brain/
← all entities
entitygenericartificial-intelligence

Cursor

Notes

Cursor

One-line summary: A VS Code fork turned into an "AI-native" IDE. The dominant IDE-centric AI coding tool for individual developers in 2026; fastest-growing SaaS by a wide margin until recently.

What it is

An AI-first IDE forked from Visual Studio Code. Headline capabilities: Supermaven-powered autocomplete, multi-model support, and Composer mode for multi-file editing with codebase indexing (2026-04-21-autoresearch-best-ai-coding-tools).

Why it matters to this thread

Cursor defines the IDE-centric tier of the AI coding tool landscape (see ai-coding-tool-landscape-2026). It is the default recommendation for "daily IDE-based development" in 2026 and the most common pairing with terminal agents like claude-code in two-tool stacks.

Key facts (from 2026-04-21-autoresearch-best-ai-coding-tools)

  • Revenue: Reported by Amplifi Labs as reaching $1B annual recurring revenue "faster than any SaaS product in history in under 24 months." (See ai-coding-tool-landscape-2026 for the contested ARR comparison with Claude Code.)
  • Customers: 360K paying customers (MorphLLM's agent round-up).
  • Pricing: $20/mo Pro; $200/mo Ultra; usage-based overages.
  • Adoption (JetBrains April 2026 survey, 10K+ devs): 18% developer adoption, 69% awareness — survey authors note Cursor's growth has stalled.
  • Task completion rate: 71% reported by MindStudio (benchmark context not specified in source).
  • Practitioner task fit: "Best overall AI code editor in 2026" for multi-file editing, daily routine coding, mid-to-senior developers; handles "80% of typical development work" via Composer + Agent mode.
  • Notable in METR context: The 2025 METR RCT (ai-coding-productivity-paradox) used Cursor Pro with Claude 3.5/3.7 Sonnet as the AI stack — i.e. the "19% slower" finding was largely a finding about Cursor specifically.

Strengths (from the synthesis perspective)

  • Multi-file understanding and refactoring; repository indexing.
  • Clean fit inside existing VS Code workflows (it's a fork).
  • Supports model choice across providers — not locked to one LLM vendor.

Weaknesses / concerns

  • Growth stalled (JetBrains April 2026).
  • Occasional "subtle bugs" per practitioner reviews.
  • The April 2026 synthesis this wiki first used is secondhand on pricing/revenue/adoption. Official Cloud Agent docs (below) are first-party for the unattended-run control surface.

Cloud Agents as isolated inner runs (from 2026-08-18-autoresearch-effective-all-day-bot-loops)

Cursor Cloud Agents run the same agent fundamentals in isolated cloud VMs (clone, deps, secrets, network). Official best-practices put environment setup first, then secrets/OIDC, egress allowlists, agents.md + skills for a "smart but low-context" developer, and tools shaped so the model can actually use them. That is the inner reliability story (can this one PR close?). All-day coverage still needs an outer trigger — a human, a cron, a GitHub event, or another bot — to start those isolated runs. Secrets via OIDC, not long-lived keys.

This is the same outer-loop shape as Claude Code Routines and openclaw heartbeat: wake → isolated workspace → git/PR → exit. It does not by itself provide a research-queue drain; the queue still has to live in git or an issue tracker. See unattended-all-day-agent-loop.

Skill discovery and Paper (from 2026-08-23-session-cursor-paper-grok-app-builder-workflow)

Vintage 2026-08-23. First-party session on Ion Wake — one environment, not a market update.

  • Local Cursor (this Mac) sees the checkout, paper-desktop if the app is open, ~/.cursor/rules and ~/.cursor/skills. AGENTS.md is always applied. .grok/skills are not on the available-skills list.
  • This session’s available_skills: Cursor built-ins, plugins (Paper, Vercel, …), ~/.cursor/skills, ~/.claude/skills, ~/.agents/skills. Not .grok/skills.
  • Cloud Agents clone to /workspace. They see repo .cursor/rules and .cursor/skills, plus home skills on the VM. Home rules (~/.cursor/rules) are not walked — discovery starts at /workspace and never reaches /home/ubuntu. Account User Rules in Cursor Settings are documented as applying across cloud sessions.
  • Do not collapse Cursor, Cloud Agents, grok-app-builder, and grok-bot into one “Grok on Cursor” blob. See agent-skill-discovery.

Sources

Related

Referenced by
Mechanisms
Grok-only skill path → AGENTS.md does not inject → repo `.cursor/skills` symlink → Cloud / Grok Bot load the packImmortal-session decay → durable queue + isolated wakes → reviewable artifacts → unattended all-day progressPaper as visual source of truth → Paper-blind agent invents layout → tight-delta session that can see Paper
Concepts
Agent skill discovery pathsAI Coding Productivity ParadoxAI Coding Tool Landscape (2026)AI Coding Tool StackingCLAUDE.md as team knowledge baseLLMs are a commodity; the moat is proprietary data + workflow integrationPaper as source of truthUnattended all-day agent loopVibe Coding / Greenfield App BuildersVibe Coding Security
Questions
what is the best way to build an effective loop to keep your bot working auto…
Entities
AnthropicClaude CodeGitHub CopilotGoogle AntigravityGrok App BuilderGrok BotOpenClawPaper DesktopWindsurf
Sources
Session: Cursor + Paper + Grok App Builder workflow (Ion Wake, 2026-08-23)