The consequences of the AI-driven coding growth are exhibiting up in areas the place one wouldn’t instantly count on.
GitHub, the world’s dominant code internet hosting platform, has been hit by a sequence of outages in latest months — and the foundation trigger isn’t a cyberattack or a botched replace. It’s the sheer weight of AI brokers writing, pushing, and merging code at a scale the platform was by no means constructed to deal with.
A Site visitors Explosion In contrast to Something Earlier than
The numbers are staggering. GitHub COO Kyle Daigle put it plainly in a latest submit: “There have been 1 billion commits in 2025. Now, it’s 275 million per week, on tempo for 14 billion this yr if development stays linear (spoiler: it received’t.)” GitHub Actions, the platform’s automation engine, has gone from 500 million minutes per week in 2023 to 1 billion in 2025 — and has since hit 2.1 billion minutes in a single week.
This isn’t extra builders signing up. It’s AI coding brokers — instruments like Claude Code, Cursor, and OpenAI Codex — autonomously cloning repositories, opening pull requests, working CI pipelines, and iterating at machine pace. AI-agent pull requests alone jumped from 4 million in September 2025 to 17 million in March 2026, a greater than 4x enhance in six months.
GitHub, web site, and app knowledge from early 2026 exhibits code pushes within the US rising 35% year-on-year — metrics that have been primarily flat simply two years prior.
The Infrastructure Cracks
GitHub’s VP Engineering Vlad Fedorov acknowledged in an April 28 weblog submit that the corporate started a plan to extend capability by 10X in October 2025 — solely to comprehend by February 2026 that they wanted to design for 30X right this moment’s scale. The corporate attributes this to a “speedy change in how software program is being constructed,” noting that “for the reason that second half of December 2025, agentic growth workflows have accelerated sharply.”
The compounding nature of the issue is critical. As Fedorov wrote, a single pull request can contact “Git storage, mergeability checks, department safety, GitHub Actions, search, notifications, permissions, webhooks, APIs, background jobs, caches, and databases.” At scale, he famous, “small inefficiencies compound: queues deepen, cache misses develop into database load, indexes fall behind, retries amplify site visitors.”
February 2026 alone noticed 37 separate platform incidents. Within the first two days of April, GitHub logged 5 extra as Copilot agent periods hit useful resource exhaustion. Unofficial monitoring put platform uptime under 90% at sure factors — towards an Enterprise SLA that guarantees 99.9%.
The Architectural Mismatch
The deeper situation is structural. AI coding brokers work together with GitHub essentially otherwise from human builders. They don’t browse the UI — they hammer the API and CLI repeatedly, across the clock, throughout 1000’s of repositories concurrently. A human developer on a free GitHub account generates a handful of commits a day. An AI agent on the identical account can produce tons of of commits, dozens of PRs, and 1000’s of Actions minutes in a day.
GitHub’s pricing mannequin hasn’t caught up. The platform’s beneficiant free-tier API limits and Actions minutes have been designed for human builders and occasional CI bots — not fleets of autonomous brokers. Agent-specific charge limits and devoted paid tiers seem inevitable; the query is whether or not they arrive proactively or solely after extra outages drive the difficulty.
GitHub has additionally been concurrently executing a migration to Azure, with a goal of fifty% of all site visitors on Azure by July 2026. Operating a significant infrastructure migration alongside an AI-driven site visitors explosion is, as one evaluation put it, an infrastructure crew “stretched skinny.”
What GitHub Is Doing
Fedorov outlined a multi-pronged response: transferring webhooks out of MySQL, redesigning the consumer session cache, remodeling authentication flows to scale back database load, isolating crucial companies like Git and GitHub Actions from different workloads, and migrating performance-sensitive code from the Ruby monolith into Go. The corporate can also be pursuing a multi-cloud path, describing it as essential to “obtain the extent of resilience, low latency, and adaptability that will probably be wanted sooner or later.”
The platform has additionally up to date its standing web page to incorporate availability numbers and dedicated to logging all incidents — massive and small — to offer customers clearer indicators when one thing is flawed on GitHub’s finish.
The precedence order, Fedorov acknowledged plainly: “availability first, then capability, then new options.”
The Greater Image
GitHub has develop into the central nervous system of AI-driven software program growth. Each main coding agent routes its output via the platform. When it goes down, it doesn’t simply sluggish builders — it stops whole AI-powered pipelines chilly. The irony, as observers have famous, is that Git was designed from the begin to be decentralized; the ecosystem constructed on prime of it has develop into something however.
GitHub’s outages are a number one indicator of a broader infrastructure reckoning. As AI brokers write an increasing number of code throughout the business, the platforms that underpin software program growth — from model management to CI/CD to bundle registries — might want to rebuild themselves for a world the place the “consumer” is commonly not a human in any respect.