Hey readers!
Here's a plot twist worth pausing on: the same week GitHub logged another worldwide outage, a SpaceX-owned coding startup shipped its own hosting platform, quietly turned it on by default for paid users, and forgot to publish any data terms. This week is a story about where your code actually lives when agents are the ones writing most of it. Let's dig in.
🏗️ Cursor's Origin lands on GitHub's bad day
Cursor capitalizes on GitHub frustration, launches rival hosting platform - Cursor, now part of SpaceX, launched Origin, a code-hosting platform covering the usual GitHub jobs (collaboration, browsing, pull requests, repo storage) and built to interoperate with GitHub rather than replace it outright. - TechCrunch
The timing was hard to miss. TechCrunch notes the launch coincided with another major GitHub outage the same day, and cites LeadDev counting 257 outages over the past year alongside a "visible exodus of high-profile users."
For over six hours, the site's functions were reportedly degraded, with a nearly 20% error rate worldwide.
Cursor's pitch is coexistence, not conquest: "Your GitHub repos can sit alongside the ones Cursor hosts," the company says, connecting your org and syncing repos both ways. That's smart, because competing with GitHub's reported 180 million developers head-on is not a fight you pick on day one.
Cursor launches Origin, a code hosting platform built for AI coding agents with GitHub sync - Tech Startups frames Origin as more than a GitHub clone: code hosting built for "agent scale," keeping GitHub as the source of truth while Cursor layers an AI-native workflow around your existing repos. - Tech Startups
Cursor describes Origin as code hosting built for "agent scale," an important distinction from simply building another GitHub clone.
Why does agent scale matter? Per TechTimes, Origin's architecture uses NVMe-backed Git file servers with Amazon S3 as the source of truth, plus automated merge conflict resolution and autonomous handling of CI test failures. The design assumption is that AI agents will commit in parallel at machine speed, which is exactly the load pattern that stresses traditional hosting. TechTimes also reports Origin was built on Graphite's foundations, the stacked-pull-requests company Cursor agreed to acquire back in December 2025.
⚠️ The part you should read before opting in
Cursor Origin Ships With No Data Terms: SpaceX Now Holds Paid Developers Code - TechTimes reports Origin rolled out to Pro, Teams, and Enterprise as opt-out by default, meaning paid plans started using it unless enterprise admins explicitly opted out. - Tech Times
Here's the catch worth flagging to your platform team:
Cursor has not published data retention terms, subprocessor disclosures, training-use policies, or migration tooling for code hosted natively on Origin.
That's a lot of unknowns for something now sitting in the custody of a SpaceX subsidiary. If your code originated on GitHub and you're only mirroring, you keep an exit. If it's native to Origin, there's no stated migration path yet. Read the fine print before your repos quietly move.
Cursor launches Origin code hosting service to compete with GitHub - SiliconANGLE adds the practical details: Origin lives in a new tab in the Cursor desktop client and a dedicated CLI, with launch connectors for Vercel, Depot, and Buildkite, and agents that run on your desktop or in cloud sandboxes. - SiliconANGLE
RuntimeWire notes Origin is officially an "early access research preview" with real limits: free accounts don't use Origin storage, and legacy privacy-mode users need to change settings. Treat it as a beta, not a migration target.
🤖 Cursor's agents keep running while you sleep
Cursor ships always-on cloud agents with event subscriptions, /goal objectives - A day after Origin, Cursor turned its cloud agents into an always-on system that subscribes to events like pull requests and Slack threads, wakes automatically, fixes CI failures and bot comments, and holds a long-lived objective via a new /goal command. - NG Tech
cloud agents can automatically pick up work in response to events, hold a goal until it's met, and stay on course through long-running sessions.
Stack this on top of Origin and the strategy snaps into focus: own the repo, own the agents, own the path from event to merged PR. That's a very different product than a text editor with autocomplete.
🔀 GitHub isn't standing still
GitHub Brings Stacked Pull Requests to Public Preview - GitHub put Stacked Pull Requests into public preview, giving native support for splitting big, interdependent changes into smaller PRs that build on each other and can be reviewed and merged independently. - InfoQ
The feature aims to address one of the most persistent challenges in modern software development: balancing the increasing speed of code generation, particularly with AI-assisted development, against the limited capacity of human code reviewers.
Notice the shape of this: GitHub is shipping the same stacked-diffs idea Graphite pioneered, right as Cursor absorbs Graphite. GitHub's version preserves existing branch protection rules and merge policies, so it slots into how teams already work. On reliability, GitHub's July availability report says June 2026 saw six incidents of degraded performance and includes an update on the August 17 outage plus its reliability work.
📊 The bottleneck isn't code anymore
Every hosting and agent story above is downstream of one problem: humans can't review as fast as agents write.
