GitHub Copilot is grappling with a series of major service disruptions and an exodus of key leaders, according to a CNBC report published on May 22, 2026. The revelations cast a shadow over Microsoft’s flagship AI developer tool, which has become the centerpiece of the company’s strategy to dominate software creation just as a wave of aggressive competitors erodes its once-unassailable market position.

Microsoft acquired GitHub for $7.5 billion in 2018, and for years the bet appeared prescient. Copilot, launched as a technical preview in 2021 and made generally available in 2022, quickly became the most widely used AI coding assistant, deeply integrated into Visual Studio Code, Visual Studio, and JetBrains IDEs. By mid-2025, GitHub boasted over 150 million developers on the platform, and Copilot had been adopted by more than 2 million organizations, including 90% of the Fortune 100. But the CNBC investigation reveals that behind the glowing adoption numbers, the service has been plagued by frequent outages that have left developers stranded during critical work hours and that a steady stream of senior talent has departed, raising questions about the product’s long-term stability.

A perfect storm of reliability failures

Developers and IT managers who spoke to CNBC described a series of cascading Copilot failures that stretched back to late 2025. On multiple occasions, the Copilot service became completely unresponsive for hours at a time, preventing code completions, chat assistance, and agentic features from functioning. In one particularly damaging incident in March 2026—confirmed by internal status logs obtained by the news outlet—a backend storage misconfiguration in GitHub’s West US data center caused Copilot to reject authentication tokens for nearly 11 hours. The outage affected not only individual developers but also enterprise pipelines that had been built to depend on Copilot’s code-review and documentation-generation agents.

“When Copilot went down, our entire CI/CD process froze,” said a senior DevOps engineer at a large financial services firm who requested anonymity. “We had built automated pull request descriptions and unit test generation into the workflow. Without it, our team lost a full day of productivity, and fixing the cascading failures took another week.”

Such incidents have become more frequent. GitHub’s own status page logged twelve major incidents classified as “service degradation” or “service outage” for Copilot between November 2025 and April 2026, compared with only two in the same period a year earlier. The underlying cause, according to two former GitHub engineering managers, is the breakneck pace at which Microsoft has layered new AI capabilities on top of aging infrastructure. Copilot originally ran on OpenAI’s Codex models, but over time Microsoft migrated to a complex mix of fine-tuned GPT-4 variants and custom Phi models. Each new model update introduced subtle compatibility issues that, when combined with a globally distributed serving layer, created unpredictable failure modes.

“The system became a Rube Goldberg machine,” one former manager told CNBC. “Trying to add multi-file editing, agent mode, and real-time context awareness all at once meant that even a minor DNS blip could cascade into a total service meltdown. We were firefighting constantly, and the best engineers burned out.”

Leadership vacuum at the top

Compounding the reliability woes is what insiders describe as a leadership crisis. GitHub CEO Thomas Dohmke, who shepherded the Copilot launch and was widely credited with its initial success, has been increasingly absent from day-to-day operations, CNBC reported, spending much of his time on Microsoft’s broader AI strategy and appearing less frequently at GitHub’s San Francisco headquarters. While Dohmke remains in place, the executive layer below him has been hollowed out. In the past nine months, the vice president of Copilot engineering, the head of AI platform, and three senior product leads have all departed—some for rival AI coding startups.

Perhaps the most damaging exit was that of Shuyin Zhao, the VP of product who oversaw Copilot’s rapid feature expansion and enterprise go-to-market. Zhao left in January 2026 to become CTO of Cursor, a fast-growing AI-native IDE developed by Anysphere. Her departure, combined with the earlier loss of Copilot’s founding architect Jaron Lanier (who left to pursue academic research), left a gaping hole in the product’s vision and execution. Current and former employees described a leadership team in disarray, with decisions bogged down by what one person called “Microsoft’s infinite committee culture.”

“Every feature now requires a series of reviews from Azure AI, Microsoft Research, the Trustworthy AI office, and the Copilot steering committee,” said an engineer who recently left GitHub. “By the time an improvement is approved, the competitor has already shipped it. It’s incredibly frustrating.”

The churn has already altered the product roadmap. Ambitious plans for Copilot Agent—a system that could autonomously write and test entire features—have been scaled back, with the public preview pushed from Q1 2026 to “late 2026.” Meanwhile, previously-shipped features like Copilot Chat’s project-aware context have become less reliable, occasionally hallucinating non-existent APIs or reverting to stale code snapshots.

Rivals seize the moment

While Copilot stumbles, rivals are moving aggressively. Amazon CodeWhisperer, rebranded last year as Amazon Q Developer, has added deep integration with AWS services and a generous free tier that undercuts Copilot’s pricing. Google’s Gemini Code Assist, launched in late 2025, leverages Gemini 2.5 Pro to offer full-codebase awareness—a feature GitHub has only half-delivered—and is bundled at no extra cost with Google Cloud subscriptions. Open-source alternatives like Tabby and Continue.dev have matured, giving enterprises a self-hosted option that avoids the cloud dependency issues that plague Copilot.

