Claude went down again, and a user recomputed its real 90-day uptime: 97.68%

A festive status page and a number it won't show you

For a few hours, Claude Code and the Claude apps returned a wall of errors — 500s for some, "API Error: 529 Overloaded" for others, a 503 fault filter abort on the completion endpoint. Anthropic's incident page lit up across products, which is roughly where the actual story starts, because the page became the joke and then the evidence.

The status board advertises numbers like 99.27% for Claude Code and 99.93% for Claude for Government — "a holiday christmas ornament," as one commenter put it, all reds and greens. Two readers then noticed the board quietly changes the displayed number with your screen size: it picks how many days to display by viewport, so a phone shows a different uptime than a desktop. So one of them scraped the page's own underlying data and ran the math.

"Actual 90d uptime: 97.6838% … Partial outage: 43h 15m 1s — Major outage: 6h 46m 48s — Total affected time: 50h 1m 49s … So, only one 9 for 10x vibes." — kordlessagain, who recomputed from the page's data

Fifty hours of degraded service in 90 days is a long way from the four-nines aesthetic the dashboard projects.

"My job is to write loops"

The debate that gave the thread its edge wasn't about minutes lost. It was about who, or what, writes the infrastructure that keeps falling over. One comment paired a quote attributed to Claude Code's head with a pointed inference.

"If even Anthropic, the company with the world's best agentic vibecoders, has horribly unreliable infrastructure, it really says something about the quality of the world's best agentically produced code." — MontyCarloHall

That drew immediate pushback on the logic itself — call it the "AI did it" fallacy.

"if you know a company uses AI, even infra scaling issues must be due to AI… The usual response goes something like 'well they made claims that AI is good' therefore anything short of perfection supposedly debunks the claim." — hombre_fatal

A third voice split the difference with a hard-won example. A developer who uses Claude Code daily described the model, asked to compare range-bound hex integers, converting the numbers to strings and doing substring matching on "0xCAFE" — "because, in a comment, it saw '0xCAFE...' as a string. And the thing saw a pattern." His verdict: useful as a bug-finder and rubber duck, dangerous as an unsupervised author, and the kludges around the models "rest on the shoulders of actual giants: projects like Linux, QEMU."

"Claude Code sucks. Claude Code CLI sucks… It's sad to have useful tools (the models) and to make such poor use of them." — TacticalCoder, who codes with LLMs daily

The dependency nobody wants to say out loud

Underneath the jokes — "time to socialize again," "I had to review a PR by hand, I felt like a savage" — sat a real anxiety about lock-in. One comment imagined a future where Anthropic raises prices 200% because no one can code by hand anymore; the replies didn't panic, they listed exits.

"I use Codex, Antigravity, Claude and GLM-5.2 interchangeably. Any sensible employer will do the same." — kordlessagain, building agentic tooling

That's the practical takeaway buried in a meme thread: the outage didn't expose bad code so much as a single point of failure people have quietly built their workday around. The ones who shrugged were the ones who'd already wired a second provider into the loop.