cursor review: the smoothest on-ramp, mind the meter

cursor's pitch has never changed: it's vs code with the AI built into the walls. you open your project and everything you already know works — extensions, themes, keybindings — except now there's a model in every surface: tab-complete, inline edits, and composer, the multi-file agent panel that does the heavy lifting.

the good parts got better

composer 1.5is the star. describe a change that touches six files and it makes the six edits, summarizes its own long-context reasoning, and shows you a reviewable diff. for the describe-and-steer workflow we care about here, it's the smoothest implementation in any editor.

background agentsare the quiet killer feature: fire off a task — “add tests for the parser” — and keep working while it runs, like a junior dev in a parallel universe. bugbot scans for problems before they ship. none of this is a gimmick; all of it produced real work in our use.

and the hobby tier is actually free— enough to learn the workflow and decide if you're in. that matters for the students and designers this scene keeps promising to welcome.

the meter

now the part you need to read twice. since the mid-2025 overhaul, paid plans come with a monthly credit pool roughly equal to what you pay: auto mode(cursor picks the model) is unlimited, but manually choosing a frontier model draws down your pool — on the $20 pro plan that's roughly 225 claude-class requests a month. a heavy build week can drain it, at which point you're choosing between auto mode, overage billing, or the $60 pro+ / $200 ultra tiers.

cursor learned this lesson publicly — the june 2025 rollout was confusing enough to force an apology and refunds. the system is better-documented now, and auto mode is genuinely good for everyday edits. but “unlimited, except for the models you actually wanted” is a sentence you should fully understand before you budget around it.

who it's for

if you live in an editor and want AI woven into it, cursor remains the default answer — and at $20 it's the cheapest way to feel how different building is now. if your work has shifted from “editing files” to “directing agents,” the ceiling is higher elsewhere — see our claude code review for that conversation.

// scored 8.4 — verdict box has the receipts