Annual code commits are projected to reach 14 billion in 2026, a fourteen-fold increase from the 1 billion recorded in 2025. Weekly commit volume now runs at 275 million. GitHub Actions compute usage hit 2.1 billion minutes per week, quadrupling since 2023. The platform designed for human developers is drowning in machine-generated traffic.

What broke GitHub's model?

The culprit is the rapid adoption of autonomous AI coding agents like Claude Code and Devin. These tools don't wait for pull request reviews or debate commit messages in Slack. They generate code, push commits, and open pull requests at machine speed. By March 2026, AI agents were initiating 17 million pull requests per month.

GitHub acknowledged the strain in an official statement, but the underlying architecture remains optimized for a world where commits arrive at human pace. Merge conflicts, CI/CD pipelines, and review queues all assume humans are in the loop. When agents remove that assumption, the entire system strains.

How is Cursor responding with Origin?

Cursor, the AI-native code editor that has grown rapidly among developers, announced new "Origin" hosting features designed specifically for agent-heavy workflows. The company argues that the bottleneck isn't code quality or compute. It's the legacy concept of centralized, human-managed repositories.

Cursor's bet is that the future requires treating code synchronization as a real-time state problem rather than a batch commit problem. Origin appears to move away from the pull request model entirely for agent-generated changes, instead managing continuous synchronization between agent workspaces and the canonical codebase.

Where do GitLab and Zed diverge?

GitLab and Zed see the same crisis but propose different fixes. GitLab's approach doubles down on their existing DevSecOps pipeline, adding agent-aware orchestration layers that can throttle, batch, and prioritize automated commits based on business rules. The philosophy: keep the centralized model but make it smarter about what traffic deserves priority.

Zed takes a more radical position. The editor, built from scratch for performance, argues that the git model itself needs rethinking. Their early experiments suggest peer-to-peer synchronization between editor instances, reducing dependence on any central repository for real-time collaboration. Whether that scales to enterprise workflows remains unproven.

Why standard scaling won't fix this

Throwing hardware at the problem has diminishing returns. The issue isn't just throughput. It's architectural assumptions baked into git's design from 2005. Linus Torvalds built git for kernel development, a workflow measured in patches per day, not per second.

Consider merge conflicts. Git assumes human intervention to resolve them. When two agents modify the same file simultaneously, the current model creates a queue of conflicts waiting for human review. That queue now grows faster than any team can process it. Some organizations report conflict backlogs measured in weeks.

CI/CD pipelines face similar strain. GitHub Actions was built to run tests after human commits, with cost structures and rate limits designed around human velocity. Agents that commit every few minutes quickly exhaust quotas or rack up bills that dwarf the value of the changes.

What happens to smaller teams?

Enterprise customers with dedicated support contracts get priority when GitHub experiences degradation. Smaller teams on free or lower tiers absorb the worst of the availability drops. A startup founder told The New Stack that their deployment pipeline failed three times last week because GitHub's API timed out during peak agent activity.

This creates an uncomfortable dynamic. The teams least able to migrate to alternative infrastructure are most affected by the platform's struggles. Meanwhile, well-funded companies can afford to run hybrid setups with self-hosted GitLab or private forks synced to GitHub only for collaboration.

Is Microsoft investing enough in GitHub?

Microsoft acquired GitHub for $7.5 billion in 2018. Since then, the platform has become central to Copilot's training data and Microsoft's AI strategy. But infrastructure investment hasn't kept pace with the demand GitHub itself helped create. Copilot trains developers to expect AI assistance. Those developers then deploy agents that overwhelm the platform hosting their code.

GitHub's response has been measured. The company points to ongoing work on infrastructure improvements but hasn't announced architectural changes on the scale Cursor or Zed propose. Whether that reflects confidence in incremental fixes or organizational inertia is unclear from the outside.

What should engineering teams do now?

For teams already experiencing GitHub degradation, immediate options are limited. Rate-limiting your own agents helps avoid account restrictions. Batching commits from agent workflows reduces API calls. Running a self-hosted mirror can insulate critical CI/CD from upstream availability issues.

Longer term, the question is whether to bet on GitHub fixing itself, migrate to GitLab, experiment with Cursor's Origin, or wait for Zed's approach to mature. None of these options are painless. The only certainty is that the status quo won't hold.

Frequently Asked Questions

Why is GitHub experiencing availability problems?

AI coding agents are generating commits and pull requests at machine speed, overwhelming infrastructure designed for human-paced development. Weekly commits hit 275 million, with GitHub Actions compute usage quadrupling since 2023.

What is Cursor Origin?

Origin is Cursor's new hosting feature that treats code synchronization as a real-time state problem rather than a batch commit problem, designed specifically for agent-heavy workflows.

How many commits do AI agents generate?

AI agents initiated 17 million pull requests per month by March 2026. Annual commits are projected to reach 14 billion in 2026, up from 1 billion in 2025.

Can GitHub scale its way out of this problem?

Experts say throwing hardware at the problem has diminishing returns because the architectural assumptions in git date to 2005 and assume human-paced workflows with human conflict resolution.

Should teams migrate away from GitHub?

Options include GitLab with agent-aware orchestration, Cursor's Origin for real-time sync, or self-hosted mirrors. Each involves trade-offs, and no solution is yet proven at scale for agent-heavy workflows.

Source: The New Stack / Paul Sawers