← All Cursor issues

This week in Cursor · Jun 15, 2026

Bugbot Cuts Review Time to 90 Seconds and Costs 22% Less; Auto-review Ships as Default for All New Users

Get the next issue in your inbox. Weekly · Free

unsubscribe anytime.

TLDR;

Two meaningful product improvements landed for Cursor users this week. Bugbot (June 10) got across-the-board better: reviews complete in 90 seconds on average (3x faster), cost 22% less per run, find 10% more bugs, and a new /review command makes on-demand analysis available anytime; billing also shifted to usage-based. Auto-review (June 11) became the default for all new users — a contextual classifier subagent running at 97% accuracy decides which agent actions need human approval before execution, governing autonomy without slowing down well-understood tasks. Together these changes mean Cursor's agentic mode is both faster to onboard and safer to run unattended. A side note: Cursor added Claude Fable 5 this week and posted a new CursorBench SOTA of 72.9%, but US export controls suspended Fable 5 on June 13 — verify model availability before committing workflows to it.

Developer Tools & Community

Cursor Bugbot: 3x faster (90s avg), 22% cheaper, 10% more bugs found, new /review command, usage-based billing

  • Released June 10, Bugbot received a comprehensive upgrade: average review time dropped from ~4.5 minutes to 90 seconds; cost per review fell 22%; the underlying model finds 10% more bugs per PR.
  • A new /review command triggers Bugbot on demand from any open PR without waiting for a push event.
  • Billing moved to usage-based pricing — you pay per review rather than a flat seat cost.

Editor’s read: 3x faster and 22% cheaper in one release is rare — if Bugbot was too slow to be in your PR workflow before, re-evaluate it now.

Auto-review now default for all new Cursor users — 97% accurate classifier subagent governs agent autonomy

  • On June 11, Cursor made Auto-review the default setting for all new accounts.
  • A contextual classifier subagent runs before each agent action and decides with 97% reported accuracy whether the action requires human approval.
  • The classifier learns from the type of task and prior session behavior to minimize interruptions on routine actions while catching potentially destructive ones.

Editor’s read: 97% accuracy is a strong claim — the actual value is making agentic mode safe for non-experts to leave running. Worth enabling if you have not already.

API & Model Updates

Claude Fable 5 added to Cursor; CursorBench reaches new SOTA at 72.9% — but Fable 5 now suspended

  • On June 9, Cursor confirmed Claude Fable 5 is available in the model picker and announced a new CursorBench SOTA score of 72.9% on their internal coding benchmark.
  • However, the US government export control order issued June 12-13 forced Anthropic to suspend Fable 5 globally.
  • Verify current model availability in Settings before relying on Fable 5 for any workflows.

Editor’s read: CursorBench SOTA is a useful competitive signal — but it is moot until Fable 5 is reinstated. Keep Sonnet 4.6 as your active model for now.

Indie & Solo Builders

Action Items

Immediate

  • Run /review on your oldest open PR to try Bugbot's faster 90-second analysis — no config change needed.
  • If you are an existing Cursor user, check Settings to confirm Auto-review is enabled (it is default only for new accounts).
  • Verify your active model in Cursor Settings — Fable 5 is currently suspended; fall back to Sonnet 4.6 or another available model.

By next sprint

  • Switch Bugbot to usage-based billing if you are on a flat plan — review the new pricing against your PR volume.
  • Read the Auto-review blog post to understand how the classifier makes approval decisions and tune the sensitivity for your team's risk tolerance.

All Resources

Every Monday morning

This week in Cursor

Five minutes, every Monday — the tools, releases and tactics for shipping solo.

unsubscribe anytime.

This week for builders

Five minutes, every Monday — the tools, releases and tactics for shipping solo.

unsubscribe anytime.