GitHub once felt like home for millions of developers. Now many call it a liability. Chronic downtime. Floods of low-quality AI output. Pricing shifts that punish heavy users. The platform Microsoft bought for $7.5 billion in 2018 shows clear signs of strain.
David Bushell laid it out plainly in late April. GitHub, he wrote, has become a “lame slop graveyard.” Uptime charts trend downward. Bots and generated code drown real contributions. The service that made open source collaboration simple now frustrates the very people who built it. (dbushell.com)
Numbers back the frustration. Third-party trackers measured GitHub at 90% reliability one month. The next month it fell to 86%. That’s zero nines. Not the standard expected from infrastructure that millions rely on daily. (The Pragmatic Engineer)
GitHub itself acknowledged the problems. In a company blog post, it pointed to a 3.5 times surge in service load. Rapid growth exposed tight coupling between services. A database overload on February 9 took down authentication for many. Failures cascaded. One change in configuration created resource contention that spread platform-wide. The firm admitted it fell short of its own reliability targets. Developer productivity and trust took hits. (InfoQ)
But load alone doesn’t tell the full story. AI features bear much of the blame. GitHub Copilot started as a helpful autocomplete tool. It evolved into agents and subagents that run long, parallel sessions. These workflows consume far more compute than flat-rate subscriptions anticipated.
On April 20 the company moved to protect existing customers. It paused new sign-ups for Copilot Pro, Pro+ and Student plans. Usage limits tightened. Advanced Opus models disappeared from basic Pro tiers. “We’ve seen usage intensify for all users as they realize the value of agents and subagents,” the announcement stated. “These long-running, parallelized workflows… have also challenged our infrastructure and pricing structure.” A VP of Product added that the changes ensure “a reliable and predictable experience for existing customers.” (GitHub Blog)
The AI Slop Problem
Agentic AI doesn’t just strain servers. It changes what lands in repositories. Maintainers now spend hours sifting through generated pull requests that ignore project standards or get abandoned. One GitHub community discussion explored solutions to the rising volume of low-quality, often AI-created contributions. The pattern repeats across popular projects.
Quality of Copilot suggestions has drawn complaints too. Developers report declining accuracy, higher latency and weaker context awareness since late 2025. A March 2026 incident saw the tool inject promotional “tips” into more than 1.5 million pull requests. Trust eroded further. (Nxcode)
Prominent voices have had enough. Mitchell Hashimoto, GitHub user number 1299 who joined in 2008, announced in April that his terminal emulator Ghostty would leave the platform. “This is no longer a place for serious work if it just blocks you out for hours per day, every day,” he wrote. “I want to ship software and it doesn’t want me to ship software.” The post carried real emotion. Hashimoto said writing it made him irrationally sad. (mitchellh.com)
His move fits a pattern. Other maintainers have published similar farewells. The list includes projects shifting to Codeberg, Forgejo or self-hosted options. Each departure chips at network effects that once made leaving unthinkable.
Yet GitHub retains enormous gravity. Billions of lines of code. Millions of users. Enterprise contracts. CI pipelines built on Actions that some engineers now describe as over-engineered. Migration takes time and effort. Many teams stay because the alternatives still feel incomplete.
But cracks widen. Recent Copilot outages hit chat features, dashboards and model performance. One January incident produced error rates that peaked at 100% for a period. Another in March affected agent sessions. Community forums fill with reports of crashes in Visual Studio, entitlement bugs and model degradation. The pattern suggests deeper architectural limits.
So developers experiment. Some explore Gitea or GitLab. Others test Sourcehut or even raw Git over SSH for smaller projects. Non-profit Codeberg gains traction among those seeking independence from corporate priorities. Federation ideas surface for Forgejo and newer platforms, though practical adoption remains early.
The shift carries risk. Git itself stays decentralized and open. The hosting layer does not define the protocol. Yet convenience built habits over 18 years. Breaking them requires pain that outweighs familiarity.
GitHub faces a hard choice. It must stabilize core services while the AI wave drives unpredictable demand. Pricing experiments continue. Usage-based billing arrives in June for some plans. Limits and model restrictions aim to buy time. Whether these steps restore confidence will decide if the platform keeps its central role or becomes one option among many.
Bushell’s original warning still echoes. Git is not GitHub. The service that popularized social coding now tests whether its users will follow it through instability or seek firmer ground. For an industry that ships software at increasing speed, hours of downtime each week prove expensive. And the bills, both literal and figurative, keep arriving.
WebProNews is an iEntry Publication