There used to be a small, private ritual after you typed git push. You watched the terminal, you waited for the checks, and you quietly hoped. Lately that hope has been getting a workout, because the platform on the other end keeps having a bad time. So we drew the ritual as what it has become: a prayer.
What happened
The short version: AI coding agents are generating commits, pull requests and Actions runs much faster than GitHub planned for. The Register, DevX and OfficeChai all describe a steady run of incidents since late 2025. They touched Actions, pull requests, search, webhooks and Copilot.
The capacity plan tells the story. According to this write-up, GitHub's engineering leadership had a plan for a large multiple of its earlier scale in autumn 2025. By early 2026 that plan had grown considerably. The target now is to design for many times the load the platform was originally built around.
A few other threads recur in the coverage:
- Agent-authored pull requests have grown several-fold in a matter of months, according to this analysis.
- Actions compute usage is reported to be a multiple of what it was a few years ago.
- Agents hit the API and CLI around the clock, while limits and pricing were designed around humans who sleep, eat and attend standup.
- Reports describe at least one worldwide outage that lasted hours, and unofficial uptime trackers put availability well below what an Enterprise SLA suggests. Treat those trackers as indicative, not official.
- GitHub is adding capacity by moving workloads to Azure. There is also a reported overflow arrangement with another cloud, which Microsoft has not confirmed.
We are deliberately not quoting exact figures here. Several of the numbers come from secondary blogs and trackers, and the shape of the problem is clear without them.
Why engineers get it
Nobody is mad at the load. Everyone has seen a service built for one traffic pattern meet a very different one. Rate limits tuned for humans, queues sized for human typing speed, and webhooks that assumed events arrive politely are all fine until something that never sleeps starts opening pull requests in parallel.
The irony is the part that sticks. The tools were built to help people ship faster, and the shipping lane is now the bottleneck. The bots are, in a very real sense, queuing behind each other to get into the building.
Then there is the daily experience. You push, the checks spin, and you can't tell whether your tests are slow, your runner is queued, or the platform is having a moment. You open the status page, you refresh it, and you accept that it may be down more often than you would like. You do what engineers have always done in this situation: you wait, you refresh, and you hope.
That last part is the whole joke. Deploy-and-hope was always a bit of a meme. Now the hope has an infrastructure dependency.
The design
The design shows a tired engineer in a hoodie kneeling in front of a glowing laptop, hands pressed together. Above the terminal floats a tiny halo containing a spinning pending-checks circle. The printed line reads "git push and pray". It works on a tee for the person who merges before lunch and a mug for the person who is still waiting on the checks from yesterday.
It is not aimed at anyone who works on the platform. They are the ones holding the pager while the queue grows. It is aimed at the situation we all share: a spinner, a queue, and a quiet hope that the green check arrives before the coffee goes cold.