GitHub Shut Down a Major AI Builder Overnight—Here's what happened why it gets worse in 2025

GitHub took down lovable overnight after the AI builder hit a terms-of-service violation, leaving the service unable to create GitHub repositories for hours and triggering a cascading outage for a major customer. Lovable’s team had checked with GitHub before the holidays about growth, quotas, and rate limits—and was told things were fine—yet during the night of January 2 in the US, GitHub effectively “put them in jail” without clear public details. When the first working day of the year arrived in Europe, the disruption became urgent: lovable’s users couldn’t ship new repos, and the company scrambled to route work through Amazon S3 as a temporary workaround while waiting for GitHub to restore access.

The episode matters less for the specific outage and more for what it signals about 2025’s scaling pressures. Lovable had been generating GitHub repositories at an astonishing pace—about one every two seconds—an exponential growth rate that outpaced what GitHub is designed to absorb. The transcript frames GitHub’s lack of responsiveness during the holiday window as a key operational risk: when growth spikes and support channels aren’t staffed, even a temporary enforcement action can turn into hours of downtime.

A second takeaway is strategic: an AI tool that depends on GitHub as its primary “engine” may eventually need to diversify away from GitHub-only workflows. The transcript suggests lovable will likely move toward Amazon S3 to scale, even though GitHub’s social and discoverability advantages—its shared ecosystem and the way developers naturally browse and trust repositories—are hard to replace. That tradeoff highlights a broader tension: AI builders want to reinforce the “GitHub flywheel” by writing code into the platform, but the platform may not be built for the volume and automation patterns that agentic systems produce.

Looking ahead, the transcript argues that 2025 will stack two exponential curves. First, the number of people interested in coding surged in 2024 because large language models make coding accessible, driving a roughly 10x increase in code activity. Second, autonomous agents are expected to start coding on their own within months, multiplying the number of commits and repository interactions again. Even if much of the output isn’t high-quality, sheer volume can still overwhelm systems, trigger enforcement, and reshape usage patterns.

The practical warning extends beyond GitHub. Any business whose architecture assumes human-paced engineering—where users interact with software in predictable ways—could be disrupted when AI agents gain access. The transcript offers an example from a SaaS marketing business: if an agentic browser and tools let marketers automate report generation and then spawn multiple agents for different tasks, those agents may interact with the service through the same user accounts, changing traffic patterns, security exposure, and operational load. In that sense, lovable’s outage is presented as an early 2025 warning: more enforcement events and scaling failures are likely unless providers refactor architectures and plan for agent-driven usage at far higher throughput.