But the most direct threat comes from a wave of AI-native coding startups. Cursor, led by former GitHub VP Zhao, has grown from 500,000 users in early 2025 to over 4 million by May 2026, backed by a $400 million funding round that valued the company at $10 billion. Cursor’s tight integration of editing, terminal, and AI via a dedicated IDE—built on top of VS Code’s open-source core—offers a seamless experience that many developers say outstrips Copilot’s plugin-based approach. Replit, with its Agent mode that can build and deploy entire applications from a prompt, has also attracted over 30 million developers, while Cognition AI’s Devin, an autonomous coding agent, has been licensed by several large banks and insurance companies to accelerate legacy migrations.

“I used to recommend Copilot to every new hire,” said a CTO at a mid-sized e-commerce company. “But after the March outage cost us a critical release window, we started a trial with Cursor. Within two weeks, our developers were more productive, and we haven’t looked back. The tool just understands our codebase better.”

Data from the latest Stack Overflow Developer Survey, released days before the CNBC report, shows Copilot’s market share among professional developers dropping from 67% in 2025 to 51% in 2026. While still the leader, the decline is steep, with Cursor leaping to 29% and Amazon Q Developer capturing 14%. Among students and hobbyists, the shift is even more pronounced.

Microsoft’s response: A vow to rebuild trust

Microsoft and GitHub have publicly acknowledged the reliability problems. In a blog post published late on May 22, GitHub’s VP of Engineering, Barry Libenson, wrote that “we have not met our own high bar for service availability and are taking immediate steps to remediate.” The post detailed a new dedicated Copilot reliability team, a week-long development freeze to focus exclusively on stability improvements, and a commitment to publish real-time per-component health metrics. A GitHub spokesperson told CNBC that the company would not comment on personnel changes but noted that “we continue to attract world-class talent and are actively hiring across the Copilot organization.”

But these assurances have done little to reassure jittery enterprise customers. One Fortune 500 CIO told CNBC that his company has invoked a disaster-recovery clause in its GitHub Enterprise contract and is now requiring Copilot outages to be included in service-level agreements with financial penalties. “Microsoft’s AI credibility is on the line,” the CIO said. “If Copilot can’t stay online, why should we trust Azure OpenAI for anything mission-critical?”

Industry analysts see the crisis as a pivotal moment. “GitHub Copilot single-handedly made AI coding mainstream, but Microsoft has become complacent,” said Julia McKenzie, lead analyst at DevTech Research. “The outages expose a deeper truth: AI development tools are infrastructure now, and infrastructure must be boringly reliable. The leadership vacuum suggests Microsoft doesn’t fully appreciate that yet. Meanwhile, the competition has figured out that developer trust is the ultimate moat.”

What’s at stake for Microsoft

Copilot is more than a product; it is the linchpin of Microsoft’s “AI developer platform” vision. The tool feeds telemetry that improves Azure’s AI models, drives adoption of GitHub Actions and Azure DevOps, and locks developers into the Microsoft ecosystem. Every developer who switches to Cursor or Q Developer is a developer who may gradually drift away from Azure services. Worse, the exodus of senior Copilot leaders to startups means that competitive know-how is rapidly diffusing.

Microsoft still holds formidable advantages: deep integration with its identity and licensing stack, massive capital to invest in infrastructure, and relationships with enterprises that no startup can replicate overnight. And Copilot remains an excellent product when it works. But the CNBC report makes clear that the clock is ticking. Developers are famously fickle, and in the fast-moving world of AI tools, a week of downtime can push a team to a competitor for good.

At the Microsoft Build developer conference just three weeks before the CNBC report, CEO Satya Nadella spent nearly half his keynote on AI, yet mentioned Copilot reliability only in passing. Internally, however, pressure is mounting. Two sources told CNBC that Amy Hood, Microsoft’s CFO, has requested a complete review of Copilot’s operational spending and outage costs, signaling that the board is paying attention.

A cautionary tale for the AI industry

GitHub Copilot’s troubles serve as a warning for the entire AI sector. As companies race to ship increasingly sophisticated capabilities, the temptation to prioritize flashy demos over boring reliability is strong. But for tools that have become as essential as a compiler or source control, downtime is existential. The coming months will test whether Microsoft can rediscover the agility that made Copilot a sensation—or whether the very disruption it unleashed will consume it.

For Windows developers who have made Copilot a daily companion, the message is clear: diversify your toolbox. Many teams are already keeping Cursor or Codeium as a backup; some are running local models with LM Studio or Ollama as an offline safety net. The era of AI monoculture may be ending, and that could be the healthiest outcome for the industry—even if it comes at Microsoft’s expense.