Cursor for New Zealand businesses.

What Cursor is now, and what it is good at

Cursor is a code editor with an AI agent built in: you describe a change, and the agent reads the repository, edits several files and runs commands while you watch the diff. In 2026 it also changed hands. SpaceX completed its acquisition on 14 August 2026, subscriptions are now sold by X Subscription Corp, and every plan bundles Grok and X (Twitter) subscription features alongside the editor.

The product itself remains strong for teams that write code every day:

• Editor-first work. Developers stay in one window while the agent edits across files, which suits frequent small changes better than handing off whole tasks.
• Shared project rules. Rules live in the repository, so a team’s conventions travel with the code.
• Team controls on paid business plans. Teams adds centralised billing, usage analytics, team-wide privacy settings and SAML/OIDC SSO; Enterprise adds SCIM, audit logs and repository, model and MCP access controls.
• Ecosystem support. Shopify’s AI Toolkit ships plugins and agent skills for Cursor plus a local Dev MCP server, useful for NZ retailers building on Shopify.

Cursor’s security page lists SOC 2 Type II, ISO/IEC 27001:2022, ISO/IEC 42001:2023 and AIUC-1. Where it is weaker for New Zealand buyers: unpredictable bills once you use OpenAI or Anthropic models, card-only USD billing below Enterprise, social-media features nobody in your business asked for, and a vendor whose terms have just changed.

Cursor plans and prices

Cursor plans as listed on 19 October 2026. Prices are in US dollars and, in Cursor’s words, “exclusive of any applicable taxes”. The page does not mention New Zealand GST, so check how tax appears on your first invoice.

Each tier includes a usage pool, shared between Cursor and Grok, that refreshes weekly; top-offs are available when it runs out. Third-party models are never in the pool. Cursor bills them separately at API list price, with optional spending caps.

Illustrative arithmetic: six developers on Teams at the US$30 entry seat cost US$180 a month before tax. If each also uses 3 million input and 300,000 output tokens a month of Claude Sonnet 5.5, and Cursor passes through Anthropic’s list price of US$2 and US$10 per million tokens, that adds US$9 per developer, or US$54. Total: about US$234 a month, plus tax and currency conversion on the card.

How Cursor handles your code

Cursor’s data use page, updated 3 September 2026, sets out two states:

• Privacy Mode on. Cursor does not use your data for training, and it holds zero data retention agreements with its model providers. Prompts that trip abuse classifiers may still be stored for investigation. Models without zero retention are labelled or need an admin to opt in.
• Privacy Mode off. Cursor may store and use codebase data, prompts, editor actions and code snippets to improve its features and train its models.

Two further details matter. Requests go through Cursor’s backend even when you supply your own API key, because that is where prompts are assembled. File contents are cached temporarily, encrypted with client-generated keys. The pages we checked do not say which country processes your requests; Cursor publishes its subprocessor list on its trust portal.

For the Privacy Act 2020: with Privacy Mode off, Cursor uses what it receives for its own purposes, so any personal information in your code (test fixtures, logs, customer exports) is being disclosed to an overseas provider, and IPPs 11 and 12 apply. With Privacy Mode on, Cursor is closer to a provider processing on your behalf under section 11, which is not a disclosure, but your organisation stays responsible for security under IPP 5.

Buying Cursor from New Zealand

Cursor’s pricing FAQ is blunt: subscriptions are sold only directly through X Subscription Corp, and no resellers are authorised. Your managed service provider cannot bundle it. Self-serve plans are paid by card in US dollars; invoices, purchase orders and wire transfers need Enterprise.

Before you commit, ask the questions in our AI procurement questions: where requests are processed, which subprocessors see code, how the ownership change affects existing privacy commitments, and how much notice you get of plan changes. Then set an exit plan. Cursor reads AGENTS.md files as well as its own rules, and so does Claude Code, so keeping instructions in AGENTS.md makes a later switch cheaper. Track who holds which seat; our lesson on managing AI subscriptions and access covers offboarding.

A starter project your team can run this week

Illustrative example: a five-developer Wellington team maintains an internal quoting service with thin test coverage. The goal is better tests without touching production code.

1. Choose the plan. For two or more developers, Teams gives you team-wide privacy settings and SSO. A single developer can trial Premium.
2. Turn on Privacy Mode for the team in the admin settings, then confirm on one developer’s machine that the setting was inherited.
3. Cap third-party spend. Set a spending cap before anyone selects an OpenAI or Anthropic model, and decide which models are approved.
4. Add project rules. Create .cursor/rules/house-rules.mdc (rules must use the .mdc extension) with the content below and commit it.
5. Ask for tests, not changes. Give the agent the prompt below. It should touch only test files.
6. Prove the tests work. Break the function deliberately and confirm the new tests fail. Tests that still pass against broken code are worthless. Our guide to evaluating AI-generated code has more checks.
7. Review usage. After a week, compare the usage figures in the admin dashboard, and any third-party model charges, against the number of tests merged.

---
description: House rules for the quoting service
alwaysApply: true
---
Use New Zealand English in comments, messages and documentation.
Store money as integer cents. GST is 15% unless a line is zero-rated.
Show dates to users as DD/MM/YYYY.
Never read or print values from .env files.
Ask before adding any dependency.

Write unit tests for calculate_quote_total() in quotes/pricing.py.
Cover GST at 15%, zero-rated lines, a discount applied before GST,
rounding to the nearest cent, and an empty quote. Do not change
pricing.py. Run the tests and report which fail and why.

Cursor versus Claude Code

• Choose Cursor if your developers want an AI-first editor with team administration in one subscription, and you are comfortable with direct USD card billing and the new owner’s terms.
• Choose Claude Code if you prefer an agent that takes whole tasks from the terminal beside any editor (including Cursor), with Anthropic’s models included in the plan rather than metered at API list price.
• Choose neither yet if nobody on staff can review code. Compare an AI agency with an internal hire first; if you do have developers, our AI training covers reviewing agent output.

Where Cursor goes wrong

• Bill surprises. OpenAI and Anthropic usage sits outside every allowance. Without caps, a week of heavy agent work can cost more than the seats.
• The wrong plan. Plus includes Grok models only. Staff who expect Claude or GPT models on an $8 plan will hit a wall.
• One developer without Privacy Mode. That is enough for code, and any customer data in it, to become training data.
• Large unreviewed diffs. Ask for small, scoped changes and read every one before merging.
• Plan churn. The 2025 line-up of Hobby, Pro at $20 and Pro+ has gone. Expect further change, and keep rules in files you can take with you.

Written by the Levy Street team in Wellington. Last updated 19 October 2026.