Stay connected with KayaToday—follow us on Instagram and Facebook for the latest news and reviews delivered straight to you.
There is a particular kind of admission that companies make only when the alternative is worse: conceding, implicitly, that a problem has outgrown their own infrastructure and asking a direct competitor for help. Microsoft made that admission this week. On June 16, the company confirmed it is routing a portion of GitHub’s traffic through Amazon Web Services — the cloud platform of the company against which Microsoft Azure competes most directly — after AI coding agents overwhelmed GitHub’s systems to the point of repeated service failure.
The scale of the strain, once disclosed, explains the decision. GitHub’s chief operating officer, Kyle Daigle, confirmed in April that the platform was processing 275 million commits per week, putting it on pace for 14 billion commits across 2026 — compared with one billion in the whole of 2025. Pull requests opened by AI agents rather than human developers surged from 4 million in September 2025 to more than 17 million by March 2026. The platform logged nine service-degrading incidents in May alone, with availability falling to approximately 88.4 per cent in June — well below the 99.9 per cent threshold that GitHub’s enterprise customers contractually expect.
A System Built for Humans, Overrun by Machines
GitHub was designed, architecturally and culturally, around the rhythms of human software development: commits made at the pace a person can type and think, pull requests reviewed at the pace a person can read and reason, infrastructure scaled to the predictable cadence of engineering teams working in roughly synchronised time zones. That architecture has functioned remarkably well for nearly two decades. It was not designed for AI coding agents that can generate, test, and submit code changes continuously, in parallel, across thousands of simultaneous sessions, with no need for sleep, lunch, or the cognitive limits that have implicitly throttled the volume of software change since the platform’s founding.
The arrival of agentic coding tools — Anthropic’s Claude Code, OpenAI’s Codex, GitHub’s own Copilot, and a growing field of competitors — has not simply added more users to GitHub’s existing demand curve. It has introduced an entirely different category of user, one whose throughput characteristics bear no resemblance to the human developers the platform’s capacity planning was built around. Microsoft’s description of the AWS arrangement as a “temporary measure” while GitHub migrates further onto Azure by 2027 is candid about the nature of the problem: this is not a momentary spike to be absorbed, but a structural shift in the platform’s load profile that requires years of infrastructure redesign to properly address.
What This Reveals About the Agentic Transition
The GitHub strain is a useful, concrete illustration of a more abstract problem that the AI industry has discussed mostly in the language of compute and chips: the infrastructure built for the human-paced internet was not designed for an AI-paced one, and the transition between the two is proving more disruptive, in practical operational terms, than most forecasts anticipated.
Read also this: https://kayatoday.com/news/anthropic-told-the-world-its-ai-was-dangerous-the-government-believed-it/
This is not confined to developer tools. Customer support platforms, financial transaction systems, content moderation pipelines, and identity verification services are all systems originally designed around assumptions about human request volume and human-comprehensible request patterns. As AI agents increasingly initiate the requests those systems process — searching, querying, executing multi-step transactions — the systems built for an earlier era are being asked to absorb demand profiles for which they were never engineered. GitHub’s public struggle this month is simply the most visible and best-documented instance of a pattern that is almost certainly occurring, with less public scrutiny, across many other categories of digital infrastructure.
The Competitive Irony
There is a layer of corporate irony in Microsoft’s solution that should not go unremarked. Microsoft owns GitHub, has built its own competing agentic coding tool in Copilot, and operates Azure as its primary cloud platform — yet found itself needing capacity from Amazon Web Services, the cloud business of the company against which it competes most fiercely for enterprise infrastructure spending, to keep GitHub functioning. That Microsoft made this choice rather than allow GitHub’s availability to continue degrading is a signal about how seriously the company is treating the reputational and contractual risk of platform failure. It is also a reminder that even the largest technology companies, with access to enormous internal cloud capacity, can find themselves capacity-constrained when demand growth compounds as quickly as agentic AI usage has compounded over the past nine months.
For the broader industry, the lesson is one that deserves more attention than it has received amid the more headline-grabbing stories of AI valuations and model releases. The agentic transition is not simply a question of which company builds the most capable model. It is a question of whether the digital infrastructure underlying nearly every category of software service can be re-engineered fast enough to absorb the volume and velocity that autonomous AI systems are now capable of generating. GitHub’s June was a preview. It will not be the last one.
Read Also: Jensen Huang Did Not Come to Taipei to Sell Chips. He Came to Declare an Empire.
