GatwAI

Cursor

Cursor for code changes, agents and reviews: current pricing, Privacy Mode, intended users and independently researched productivity evidence.

Provider sources checked

What does Cursor do?

Cursor combines a development environment with agents that inspect existing code, modify files and work on engineering tasks. The current platform goes beyond typing suggestions: desktop, command-line and cloud agents, together with code review, form part of the offering. It is particularly relevant when work already takes place inside a repository. Evaluate it on an actual maintenance task: does the agent locate the right code, keep changes understandable and reproduce the defect? A quick draft offers a different benefit from a dependable change ready for release. GatwAI recommends including review time and rework in a pilot comparison.
Editorial research · 2026-11-14How we research →

Who is this tool for?

  • Developers regularly modifying existing projects.
  • Teams with Git review and a measurable pilot task.

Strengths

  • Agent work and manual coding can coexist in one development workflow.
  • Cloud tasks and reviews extend the offering to parallel work.
  • An optional Privacy Mode is documented even for free accounts.

Limitations and requirements

  • Repository context cannot replace correct requirements or meaningful tests.
  • Available models and additional usage charges differ substantially between plans.
  • Agents with terminal or network access need scoped permissions and reviewable changes.

Who is behind the product?

Company
Anysphere, Inc.
Headquarters
Not clearly dated by the primary sources reviewed.
Company founded
Not clearly dated by the primary sources reviewed.
Product launch
Not clearly dated by the primary sources reviewed.

The current privacy page identifies Anysphere as the provider. Company formation and original product launch were not separately established here. Current pricing connects Cursor with a shared Grok allowance; this does not establish an otherwise unverified corporate acquisition.

Suggested ways to use it

Suggested workflows or verification steps. These are not hands-on tests; for unverified products they are not a recommendation to use the service.

Suggested: fix a bug with a regression test

  1. Reproduce the failure and identify the relevant project area.
  2. Request a scoped patch and an appropriate regression test.
  3. Review the diff, what the test proves and effects on adjacent behavior.

Goal: A traceable change proposal rather than an unreviewed production patch.

Suggested: prepare separate maintenance tasks

  1. Define two independent tasks with separate acceptance criteria.
  2. Let agents work separately and review each result.
  3. Handle conflicts and integration checks before merging.

Goal: Comparable proposals with visible review effort.

Pricing, allowances and data

What should you check before choosing a plan?

At review: limited Free access; Plus at USD 8/month with Grok models only; Premium at USD 30/month with cloud agents and reviews. Third-party models incur separate usage charges according to the pricing page. Super, Ultra and organizational tiers provide larger allowances. Applicable taxes may be additional. Compare the subscription and extra consumption together.

What should you check about data and permissions?

According to the security page, enabling Privacy Mode prevents Cursor training on your data, with technical and contractual controls for model providers. It does not mean processing is entirely local. Teams should review model selection, repository access, retention and cloud agents separately. The privacy policy was updated on 6 November 2026.

External reviews and available evidence

No GatwAI hands-on test. METR studied experienced open-source developers in 2025, primarily using Cursor with contemporary Claude models. Its measured 19% slowdown is not an assessment of the current version.

Reported positives

  • The study examines real work rather than provider benchmarks alone.

Reported cautions

  • METR's 24 February 2026 follow-up describes selection effects and does not establish a reliable current speedup.

These reports reflect the linked authors’ experiences. Test date, plan and task affect results; they do not establish a representative overall rating.

Which alternative fits which task?

Frequently asked questions

Does every plan support every model?

No. Current Plus pricing specifies Grok only; third-party models on higher tiers incur separate charges.

Does METR prove Cursor always slows developers down?

No. Its result concerns particular developers, tasks and early-2025 models; the authors explicitly caution against generalization.