Cursor 3.8 vs Cursor 3.7 vs Claude Code Automations June 2026
Cursor 3.8 vs Cursor 3.7 vs Claude Code Automations June 2026
Cursor shipped two major releases in 48 hours. Cursor 3.7 (June 17) added Cloud Subagents in the Agents Window. Cursor 3.8 (June 18) shipped the /automate skill, Slack triggers, and event-driven GitHub automations. Claude Code remains the best terminal autonomous agent. Here is how the three stack up for event-driven AI engineering work in June 2026.
Last verified: June 20, 2026.
TL;DR
- Cursor 3.8 (June 18, 2026): /automate skill, Slack emoji + @Cursor triggers, expanded GitHub triggers, computer-use for cloud agents.
- Cursor 3.7 (June 17, 2026): Cloud Environment Setup, Cloud Subagents in the Agents Window.
- Claude Code: Best terminal autonomous agent, sub-agents, dynamic workflows, MCP integration, Cowork for async parallel work. Rate limits doubled in May 2026 via the SpaceX Colossus 1 compute deal.
- Event-driven automations: Cursor 3.8 wins for first-class IDE integration; Claude Code wins for autonomous depth.
- For most teams: use both. Cursor 3.8 for GitHub/Slack-triggered work, Claude Code for autonomous IDE/terminal work.
Direct comparison
| Dimension | Cursor 3.8 | Cursor 3.7 | Claude Code |
|---|---|---|---|
| Released | June 18, 2026 | June 17, 2026 | Continuously updated; rate limits doubled May 2026 |
| Primary surface | IDE + cloud agent automations | IDE + cloud subagents | Terminal autonomous agent |
| Event triggers | Slack emoji react, @Cursor mention, GitHub issue/PR/review/workflow events | Limited | Via GitHub Actions or external webhook |
| Plain-English automation setup | /automate skill | No | No (write skill or hook) |
| Cloud agent isolation | Yes — Cursor Cloud Agents | Yes — Cloud Subagents | Cowork sandbox for parallel work |
| Computer-use | Yes for cloud agents (demo/artifact generation) | No | No (terminal-native) |
| Sub-agents | Yes (cloud subagents) | Yes (cloud subagents) | Yes (Claude Code sub-agents, MCP, dynamic workflows) |
| MCP integration | Limited | Limited | Native, mature |
| Marketplace | Cursor Marketplace with automation templates | Cursor Marketplace | Skills + community hooks |
| Best for | Event-driven team workflows | Cloud-based dev environments | Autonomous engineering work |
| Pricing | $20/mo Pro, $40/mo Ultra | Same | Claude Pro $20/mo, Max $100/mo, doubled rate limits |
What Cursor 3.8 actually changes
Cursor 3.8 makes event-driven AI engineering a first-class IDE feature. Before 3.8, you stitched GitHub Actions + Claude Code + custom scripts to get “when X happens in PR, do Y.” 3.8 collapses that into:
/automateinside a local Cursor agent session. Describe the workflow in plain English. Cursor configures triggers, instructions, and tools.- Slack triggers:
- Emoji react — react to any Slack message with a designated emoji to start an automation.
- @Cursor mention — mention @Cursor in a channel or thread; a background agent reads the conversation, applies fixes to a repo, and opens a PR.
- Slack messaging — automations can post to Slack channels and read them.
- GitHub triggers (expanded):
- Issue comments (non-PR)
- PR review comments (inline)
- PR review submitted
- Review thread resolved/unresolved
- Workflow run completed
- Computer-use for cloud agents. Generate demos, screenshots, and artifacts of completed work.
- Cursor Marketplace. Templates like “triage failed GitHub Actions,” “auto-fix PR review comments,” “weekly dependency PRs.”
The strategic move: Cursor is becoming the orchestration plane for team AI engineering, not just the editor. The Agents Window from Cursor 3 was the foundation. Bugbot was the first cloud agent product. /automate is the workflow surface that ties it together.
What Cursor 3.7 changes
Cursor 3.7 (June 17, 2026) shipped the day before 3.8 and focused on cloud-side infrastructure:
- Cloud Environment Setup. Configure cloud agents’ environments declaratively — Node version, system deps, env vars.
