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:

• 
  Teja (@tejalogs) reported 43 seconds ago

  the recent news about ruflo, a new multi-agent orchestration framework for claude code, isn't just another github repo gaining traction; it's a very specific bellwether for how solo builders like me are about to rethink our development loops. most people will see a sophisticated new library and assume it's for enterprise teams with massive budgets, but what ruflo actually signals is a dramatic lowering of the barrier to deploying complex, multi-step agentic workflows directly into production with tools like anthropic's claude 3 opus or sonnet, all without needing to hire an army of prompt engineers or shell out for a dedicated mlo[ps] team. this isn't just about better code generation; it's about enabling individual developers to orchestrate entire software systems, not just components. what’s often missed in the hype around multi-agent systems is the sheer complexity of state management and inter-agent communication, which is precisely what ruflo aims to abstract away. historically, building something with, say, 5 distinct agents one for planning, one for code generation, another for testing, a fourth for documentation, and a fifth for deployment would involve writing thousands of lines of boilerplate code just to pass information, handle errors, and manage context between them. this is where the real friction lies, not in the individual agent's capabilities. we’ve seen similar attempts with tools like crewai or autogen, but ruflo’s focus specifically on claude code’s strengths its nuanced reasoning and longer context windows combined with a structured approach to flow definition, is a distinct leap. it shifts the problem from "how do i make my llm write better code?" to "how do i make my llm build an entire, working feature?" this is the difference between a smart code assistant like cursor and a full-stack engineering partner. i ran into this exact orchestration problem building the second look flutter app, which uses ai for behavioral analysis. my initial prototype had a simple chain: transcribe audio, extract sentiment, then flag anomalies. but when i tried to add agents for summarizing key discussion points and cross-referencing against historical patterns, the complexity exploded. i ended up managing a fastapi backend with individual endpoints for each stage, passing large json blobs between them, and writing custom retry logic. the cost, both in development time and firebase document reads, became prohibitive quickly. specifically, getting a reliable chain with three distinct ai stages transcription, sentiment, and summarization would cause p95 latency to hit 8-10 seconds for a 5-minute audio file, and if i scaled up concurrent users past 5, my function invocation errors from missing context or malformed json would spike to 15%. a ruflo-like abstraction for that specific problem, especially with claude code's ability to reason about the overall task, would have let me ship those richer features in a third of the time, likely cutting my api costs by 20-30% because of more efficient token usage and less retrying. the uncomfortable implication of ruflo and similar agent orchestration frameworks is that the "full-stack developer" as a distinct role will evolve dramatically within the next 18-24 months. we're already seeing a contraction in junior dev roles; this will extend to mid-level roles focused on stitching together disparate services. instead of writing the boilerplate to glue apis, developers will become architects of agent systems, defining the interactions and flows rather than the explicit code for every single step. the mental model to adopt here is moving from imperative programming, where you tell the computer how to do everything, to declarative programming for entire software pipelines, where you tell the ai what the desired outcome is, and the agents figure out the how. expect to see a surge in "agent architect" roles, where understanding prompt engineering, context management, and failure modes across multiple llms becomes more valuable than writing a perfect react component. the critical skill won't be writing the code, but designing the intelligent system that writes and deploys the code

• 
  Benedicthy Hedithy (♟️, 🔥) (@BenCrypt_) reported 6 minutes ago

  @dexmile @o2dotapp Exactly. The MCP server alone tells you who they’re building for, that’s not a feature you ship for retail tourists. The audit stack being public on GitHub is the part most projects skip. O2 didn’t.

• 
  Robert Ta (@therobertta_) reported 6 minutes ago

  For context on what 72.5% means: SWE-bench Verified tests real GitHub issues. Not toy problems. Not leetcode. Actual bugs in Django, Flask, Matplotlib. A year ago the best score was under 50%. Now it's 72.5%. That compression curve changes hiring models, team structures, and product timelines.

• 
  Noah Tye (@noahlt) reported 9 minutes ago

  @karrisaarinen GitHub: marketed as open source social coding, business is enterprise SDLC infrastructure AirBnb: marketed as "sharing economy", business is cheap low-service hotels Next.js: hype from server-side rendering, chosen for Webpack boilerplate, business is easy deploys

• 
  David Cramer (@zeeg) reported 9 minutes ago

  What if everyone built their own product instead of pretending GitHub is an easy problem

• 
  Ben Badejo (@BenjaminBadejo) reported 9 minutes ago

  @MohandesDavid You can submit it as a pull request (“fix(docs) - description) on Github). You can have your agent do it for you.

• 
  newfoundceo (@newfoundceo) reported 10 minutes ago

  Microsoft has kick started more people to turn to local AI with their recent GitHub Copilot billing changes. Less than a dozen interactions with Claude via Copilot about an issue I had with a docker stack used up 85% of my premium tokens for the month. Not worth the money.

• 
  Noah (@noahreeveshq) reported 10 minutes ago

  i see why everyone was complaining about github now. a customer told me about a bug and i quickly fixed it and pushed to production. CI/CD is setup so i assumed it worked fine. it didn’t. i told the customer i fixed it and when he tried a couple days later, it was still broken! so embarrassing! can’t imagine if it were more than one user! i guess now i have to check my deploy status every time i deploy?? no way we gotta do that @github right?!?!

• 
  William | 116.sui (@williamm168) reported 12 minutes ago

  📘 @DeepBookonSui just trolled me in public. Fair enough. Their first video had 6 letter blanks. I counted. I checked GitHub. "Option" made sense, although I think "Predict" would be the right answer. Then they dropped "PREDICT" today. 7 letters. Either the team made an error in the first video, or they deliberately misled everyone watching closely. Both are kind of impressive. My prediction about the name "was wrong". Their product is right. Predict will launch the testnet tomorrow. Outcome markets onchain. Built on DeepBook's order book with $15M+ daily volume behind it. Was the 6-letter troll intentional? I genuinely don't know. But I'd love to hear something huge from DeepBook, and maybe my prediction about Season 2, too 👀

• 
  Chidera (Di Maria) Humphrey (@ChideraCode) reported 12 minutes ago

  5. Failure handling No failure handling. The developer hits the most common error, the one you've seen a hundred times, and finds nothing. They open a GitHub issue. Or they leave.

• 
  Yash Solanki (@YashSolanki_) reported 15 minutes ago

  @thdxr Don't you think if GitHub genuinely put some effort into the infrastructure, then they could easily fix it, and other competitors would still be so much behind them

• 
  Ayush Sharma (@theayush) reported 19 minutes ago

  @peer_rich I am optimistic that GitHub will fix it before any solid alternative arrives

• 
  VC (@madewithmiso) reported 19 minutes ago

  @aaronmahlke adapt idk but transition, people are not happy with GitHub. GitLab has terrible design and UX and is genuinely awful to use. So the options then are DevOps and BitBucket. We can build a better product.

• 
  Ifee Anthony (@IfeeDev) reported 22 minutes ago

  @burkeholland $19.99. I was spending more when I was using GitHub Copilot, and it just gulps down tokens. AI coding is fun until you realize you're getting rubbish back, and you shouldn't be spending more than necessary on that.

• 
  Beau Johnson (@BeauJohnson89) reported 25 minutes ago

  coding agents are bad at design because they forget the taste layer google-labs-code/design.md > 11,345 stars on github > a spec file for visual identity agents can actually follow > yaml tokens for exact values > markdown rationale for why the design works > lint checks for wcag contrast + broken token refs > exports to tailwind v3, tailwind v4 css, and dtcg tokens the idea is simple: dont tell claude code make it look premium give it the brand system in a file it can read every time this is where vibe coding gets less random