2 min lesson
Name the era to locate the org
Use two rows in "Name the era to locate the org" to state the practical decision rules.
Step 1 of 2
Name the era to locate the orgthe field team's vocabulary, borrowed
The seat-based curve above is your account diagnostic. Cursor's own leadership pairs it with a capability diagnostic: the three eras of AI coding, each defined by where developers spend their attention. Naming the era an org is stuck in is a sharper way to say "here's why depth is shallow" - and it gives a leader the vocabulary to see the next step.
- Era
- Era 1 - Tab / autocomplete
- Where attention goes
- Keystrokes
- Tell-tale in an account
- Usage is Tab-only; engineers treat Cursor as faster typing
- Era
- Era 2 - Synchronous agents
- Where attention goes
- Steering (riding shotgun)
- Tell-tale in an account
- Agent used live, watched step by step; this era lasted under a year
- Era
- Era 3 - Async cloud agents
- Where attention goes
- Reviewing / managing the outcome
- Tell-tale in an account
- Agents run on their own VMs and return artifacts; humans review, not type
| Era | Where attention goes | Tell-tale in an account |
|---|---|---|
| Era 1 - Tab / autocomplete | Keystrokes | Usage is Tab-only; engineers treat Cursor as faster typing |
| Era 2 - Synchronous agents | Steering (riding shotgun) | Agent used live, watched step by step; this era lasted under a year |
| Era 3 - Async cloud agents | Reviewing / managing the outcome | Agents run on their own VMs and return artifacts; humans review, not type |
The Agents Window in Cursor 3 is the surface built for Era 3 - it focuses on the model's outcome, not its exact output.
A complementary ladder starts at manual coding and ends with AI as a co-founder or builder that works backwards from a desired end state. The eras table above covers the middle of that climb; the two ends are what tell a leader how far their org still has to travel.
The aspirational end state: humans manage agents while agents passively watch over the codebase to keep it safe - automations are the next step toward that.
Learn more
Full explanation
Review is the new bottleneck - design for it
With agent-driven coding producing code at 10x-100x prior rates, far more PRs get generated than humans can read, so quality and review - not generation - become the constraint. An org that only counts "more code shipped" is measuring the wrong end of the pipe.
The target behavior: humans stop reading every line and instead read a tight agent-written summary - "the agent generated 3M lines; here are the 7 things you, the expert, should decide on." Flag this early, because it reframes where the account should invest tooling and process as adoption deepens.