- Cloud Subagents. Spawn parallel cloud agents from within the Agents Window — useful for “fan out three approaches and let me pick.”
3.7 set up the infrastructure that 3.8 makes useful. The two-day release cadence is interesting: it suggests Cursor’s release engineering is now mature enough to ship rapid increments.
How Claude Code compares
Claude Code’s posture is different — and complementary.
- Terminal-native autonomous agent. Best for “open a terminal, ask Claude to ship a feature, walk away” workflows.
- Sub-agents. Mature parallel-work primitive.
- Dynamic Workflows (research preview from May 2026). Codebase-scale migrations.
- MCP integration. Native and mature; Claude Code is the reference implementation of MCP in coding agents.
- Cowork. Async parallel work in sandboxed VMs. Good for “Claude, refactor while I do something else.”
- Rate limits doubled in May 2026 via Anthropic’s deal with SpaceX for the Colossus 1 data center (220,000+ NVIDIA GPUs). 5-hour limits doubled on Pro/Max/Team/seat-based Enterprise. Peak-hour throttling removed.
- Claude Opus 4.8 (May 28, 2026) added Dynamic Workflows and Effort Control.
Where Claude Code wins over Cursor 3.8 for event-driven work: depth of autonomous behavior, MCP ecosystem, terminal-native composition. Where Cursor 3.8 wins: out-of-the-box Slack and GitHub event integration, IDE-native automation setup.
When each tool wins
Use Cursor 3.8 when
- You want GitHub event → AI work → PR as a first-class team workflow.
- Your team coordinates in Slack and wants AI work triggered from Slack messages.
- You want the /automate plain-English setup rather than writing scripts.
- You need computer-use artifacts (demos, screenshots) from cloud agents.
Use Cursor 3.7 cloud subagents when
- You want parallel agent approaches to a problem in the IDE.
- You want declarative cloud environment configuration for AI agents.
Use Claude Code when
- You want autonomous terminal-native engineering work.
- You need MCP ecosystem depth (filesystem, browser, search, custom tool integrations).
- You’re running codebase-scale migrations (Dynamic Workflows).
- You want Cowork for async parallel work in sandboxed VMs.
- You have Claude Max and benefit from the doubled rate limits.
Use both when
- You’re a team doing serious AI-assisted engineering in 2026. The realistic posture is Cursor 3.8 for the IDE and event-driven workflows, Claude Code for autonomous work and Anthropic-specific capabilities (sub-agents, MCP, Cowork).
Pricing snapshot June 2026
| Plan | Cursor | Claude |
|---|---|---|
| Free | Free tier | Limited |
| Pro | $20/mo | $20/mo |
| Pro+ / Max | $40/mo (Ultra) | $100/mo (Max), $200/mo (Max 20x) |
| Team | $20/seat/mo | $30/seat/mo |
| Enterprise | Custom | Custom |
What to expect next
- Cursor 3.9 / 4.0: Expect deeper Slack and Linear integration, more marketplace templates, possibly a Cursor + Claude Code interop story.
- Claude Code: Continued MCP expansion, more Cowork polish, likely Claude Fable 5 integration ahead of GA (currently Claude Opus 4.8 / Claude Fable 5 preview).
- Convergence: Both tools are converging on “AI engineering orchestration.” Expect more overlap and more user-side choice.
Sources
- Cursor changelog 06/18/26: Cursor 3.8 release notes
- Cursor changelog 06/17/26: Cursor 3.7 release notes
- Kingy.ai: “Cursor /automate skill AI coding agents”
- DigitalApplied: “Cursor Automations always-on agentic coding agents guide”
- Anthropic news: Higher limits via SpaceX (May 2026)
- Anthropic Code with Claude 2026 conference materials (SF May 6, London May 19, Tokyo June 10)
- Mindstudio.ai: “Claude Code rate limits doubled, Colossus 1 API limits”
Published June 20, 2026 by andrew.ooo. See related: Cursor 3.7 vs Cursor 4 vs Windsurf vs Claude Code and Claude Code rate limit doubling.