Cursor · by Anysphere
Implementation & fluency
The EditorThat BuildsWith You
Cursor isn't autocomplete with a new coat of paint. It's a spectrum of autonomy—from a keystroke ahead of you to a fleet of agents shipping in parallel. Most developers live on the bottom rung. We teach you to climb.
The Power
One editor, five gears
Cursor scales with how much you trust it. Knowing which gear a task wants—and when to shift up— is the whole skill.
Tab
Gear 1Project-aware autocomplete that predicts your next multi-line edit once the codebase is indexed. The baseline almost everyone uses.
Inline edit (Cmd-K)
Gear 2Select code, describe the change in plain English, and Cursor rewrites it in place—no chat window, no copy-paste.
Composer
Gear 3Describe a change and it touches every relevant file at once—new files, imports, tests, config—then hands you a unified diff to step through before anything applies.
Agent
Gear 4Hand it a goal and it reads, writes, runs terminal commands, and iterates on its own results until the task is done.
Parallel agents
Gear 5Fan a task across multiple agents on isolated git worktrees—several branches advancing at once while you orchestrate and review.
↑ more autonomy · more leverage · more judgment required
Under the Hood
The parts that make it fly
Rules files
Encode your conventions in .cursor/rules so every suggestion respects your architecture instead of fighting it.
MCP servers
Connect Cursor to your real tools and data through the Model Context Protocol—the same setup across editor and CLI.
Codebase indexing
Local embeddings make the whole repo context, so answers and edits actually understand your code.
CLI & background agents
Kick off agent work from the terminal or the cloud and keep moving while it runs.
Model choice
In-house low-latency models for fast edits, frontier models for hard reasoning—pick the right one per task.
Reviewable diffs
Every agentic change lands as a diff you approve. Speed without surrendering the review.
Where Thinking Backward Comes In
This is about your team.
We make them faster at it.
There's a real difference between code an agent produced and code your team understands well enough to own. We focus on the second.
We coach your engineers to drive Cursor well—when to let an agent run, how to keep diffs reviewable, where the rules and guardrails go. You ship faster and stay in control of what shipped. Capability that stays with you.
Hands-on, with you
Your engineers build the muscle with us, hands-on, in the room.
Reviewable by default
Velocity that keeps diffs small and human-checked, so quality holds.
Outcome first
We start from the delivery goal and work backward to the setup.
Workflows We Build
Habits your team keeps shipping with
We start from the outcome—faster delivery without losing the plot—and work backward to the setup that gets you there.
A rules + guardrails setup
We codify your team's conventions into rules files so Cursor produces code that fits your codebase from the first suggestion.
An agent playbook
Which tasks to hand to Agent, how to scope them, and how to keep the diffs small enough to actually review.
A parallel-agent pattern
For the right work, fan tasks across worktrees—so your team gets real throughput without merge chaos.
A review discipline that scales
The habits that keep AI-written code trustworthy as volume goes up—the part that protects you.