Cursor
The AI code editor that won the market — a VS Code fork where the agent is the primary interface, now the center of two stories at once. Product-wise it ships its own frontier coding model (Composer) alongside every major API model, an agent mode that runs multi-file changes and terminal commands, and a background-agents window. Business-wise SpaceX agreed in June 2026 to buy its parent Anysphere for $60B (closing Q3 2026), folding the dominant editor into the same SpaceXAI umbrella that makes Grok — so it's no longer a neutral tool but the front end of a model-plus-agent-plus-editor stack. Worth a fresh look after years away; a GUI-first paid subscription that's the opposite pole from a terminal-and-local workflow.
Cursor is the AI-native code editor that became the category default — a VS Code fork rebuilt so the agent, not the keyboard, is the primary way you change code. Two things make it worth re-opening in mid-2026: it now ships its own frontier coding model, and its parent Anysphere is being acquired by SpaceX for $60B, pulling the dominant editor into the same house as Grok. This page is the orient-and-decide surface — cursor.com and its docs own the current plans, models, and limits, which move fast.
What it is
A desktop code editor forked from VS Code (so your extensions, keybindings, and themes mostly carry over) with AI wired into the core loop rather than bolted on: inline autocomplete that predicts multi-line edits, a chat that can read and edit across the whole repo, and an agent mode that plans a change, edits multiple files, and runs terminal commands under your review. As of the Cursor 3 line (2026) it added a dedicated agents window for managing background coding tasks and multi-repo work.
The model story is the update that matters. Cursor now ships Composer, Anysphere's own in-house frontier coding model, alongside the option to route to the major API models (Claude, GPT, Gemini, Grok). So it's no longer just a nice front end over someone else's model — Anysphere trains the brain too, which is exactly the capability the SpaceX deal is meant to supercharge with Colossus-scale compute.
When to use it
Reach for it when:
- You want an AI-native editor, not a terminal agent — a GUI where autocomplete, chat, and agent runs live inside the file you're looking at. This is the opposite ergonomic pole from Claude Code or Aider.
- You're coming back to it after years (as many are) and want to judge the current thing — Composer, the agents window, multi-repo — not the 2024 autocomplete tool you remember.
- You work in a team standardizing on one AI editor — Cursor is where over half the Fortune 500 landed, so it's the low-friction default for onboarding and shared workflows.
- You want one surface that routes to any model — try Composer, Claude, and Grok on the same task without changing tools.
Skip it when:
- Your workflow is terminal-first — if you live in a shell and drive an agent from the command line, a GUI editor is a context switch, not a home (see Claude Code).
- You want $0 or fully local — Cursor's real power is on paid tiers against cloud models; the free/local answer is Aider on Ollama.
- The code is sensitive — it's a closed cloud product; your code and context go to Anysphere/SpaceX servers by design.
- Vendor concentration worries you — post-acquisition, adopting Cursor means leaning on the same SpaceXAI stack that owns the model and the terminal agent (see Risks).
At a glance
The product
| Dimension | Cursor |
|---|---|
| Form | Desktop editor — VS Code fork, GUI-first |
| Own model | Composer, Anysphere's in-house frontier coding model |
| Also routes to | Claude, GPT, Gemini, Grok (bring the model you trust) |
| Agent | Agent mode (multi-file edits + terminal) + a background agents window |
| Scope | Multi-repo workspaces as of the Cursor 3 line |
| Pricing | Free Hobby tier; paid Pro / higher tiers (usage-based) — see cursor.com |
The acquisition (why it's back in the conversation)
- SpaceX agreed in June 2026 to acquire Anysphere — Cursor's parent — in a deal reported around $60B, expected to close in Q3 2026. It is agreed and pending close, not yet completed.
- xAI merged into SpaceX in February 2026, so this folds Cursor into the same SpaceXAI entity that builds Grok and Grok Build. "xAI bought Cursor" is right in spirit; the precise buyer is the merged SpaceX/xAI.
- The strategic logic: Cursor had the distribution and the revenue but was compute-constrained training its own models; SpaceX/xAI brings the Colossus supercluster. Model + agent + editor, one owner.
Exact plan names, per-seat prices, model options, and usage limits live on cursor.com and the docs — canonical over anything here. Acquisition terms are as reported by press (TechCrunch, Forbes, Qz) as of July 2026 and subject to close.
Where it fits on the GL stack
Framed as a re-evaluation — watching, not adopted.
- It's the GUI pole of the coding-agents cluster. The working setup here is terminal-first (Claude Code on a flat Max subscription). Cursor is the other ergonomic model entirely — the agent inside the editor. The honest question isn't "is Cursor good" (it plainly is); it's "does a GUI editor beat a terminal agent for how you actually work."
- The hype is real but partly the deal. Cursor's revenue and adoption are genuinely dominant, and Composer is a real model, not a wrapper. Some of the current noise is also the $60B acquisition — separate the product (worth trying) from the headline (a business event).
- The acquisition cuts both ways. Upside: a Cursor with Colossus-scale compute behind Composer could pull further ahead. Downside: adopting it now means betting on the same SpaceXAI stack across model, agent, and editor — convenient, and concentrated. That's the thing to weigh, not the feature list.
Gotchas
- The acquisition hasn't closed. It's agreed and expected to close Q3 2026 — roadmap, pricing, and model strategy under new ownership are forecasts, not facts. Don't make a durable bet on the post-deal Cursor yet.
- Usage-based pricing bites. The paid tiers meter model usage; a heavy agent workflow can run up a bill in a way a flat subscription (Claude Max) does not. Watch the meter, not just the sticker price.
- It's a fork you don't control. VS Code compatibility is high but not total — some extensions and update cadences diverge, and you're on Anysphere's fork, not upstream.
Risks
- Data leaves the machine. Cursor is a closed cloud editor — your code and repo context go to its servers to power the agent. For a local-first posture, that's the headline risk, not a footnote.
- Vendor concentration under SpaceXAI. Post-close, Cursor, Grok, and Grok Build share one owner. Standardizing on Cursor is a bet on that single stack's pricing, policy, and direction — a lot of power concentrated in one vendor's hands.
- Post-acquisition drift. Ownership changes change products: pricing, model routing (will third-party models stay first-class, or get nudged toward Composer/Grok?), and priorities can all shift. What you'd be adopting today may not be what it becomes.
Related
- Claude Code — the terminal-first, flat-rate daily driver; Cursor is the GUI-first opposite pole.
- Grok Build — the terminal agent from the same post-acquisition SpaceXAI stack.
- Aider — the open-source, $0, local-Ollama alternative to any paid cloud editor.
- xAI Grok API — the model side of the SpaceXAI stack Cursor is being folded into.