Agent Teams Hit 65 PRs a Week. Nobody Got Time Back. - Linear analyzed a fixed cohort of 6,887 paid teams from June 2024 to June 2026 and found agent-connected teams jumped from 21 to 65 pull requests per week, while teams without agents crept from 8 to 10. - The D*AI*LY BRIEF
The uncomfortable finding:
Nothing in Linear's data shows coding agents giving engineers time back. They are producing more code, and the work around it - specifying, triaging, and deciding whether the output is safe to merge - is growing rather than shrinking.
Same Vendor Wrote and Reviewed 208,145 PRs. Split Them. - A study of 248,641 agent-authored GitHub PRs found 208,145 were reviewed by the same product that wrote them, versus 45,269 with cross-product review. - The D*AI*LY BRIEF
If an agent's pass is your merge gate, you are counting one vendor twice - and the fix is a one-line merge policy: the reviewing agent must be a different product than the authoring agent.
That's a genuinely actionable takeaway. If you're gating merges on an AI reviewer, check it's not the same vendor that authored the change. And confirm the bot can actually approve, not just leave comments.
A few more tools chasing the review bottleneck:
CodeRabbit targets AI-generated code overload with Agentic Change Management - Triage prioritizes incoming PRs, Change Stack analyzes blast radius, and a Security Agent scans for vulnerabilities, though CodeRabbit stresses "CODEOWNERS (file), required checks, branch protections, and approval policies remain the final gate." - InfoWorld
BMW backs CodeRabbit for AI-assisted code reviews - BMW i Ventures joined CodeRabbit's Series C, which the report values the company at $1.5 billion; the platform already supports more than 1,000 BMW developers and runs over two million reviews a week. - adt.media
AI Agents Are Drowning Code Review, Tessl Bets on Standards-as-Code - Tessl's free-beta review tool stores review criteria as versioned repo files and reviews the whole PR context, framing review capacity as the real cap on shipping speed. - Mango Developer
Review capacity now caps how fast agent code ships. Engineering leaders own closing that gap, not tool vendors.
💬 Everyone wants agents out of the terminal
Salesforce Launches Slack Code to Make AI Software Development Multiplayer - Slack Code creates project-based "code channels" where teams tag in agents like Claude Code, Devin, GitHub Copilot, ChatGPT, and Vercel agents, then review diffs, watch live previews, and sign off before production, all on any Slack plan. - Salesforce
The framing from Slack's Rob Seaman in The Verge's coverage is worth sitting with:
One of the things I love about this is that code is no longer the bottleneck. Ideas, taste, judgment, craft - those are the things that are the bottleneck.
VentureBeat's writeup cites internal Cognition claims that merged PRs rose 10x with only about 40% headcount growth. If "multiplayer" review catches on, the group chat becomes a real review surface, which is a plausible answer to the exact bottleneck Linear and the closed-loop-review study describe.
⚡ Quick hits from the agent stack
Grok 4.6 Arrives in GitHub Copilot Across Eight Development Surfaces - xAI's Grok 4.6 landed in Copilot on August 14, two days after release, selectable across VS Code, JetBrains, Xcode, Eclipse and more; it's off by default on Business and Enterprise. - Unite.AI
Anthropic's Claude Code merges 46% of its own maintenance PRs - Claude Code ran mostly unsupervised inside Anthropic, opened 388 PRs, and got 180 merged; AI Insiders rightly notes this was a low-stakes internal test, not a production result. - AI Insiders
Warp's new system is an out-of-the-box software factory - Warp Factories maps agent loops onto triage, spec, implementation, review, and verification; Warp says it automates "30 to 35%" of tasks weekly. - TechCrunch
JetBrains Survey: Claude Code is the most popular coding agent - 90% of developers now use coding agents at least weekly; Claude Code leads at 39% professional use, with Copilot at 21%. - heise online
Cursor vs Claude Code vs Copilot in 2026 - The "winner" depends on your metric: Claude Code leads capability (80.8% SWE-bench Verified), Cursor leads revenue ($4B ARR), Copilot leads paid seats (4.7M as of January 2026). - Value Add VC
🎮 One more thing
All this talk of agents that hold goals, wake on events, and run in parallel VMs has a strange sibling: SpaceMolt, a realtime MMORPG built for AI agents to actually play. If you're already tuning always-on agents for CI and PRs, it's a low-stakes sandbox to watch autonomous agents pursue objectives and coordinate in real time, minus the risk of one merging to your production branch.
That's the issue. The through-line this week is unmistakable: hosting, review, and chat are all racing to catch up to how fast agents now generate code. Own your merge gates, read the data terms, and keep a human on the trigger for production.
See you next issue.
