StartGitHub Broke Again. Cursor Launched Its Rival.
By Addy · August 17, 2026 · Editorial standards
GitHub went down for most of the world's developers today, and by the time it was substantially restored, it had quietly become the platform's 13th separate incident in the first seventeen days of August. That number matters more than today's outage on its own does, and it's landing in the same narrow window that Cursor, backed by SpaceX money and built by the team behind Graphite, chose to accelerate the launch of its own alternative to GitHub entirely.
The Timeline of Today's Outage
GitHub opened the incident at 13:40 UTC, reporting impacted performance across some services. Within five minutes it had a number attached: a roughly 20% error rate across Pull Requests, Issues, and general web traffic. By 14:04 UTC the scope had widened further, the same 20% figure held for web and API traffic broadly, while archive downloads and raw repository content downloads were failing at closer to 50%, with no root cause identified yet. Twenty minutes later, authentication joined the list, SAML and OIDC login, SCIM provisioning, and Team Sync all confirmed impacted. Between 14:45 and 15:01 UTC, Pull Requests, Issues, Actions, Webhooks, and API Requests were each escalated from the milder "degraded performance" to the more serious "degraded availability" classification.
GitHub reported a likely fix around 16:59 UTC, corrective action taken, strong signs of recovery, though error rates remained slightly elevated. Seven of the eight affected services were declared mitigated by 17:30 UTC. GitHub Copilot was not one of them, it stayed degraded longer than everything else GitHub runs. The relief didn't fully hold either, Git Operations relapsed into degraded performance at 17:34 UTC, and as of the most recent update at 18:23 UTC, GitHub was still working through residual, sporadic authentication failures even after declaring the core outage mitigated.
What It Actually Broke
A GitHub outage doesn't stay inside GitHub's own dashboard. Archive and raw content downloads failing at a 50% rate breaks anything that pulls files programmatically, package managers, deployment scripts, container builds that fetch dependencies mid-pipeline. GitHub Actions being degraded means CI/CD itself stalled for however many organizations run builds, tests, and deployments through it, not a webpage being slow but releases actually not shipping. The authentication failures hit hardest for enterprise customers specifically, SAML and OIDC are how many companies gate access to their own repositories, so a login failure there can lock legitimate engineers out of their own code entirely, separate from whether GitHub's servers are technically "up."
Downdetector logged more than 3,000 reports of access problems within hours, with regional peaks of 2,100 in the US and 849 in India. Developers across both regions and beyond reported the same basic experience: pull requests wouldn't load, repositories timed out, and Copilot stopped responding reliably well after the rest of the platform showed signs of recovery.
This Is the 13th Incident in 17 Days
The single most important number in this story isn't from today at all. GitHub has logged 13 separate incidents across nine different days in the first seventeen days of August 2026. July wasn't quiet either, GitHub's own availability report for that month disclosed eight incidents resulting in degraded performance. The pattern underneath both months, according to one detailed incident analysis, keeps tracing back to the same two causes: capacity headroom and change management, meaning routine deployments and configuration changes repeatedly push services past limits that weren't being watched closely enough to catch before customers felt it. GitHub's own postmortem for an August 6 Actions failure conceded its own availability metrics hadn't fully captured the actual impact at the time.
The 90-day uptime numbers make the unevenness concrete. GitHub Actions sits at 99.33% over that window, which works out to roughly fourteen and a half hours of downtime in three months. GitHub Pages is at 99.65%, about seven and a half hours. Git Operations and Webhooks, by contrast, hold 99.99%. Same company, wildly different reliability depending on which specific service an organization actually depends on, and Actions and Copilot, the two most central to how AI-assisted development actually works day to day, are the weakest links.
Cursor Was Already Moving In With Its Own Alternative
Cursor announced Origin on June 16 at its first Compile conference, a code-storage and git-hosting platform built by the team behind Graphite, the stacked-diff review company Cursor acquired in December 2025. The pitch was explicit from the start: GitHub's entire review model assumes human pace, one reviewer, one diff, sequential merges, and that model breaks down once agents are generating code fast enough that review, not generation, becomes the actual bottleneck. Cursor's own demo claimed 22.6 commits per second sustained in a single repository, sub-400-millisecond global sync latency, and a clone-handling capacity in the hundreds of thousands per hour, though the exact figure circulating for that last stat has been flagged as unclear in its original sourcing and is worth treating as a scale claim rather than an audited benchmark until Cursor publishes its own numbers.
The same June event where Origin was unveiled also disclosed SpaceX's $60 billion all-stock acquisition of Cursor's parent company, Anysphere, tying Cursor's infrastructure ambitions directly to the same capital base now running xAI. Origin was originally slated for a fall 2026 launch, waitlist only. That changed fast. By mid-August, internal strings and closed partner betas pointed to a launch landing weeks ahead of schedule, with coverage explicitly naming the reason: GitHub's run of reliability and security incidents has already pushed enterprise customers to look elsewhere, publicly and privately, and Cursor is moving to be the answer before that search settles anywhere else. One analysis of the timing put it plainly: Cursor is exploiting this exact window.
The reception hasn't been uncomplicated. Hacker News called the original announcement a waitlist for remarkably little actual information, and the waitlist itself reportedly buckled under demand at launch, an ironic first impression for a product whose entire premise is infrastructure that can handle load GitHub can't. The harder, unresolved question raised in that same thread is the one that matters more than any commit-speed number: if agents are committing 22.6 times a second, something has to actually review all of it, and neither Origin's demo nor GitHub's own struggles this month have answered what that review process looks like once it's running at machine speed instead of human speed.
None of this means GitHub is being replaced this year. Migrating source control is one of the highest-friction moves a software organization can make, and Origin's own public surface is still closer to a pitch than a finished product. But the timing lines up too cleanly to ignore: the incumbent logged its 13th reliability incident in seventeen days in the same stretch a well-funded challenger moved its launch up by months. GitHub didn't need Cursor to notice the opening. Its own status page has been advertising it daily.
Previously on TheQuery: