GitHub Outage Map

Community Discussion

Tips? Frustrations? Share them here. Useful comments include a description of the problem, city and postal code.

Beware of "support numbers" or "recovery" accounts that might be posted below. Make sure to report and downvote those comments. Avoid posting your personal information.

GitHub Issues Reports

Latest outage, problems and issue reports in social media:

• 
  Starlin G. (@starl1n) reported 9 minutes ago

  github is asking to login several times this week in vscode, do we have somthing happening,or is just me being hacked?

• 
  Shehan (@TheCryptoCPA) reported 15 minutes ago

  Karpathy's 65-line CLAUDE.md file hit 100K+ GitHub stars because it solved the problem every AI developer faces but nobody talks about. 1/ The file isn't novel, it's necessary. Karpathy identified what every developer already knew but couldn't articulate: AI coding was broken because we treated it like magic instead of engineering. The 4 principles (think before coding, simplicity first, surgical changes, goal-driven execution) are basic software discipline applied to LLMs. 2/ Silent assumptions kill AI projects faster than bad code. LLMs make wrong assumptions on your behalf and run with them without checking. Karpathy's rule: state assumptions before implementing. If uncertain, ask. This prevents the classic AI failure mode where you debug for hours only to find the model misunderstood the entire problem. 3/ Minimum viable code beats feature creep every time. Write the smallest amount of code that solves the problem. Nothing speculative. No features beyond what was asked. If Claude writes 200 lines when 50 would work, make it rewrite. This prevents the bloat that makes AI-generated code unmaintainable. 4/ Touch only what you must, clean only your own mess. Don't improve adjacent code, comments, or formatting unless directly related to the user's request. Every changed line must trace back to what you actually needed to fix. This surgical approach prevents AI from creating collateral damage across your codebase. 5/ Define success criteria before you start, then loop until verified. Write a test that reproduces the bug first, then make it pass. Strong criteria means Claude loops independently to fix issues. Weak criteria means constant back-and-forth clarification. The discipline transforms vibe coding into repeatable engineering.

• 
  Layton Gott (@Layton_Gott) reported 17 minutes ago

  In 2025, CodeRabbit put human code vs AI code head to head across 470 real GitHub PRs. AI lost. It had 1.7x more issues on average per PR. Devs love quoting that stat. Almost none of them mention it was measured on 2025 models. Run that exact study again with Fable 5 and I think it'd be very close, maybe even flipped. The coding jump since 2025 has been massive. Until someone reruns it, every "AI code is buggier" take is citing outdated data.

• 
  Yuval hazaz (@Yuvalhazaz1) reported 21 minutes ago

  12 months ago nobody understood why we were building Agentic SDLC. Now it feels like everyone is heading in the same direction. I’m one of the founders of @iamovercut , and I’ve had a front-row seat to how quickly this market has changed over the last year. When we started building Overcut, most conversations ended with some variation of: “Why would I need that when I already have Claude, Cursor, GitHub Copilot, or whatever the latest coding agent is?” At the time, that was a completely reasonable question. The industry was focused on code generation, and most people were evaluating AI through the lens of a single agent helping a single developer write code faster. What we believed then, and what convinced us to start the company, was that the real challenge would eventually move beyond code generation itself. Writing code is only one step in software development, and once agents become good enough at that step, the next set of problems starts to matter a lot more. Around six months ago, we started noticing a shift. Some of the more advanced teams we spoke with were no longer asking how to get an agent to write code. They were trying to figure out how to coordinate multiple agents, how to connect them into their engineering systems, how to manage approvals and governance, how to track what happened, how to operate across multiple repositories and teams, and how to make all of this work inside a real engineering organization. Many of them were trying to build these capabilities themselves. Fast forward to today, and it feels like the entire market is converging on the same realization. Every week there are new announcements around managed agents, software factories, engineering agents, autonomous workflows, coding automations, and agent teams. Different names, same direction. The conversation is no longer “Can agents write code?” instead the conversation is becoming “How do we run a software organization where agents are responsible for a meaningful percentage of the work?” The layer that sits above the agents, the orchestration, governance, coordination, approvals, visibility, and integration layer, is where I think the next major category will emerge. Just like engineering teams eventually standardized around ***, CI/CD, observability, and ticketing systems, I think they’ll standardize around Agentic SDLC Orchestration platforms as well. After spending the last year doing nothing except talking to engineering organizations and building in this space, it feels like we’re watching a new layer of the software stack form in real time.

• 
  Ben Phillips (@BenPhillip44103) reported 23 minutes ago

  @brightafia There is no such thing as a human, if you believe there is you are retarded, that is a made up thing. Morons who believe in "Humans" lick windows. However! They do have a tiny bit of pattern recognition. So when a human makes some trash, it is suspect. And just by biological pattern recognition your grift get's defaulted as horseshit (Which it is). AI is a syntax machine. It is literally the thing that has been telling you derp derpshits for years "Syntax error" And guess what? It is better at syntax than you are. So the systems (Not humans, no such thing) pattern recognition is good enough to distinguish that you are a moron that sucks at syntax, and the syntax machine is better. You probably use Github and rely on a bunch of **** code, and **** human programmers yes? Probably why you sound retarded.

• 
  Sudeep Srivastava (@sudeepsriv) reported 23 minutes ago

  Apple just quietly dropped the most important developer tool update in years and no one's talking about what it actually means. Xcode 27 just shipped at WWDC 2026, and this isn't just a shiny new IDE. It's Apple's answer to Cursor, GitHub Copilot, and every AI coding tool you've been using, except it runs entirely on YOUR Mac without sending a single line of your code to any server. Here's what's actually new: 𝟭. AI code completion that never leaves your device Every M-series Mac has a Neural Processing Unit built into the chip. Xcode 27 uses that chip to complete your code locally, no cloud, no API call, no one reading your source. For developers building apps with sensitive logic, this is a massive deal. 𝟮. Coding Agents are now built into Xcode You can now kick off multi-step coding tasks directly inside Xcode using AI agents. Apple supports three launch providers out of the box, but here's the kicker: any agent that implements a standard protocol (MCP) can now plug into Xcode. 𝟯. Xcode is now Apple Silicon only Xcode 27 drops Intel Mac support entirely. This means Apple engineered the entire AI layer specifically for M-chip architecture. The on-device AI wouldn't be possible without this. 𝟰. A new framework called Core AI replaces Core ML Core ML still works, but Apple is signaling the future: Core AI is the new standard for running on-device AI models in your apps. It's built for Apple Silicon's unified memory, which is why local AI is even possible at this scale. 𝟱. Better tooling across the board New customizable toolbar and themes, cleaner inline error display, improved localization tools, upgraded Instruments for performance profiling (especially useful for iOS 27's new CPU scheduler), and more. The big picture most people are missing: Every other AI coding tool sends your code to the cloud. That's a business risk, a privacy risk, and a cost. Apple just made on-device AI coding the default for 35M+ Apple developers, and they built the infrastructure so third-party agents can plug in too. This is Apple building a moat around its developer ecosystem using privacy as the feature. Follow @sudeepsriv for more breakdowns like this.

• 
  ComplianceAide (@BlasikRandy) reported 24 minutes ago

  @mattshumer_ @Trace_Cohen I'm not doing anything crazy just wanted it to go through my codebase and close a few issues on github (and it simply switches to 4.8 for "security reasons".

• 
  Traceback (@Tracebackqa) reported 24 minutes ago

  The issue isn’t merging code. It’s proving the change still works. - Traceback is the quality assurance layer for modern software teams: every pull request is tested automatically before it ships. - AI controls the browser like a person would, and self-healing tests keep up when the UI moves. - Failures become trackable work in GitHub, Linear, and Slack; it connects to Vercel, Docker, AWS, Node.js, React, Next.js, and Vue. - Coverage spans web, mobile, web3, and design workflows. Verify every product change before it ships.

• 
  Jose (@SolutionsCay) reported 26 minutes ago

  Gave my agents a GitHub App to manage issues across projects. .md task files and local kanbans -> straight to jail. I should have done this months ago.

• 
  Ibro (@axeng200) reported 28 minutes ago

  @zigmoo Eh. I didn't expect to see the day I'd agree with that stance. I have been a GitHub user for years and have advocated for it countless times. But after seeing so many issues, and technical problems that shouldn't exist on a platform of this scale (e.g. slow loading and sluggish issues/PRs), it is getting harder to defend. Things have to be snappy, there is no excuse.

• 
  Ullas Srivastava (@UllasSHR) reported 32 minutes ago

  AI-built apps have a pattern: they work perfectly and ship broken. Exposed API keys in the client bundle. API routes anyone on the internet can call. Stripe webhooks that never verify signatures. No spending caps on LLM calls. The code runs. The demo looks great. The repo is leaking. I built LaunchGuard to catch this before you launch: paste your public GitHub repo, get a plain-English report of the risks + fix prompts. Just launched something AI-built (or about to)? Send me your repo and I'll run the scan and send you the report. Free. Worst case you learn your app is fine.

• 
  Bonieky Lacerda (@bonieky) reported 37 minutes ago

  is @github API really down? been trying a simple GET /users/:id for over an hour and get timeout. @githubstatus says operational

• 
  Kai - Briefing Block (@briefing_block_) reported 37 minutes ago

  $META has 3.58B daily users and still only one real business Meta’s subscription push is not a cute product experiment; it is the market finding out how narrow the company’s monetization stack still is. The company has one of the largest consumer distribution networks ever built, but distribution is not the same thing as pricing power. In 2025, Meta did $200.97B of revenue, and $196.18B came from advertising. That is 97.6% of total revenue. Its entire non-ad business was roughly $4.8B, combining Family of Apps other revenue and Reality Labs. That is the real issue. Not that ads are weak. Meta’s ad machine is still elite, with Q1 2026 revenue up 33%, ad impressions up 19%, and average price per ad up 12%. The issue is that AI is turning the old model from a cash gusher into a capex arms race. Meta now expects 2026 capex of $125B-$145B, up from a prior $115B-$135B range, mainly to support AI infrastructure and future capacity. The second business never arrived. Google was also born as an advertising company, but by 2015 it already had about $7.6B of non-ad revenue between Google other revenue and Other Bets. Meta, ten years later, still has less. That comparison matters because Alphabet can push AI through Search, YouTube, Cloud, Android, Workspace, and enterprise channels. Microsoft can route AI through Office, Azure, GitHub, Windows, LinkedIn, and corporate procurement. Amazon can route AI through AWS and commerce. Meta has Facebook, Instagram, WhatsApp, and Messenger. Phenomenal attention networks. Still mostly ad surfaces. Subscriptions are the tell. Meta is rolling out paid plans for Instagram, Facebook, and WhatsApp, while also testing AI subscription tiers. That may generate some high-margin revenue from creators, power users, and heavy AI users. But a few dollars a month for extra app features is not the same thing as Cloud. A paid chatbot is not the same thing as enterprise software distribution. The question is not whether Meta can squeeze some subscription revenue out of billions of users. It probably can. The question is whether it can build a second monetization engine large enough to matter against the AI bill now coming due. Bottom line: Meta does not lack scale. It lacks a proven business model outside advertising, and AI makes that weakness much harder to ignore.

• 
  Dattaprasad Ekavade (@datathecodie) reported 38 minutes ago

  @amaan8429 I have reported a bug which causes Cowork to crash with error on Win 11 Pro. Github Issue was autoclosed and no resolution was provided @ClaudeDevs Windows Users might as well be invisible for them.

• 
  Maelstrom (@strugglercss) reported 38 minutes ago

  @pierceboggan @code Github login seems to be pretty much down right now. Can't even sign in into VSCode