☕️ GitHub back up after 8-hour outage

Hi there, this is your daily ☕️ Devshot.

In today's Devshot:

🐙 GitHub back up after 8-hour outage

🖥️ Cursor launches Origin to rival GitHub

⚙️ Cloudflare CI pipelines are now TypeScript

☸️ Kubeflow graduates from CNCF

⚡ Oxlint adds React Compiler lint

🦆 DuckDB v2.0 adds JSON patch functions

Plus: 🎁 6 other news you might like, 🧰 6 tools, and 📚 5 papers.

🐙 GitHub back up after 8-hour outage LINK

GitHub is back online after an 8-hour disruption on Monday that spiked error rates across the platform and left thousands of developers reporting failures on Downdetector before the company rolled out mitigations.

At its peak the outage hit roughly 20% of web traffic with errors, while archive and content downloads failed at about 50%, and GitHub said it was still hunting for the root cause hours in.

Some other Microsoft-owned products stumbled too, with hundreds of users flagging problems with Copilot and Teams, as GitHub's CTO has noted the platform is scaling capacity 30x to keep up with AI-driven development.

🖥️ Cursor launches Origin to rival GitHub LINK

Cursor has launched Origin, a native code hosting platform that pulls repositories, pull requests, reviews, merges, and CI connections directly into the editor, positioning it as a direct competitor to GitHub.

Now in beta for paid users, Origin lets teams sync existing GitHub repos as the source of truth and ships with a unified dashboard, a Codebase tab, and integrations for Vercel, Buildkite, and Depot.

Cursor pitches Origin as hosting built for "agent scale," aiming to own the repository layer so AI agents creating branches, editing files, and opening pull requests can operate closer to the deployment pipeline.

⚙️ Cloudflare CI pipelines are now TypeScript LINK

Cloudflare shipped a CI SDK that lets developers write continuous integration pipelines in TypeScript instead of YAML, running each step as a durable Cloudflare Workflow targeting the Workers runtime rather than Node.js.

Because every stage maps to a Workflow step, execution is checkpointed: a failed step retries with preserved state and resumes from that point, while independent checks like lint, test and build run concurrently before deploy.

Dependency caching stores install results as Sandbox filesystem snapshots in R2, but the tradeoff binds you to Workflows, Sandboxes, Containers and Durable Objects, and commands must be idempotent since retries can duplicate side effects.

☸️ Kubeflow graduates from CNCF LINK

Kubeflow, the Kubernetes-based platform for standardizing data and AI workloads, has graduated from the CNCF, confirming its technical maturity as a production backbone for data processing, model training, fine-tuning, and inference.

Created at Google in 2017 and a CNCF incubating project since 2023, Kubeflow now has more than 6,600 contributors across 1,000+ organizations, 33,000+ GitHub stars, and nearly 260 million PyPI downloads of its Python packages.

To graduate, the project passed a third-party security audit, formed a steering committee, and adopted the CNCF Code of Conduct; its roadmap targets LLM orchestration, post-training fine-tuning, large-scale data engineering, and agentic workloads.

⚡ Oxlint adds React Compiler lint LINK

Oxlint shipped a native React Compiler lint rule in v1.70.0, adding a nursery `react/react-compiler` check that runs the vendored `oxc_react_compiler` crate in lint-only mode instead of leaning on the Babel-based JS plugin escape hatch.

The native Rust implementation cut one author's lint task from ~29s to ~9s, a 3.2× speedup, and it catches Rules of React violations like conditional hook calls, setState during render, and ref access during render.

Enabling the Oxlint-only `reportAllBailouts` option flags every case where the compiler silently skips memoization, letting you enforce a fully compatible codebase and prevent accidental performance regressions from things like nullish prop reassignment.

🦆 DuckDB v2.0 adds JSON patch functions LINK

DuckDB v2.0 will extend its JSON extension with four new scalar functions, json_merge_patch_diff, json_deep_merge, json_normalize, and json_strip_nulls, already available to try in the v2.0-dev preview build.

json_merge_patch_diff computes the minimal RFC 7396 patch to turn one document into another, json_deep_merge treats null as "keep original" rather than delete, json_normalize sorts keys for hashing, and json_strip_nulls recursively removes null-valued keys.

The functions target data-pipeline reconciliation, letting you strip nulls, compute minimal CDC patches, and hash canonicalized documents in a single query rather than leaving the database, benchmarked against equivalent Python on 500,000 synthetic events.

Other news you might like

• SpaceX officially closes its Cursor acquisitionLINK

• Linux 7.3 merge window kicks off by throwing out two ancient file systemsLINK

• Same Cluster, 33 Points More Utilization: What Changed Was the OrderLINK

• Quake shareware, a CD just a little too fullLINK

• Qwen3.8-27B runs frontier-class coding agents and reasoning locally, no cloud API requiredLINK

• How Databricks Feature Store serves features with sub-second freshnessLINK

🧰 Trending tools

Sim: a visual and code-based workspace for building and deploying AI agents, connecting to 1,000+ integrations and every major LLM provider.LINK

Prefactor: an evaluation layer that scores agent runs in real time, catching quality regressions and drift before they hit production.LINK

ZooData: converts webpages into structured JSON, cutting LLM token usage by 75% while delivering pre-analyzed Amazon and TikTok e-commerce data for AI agents.LINK

CartAI: automates checkout across merchant sites via one API call, managing payments, shipping, and order tracking without per-site integration work.LINK

ditto.site: converts any public URL into clean Next.js or Vite code deterministically, preserving components, design tokens, and hover states via a free API or MCP server.LINK

Replay QA: records website sessions with full devtools access, letting you create detailed, shareable bug reports without manual QA processes.LINK

📚 Trending papers & reports

Robot mission planning tools compares four common ways of specifying what robots should do in the field, finding no single approach covers all needs, helping teams pick the right one for real deployments instead of guessing.LINK

AI code-fixing shortcuts that rely on a model's own confidence to catch its coding mistakes actually make accuracy worse in most tests, while having it verify and rerun code correctly boosts pass rates by ~6 to ~26 points.LINK

Agent failure detection tests whether AI systems can spot and pinpoint what went wrong when automated agents and tools interact, finding even top models identify the fault type only ~25% of the time.LINK

Long-horizon agent failures get a new diagnostic benchmark of 1,140 failed AI task-runs, testing whether tools can pinpoint which step, out of hundreds, actually caused the breakdown.LINK

AI coding teams reveal that letting multiple AI agents share files instead of messaging each other cuts communication costs by about 42% at eight agents on message-heavy work, showing how to make multi-agent AI cheaper to run.LINK

See you tomorrow for a new dose of ☕️ Devshot!