Skip to main content

Buyer guide

Cursor Cloud Agents in Your Review Workflow

Choose cloud-agent work, confirm build and test access, connect Bugbot to repository rules, and name the merge approver.

Reading time
3 min read
Sources
12 cited
In this piece
Cursor+11 Mixed

The call

Delegate well-specified tickets with runnable tests to cloud agents, add Bugbot with BUGBOT.md rules as the automated review, and save a named human reviewer for the final approval on every merge.

Cloud agents and Bugbot produce pull requests and review comments faster than people can read them, so the human reviewer is the step to protect. A company running Cursor on a large monorepo puts that reviewer last, after the developer and Bugbot. This guide maps which work to delegate, how Bugbot fits, and who approves.

When you need this

  • Engineers want to hand Linear or Slack tickets straight to an agent.
  • Pull requests arrive faster than reviewers can check them.
  • You are weighing Bugbot against another AI reviewer.

Worked example: one company's order of review on a large monorepo

  1. Plan

    The developer starts in Plan Mode and has the agent implement against tests.

  2. First review

    The developer reviews the diff in the IDE before anything else sees it.

  3. Automated review

    Bugbot reviews each pull request update against BUGBOT.md. Public reports range from about ten real bugs caught in a month for a small team to bugs surfacing only across several review cycles.

  4. Final approval

    A second engineer approves last, because review time limits how much the team ships.

Evaluation criteria

  1. Which work to delegate

    What should a cloud agent take on without a human in the loop?

    Look forDelegate simple or well-specified tasks with tests the agent can run. Engineers on Hacker News describe cloud agents handling simpler Linear tickets and investigating support issues or alarms, and they keep complex work in the editor.

  2. Environment and access

    Can the agent build and test the change it makes?

    Look forCloud agents run in isolated VMs with cloned repositories, installed dependencies, secrets, and network access. They support MCP servers and can work across several repositories in one task. An admin first connects GitHub, GitLab, Bitbucket Cloud, or Azure DevOps. Internet access is on by default, so restrict it to an allowlist, and mark secrets as Runtime Secrets to keep their values out of transcripts, commits, and the model.

  3. Bugbot rules and effort

    What does Bugbot review, and by whose standards?

    Look forBugbot reviews each pull request update, or runs when someone comments bugbot run, and reads the .cursor/BUGBOT.md files committed to the repository. Effort levels exist on usage-based Bugbot plans, and higher effort finds more bugs at more usage. Cursor bills each review once, to the account that turned Bugbot on for the repository, and autofix draws Cloud Agent credits at plan rates. Findings default to a neutral check status, so Bugbot blocks a merge only after you configure its check to fail on unresolved issues.

  4. Review order

    Who reviews what, and when?

    Look forOrder review from cheapest to scarcest. The developer reads the diff in the IDE first, and Agent Review checks local changes after each commit against the same BUGBOT.md rules. Bugbot reviews the pull request next, and the second engineer approves last.

  5. Merge approval

    Who owns the merge for an agent-authored pull request?

    Look forName the engineer who approves each merge. Non-engineers can start changes through web agents when engineering reviews before merge, as one company does for its marketing site.

  6. Delegation and review map

    Is it written down which work goes where?

    Look forFor each task type, record the trigger and whether it runs in the editor, CLI, or cloud. List the required checks and whether Bugbot reviews it. Name the human approver, and revisit the assignments when review time per change rises.

Questions to ask vendors

  1. Which repositories and integrations can start cloud agents, and who can trigger them?
  2. Which account has Bugbot turned on for each repository? Cursor bills every review there, whichever team the pull request author belongs to.
  3. Which Bugbot findings should fail the check and block a merge, and who owns that configuration?
  4. Who limits cloud agents' default internet access to an allowlist, and which secrets do they receive?

Red flags

  • Agent pull requests merged without a named human approver.
  • Bugbot comments treated as the review itself.
  • Cloud agents enabled before source control access and secrets are scoped.
  • Review load measured by pull request count alone.

Worth evaluating

Listed in no particular order.

Sources

Official documentation
6
Public source
6
  1. Cursor docs: Cloud AgentsOfficial documentationcursor.com
  2. Cursor docs: Cloud Agents security overviewOfficial documentationcursor.com
  3. Cursor docs: BugbotOfficial documentationcursor.com
  4. Cursor docs: Agent ReviewOfficial documentationcursor.com
  5. Cursor docs: Plan ModeOfficial documentationcursor.com
  6. Cursor pricingOfficial documentationcursor.com
  7. Peter Eysermans, madewithlove: automatic pull request reviewing with Bugbot (Sep 2, 2025)Public sourcemadewithlove.com
  8. Cursor forum: Bugbot needing several review cycles (Feb 10, 2026)Public sourceforum.cursor.com
  9. Cursor forum: layered review on a large monorepo (Feb 18, 2026)Public sourceforum.cursor.com
  10. Hacker News: delegating simpler Linear tasks to cloud agents (Apr 2, 2026)Public sourcenews.ycombinator.com
  11. Hacker News: cloud agents for automated investigations (Aug 29, 2026)Public sourcenews.ycombinator.com
  12. Hacker News: a marketing team using web agents with engineering review (May 5, 2026)Public sourcenews.ycombinator.com

Keep reading

  1. Buyer guideAmp Team Rollout and Spend ChecklistWhat to settle before engineers move from personal Amp accounts to a shared workspace, and how to grade the pilot when it ends.
  2. Buyer guideHanding a Bolt App to EngineeringHow GitHub sync, branches, the Supabase database claim, Netlify, and Expo work, where each stops, and the setup an engineer or a coding agent needs to take the code over.
  3. Buyer guideCodeRabbit in Your Coding Agents' LoopHow to wire the CodeRabbit CLI into Claude Code, Codex, Cursor, or Gemini, for teams whose agents open pull requests faster than reviewers read them.
Read the research