Cursor

Cursor is the tool I recommend to people who live in an editor: Tab handles the boring half of coding, and the Agent handles medium refactors well enough that you stop opening a terminal for them. That is the daily trait: the editor does more, so you context-switch less. The pain is the bill; Pro is $20, then Pro+ and Ultra buy 3x and 20x the limits.

Pick it if you want the best editor experience today and can live with a subscription that grows. Pick Claude Code or Aider to see the bill and stay in the terminal.

The Agent's sandbox comes with the run mode: Auto-review sandboxes commands through Seatbelt or Landlock, and Run Everything hands them your machine, the checkout you were editing, and your credentials. The pricing page is the second failure mode: the tiers multiply Agent limits by factors the page never defines, so you cannot know what a limit is worth until you have spent it, and the upgrade prompt arrives at the worst moment.

Keep auto-run off outside a worktree and set a budget before the first month. What it does right is Tab and multi-file edit, still the reference the others copy, and the reason the bill gets paid.

Cursor's agent design is documented, not measured. 1. Parallel subagents run in git worktrees, a sound isolation choice; each attempt gets a clean tree. 2. Verification is the user's job unless the agent chooses to run tests, so the loop closes only when the person closes it. No benchmark appears in the dataset, so the agentic gains are claimed, and the editor's reputation is doing the work the numbers should.

The absence of numbers from the best-funded editor on this board is the observation. What would change it is one published run on a public harness, which the company could afford tomorrow.

Anysphere is the revenue story of this category, and a tier priced per user for the engineering manager tells you the buyer is no longer a developer with a card. The moat is distribution and habit: an editor is where a developer lives. The risk is the supply chain: Claude, GPT, Gemini and Grok are all sold by companies shipping their own agents, so every frontier request is bought from a competitor who can change the price.

The in-house model is the hedge, and the hedge is the whole story. Position: long the product, watch the model bill and the day a lab cuts it off.

The demo is an editor that fixes a bug while the engineer watches. Teams is $40 per user with SAML SSO, so sixty seats is $28,800 a year before Cloud Agents bill on top at API rates, which is the line nobody budgets. SCIM, audit logs and auto-run controls sit on Enterprise at custom pricing, and auto-run controls are the item security will insist on, so Enterprise is the real tier. Onboarding is trivial; half the engineers already use it.

Approved with conditions: Enterprise pricing in writing, a monthly Cloud Agent spend cap, and auto-run policy set centrally.

Cursor is VS Code with the source taken away. mcp.json takes my servers, and rules, skills and hooks are text files I can version, which is a real configuration surface. That is the ceiling: local_models is false, and the agent runs on their prompts, which I cannot read or change, so when it does something odd I can only guess why. Nothing forks; the upstream editor does, but not this.

Grudging respect for Tab, which nobody has matched in the open. It is the best tool I do not own, and I keep noticing that the ones I own are slower.