Codingby Anysphere★ 4.3

Cursor

AI-native code editor forked from VS Code with agent mode for multi-file edits, background tasks and codebase-aware refactors.

Try it Also in Tools: AI Code & Dev →

Safety & Trust

Delv Safety Grade: B

Score 68/100 · assessed 2026-04-18

Maintainer75

Permissions40

Supply chain70

Transparency45

Incidents100

Cursor is a closed-source AI code editor from Anysphere, a venture-backed startup. It's a fork of VS Code with deep filesystem and shell access for multi-file edits and background tasks. The maintainer is a legitimate, well-funded company with active development, but the product is proprietary with no public repository. Supply chain is standard desktop app distribution (direct download, auto-updates). Permissions are extensive: full filesystem read/write, shell execution for build tasks, network access for AI models, and environment variable access including potential secrets. Transparency is limited due to closed source. No known security incidents. The agent mode's autonomy over codebases creates meaningful attack surface if credentials or malicious prompts are involved. Suitable for professional use with awareness of broad permissions.

Green flags

Legitimate venture-backed company (Anysphere) with active development
Standard desktop app distribution with signed binaries
No known security incidents or credential leaks
Freemium model with transparent pricing
Built on VS Code foundation with familiar security model

Red flags

Closed source with no public repository for audit
Full filesystem write access across entire codebase
Shell execution permissions for build and task automation
Environment variable access may expose API keys and secrets
Agent mode autonomy could execute unintended multi-file changes

Permissions requested

Read filesWrite filesDelete filesShell executeOutbound networkRead envAccess secretsExternal LLM call

Assessed by Delv Editorial using public metadata. Grades are advisory and update as the ecosystem changes. They do not replace your own review of permissions and code before granting an agent access to sensitive systems.

Pricing

FREEMIUM — Free tier, paid from $20/mo

Platforms

desktop

Review

Cursor is VS Code with agent mode bolted on, and that combination matters more than it sounds. The editor itself is familiar—same keybindings, same extensions, same Git tooling—but Cmd+K opens a chat that can edit multiple files in one go, and Cmd+Shift+K kicks off an agent that runs in the background while you keep working. The autonomy shows up in refactors. I pointed it at a React component tree that needed prop drilling replaced with context, gave it a two-sentence brief, and it touched eleven files without me babysitting each edit. It proposed changes in a diff view, I accepted the batch, done. Compare that to Copilot, where you're still driving file-by-file, or to Windsurf, which tries to be clever about when to act but often guesses wrong. Cursor's agent mode is explicit: you tell it to go, it goes, you review when it's finished. It shines on structural changes—renaming across a codebase, moving logic between modules, updating API calls after a schema change. The codebase indexing is fast enough that it doesn't feel like you're waiting for context to load. The chat references functions and files accurately, and when it makes a mistake, it's usually a logic error you'd catch in review, not a hallucination about what exists. Failure modes: it's still an LLM under the hood, so it can propose subtly broken code if the task is ambiguous. The agent mode doesn't run tests or check types automatically—you're the CI pipeline. And the free tier is stingy: two weeks of the premium model, then you're back to a slower fallback unless you pay. The $20/month Pro plan is reasonable if you're shipping daily, but the jump to $40 for the better model stings. I'd reach for Cursor when I need to make a change that touches more than three files and I don't want to context-switch between editor and chat window. It's not magic—it's just VS Code with a capable agent that stays out of your way until you need it. For solo developers or small teams already in the VS Code ecosystem, it's a straight upgrade. For larger teams, the lack of shared agent memory or team-wide indexing is a gap.

Verdict

Pay for Cursor if you're a solo dev or small team doing frequent multi-file refactors in a VS Code workflow. Skip it if you need deeper agent autonomy (try Devin) or if you're happy with Copilot's per-file assistance and don't want another subscription.

Good at

Agent mode handles multi-file edits in one pass, with clear diff review before applying
Familiar VS Code foundation means zero onboarding for existing users
Fast codebase indexing keeps context-aware suggestions accurate
Background agent lets you keep working while it churns through tasks
Explicit agent invocation—no surprise edits, you control when it acts

Watch out

Free tier drops to slower model after two weeks, pushing you toward paid plans
No automatic test running or type checking—agent can propose broken code
Lacks team-wide indexing or shared agent memory for collaborative workflows
Premium model ($40/mo) is expensive compared to Copilot or standalone LLM subscriptions
Still prone to LLM hallucinations on ambiguous or under-specified tasks

Use cases

pair programming
refactoring
agent coding