Cursor didn't remove the IDE. Here's how to get it back
Cursor's Agents Window can open first, but Editor Window remains supported. I map the commands and startup setting that put code back in front of you.
Cursor did not remove its IDE when it introduced Agents Window. Editor Window still exists, and Cursor documents several routes back to it. On a cold launch with nothing to restore, the agent-first window can appear first. To me, that makes agents feel primary, but it does not eliminate the code-first workflow. [1] [5]
How do I get the Cursor IDE back?
Choose Open IDE from the command palette in Agents Window. Cursor documents that this opens the current workspace in Editor Window, while Cursor staff also points to File → Open IDE. [1] [5] For a new editor window, use Cmd/Ctrl+Shift+N; for an explicit classic launch, run cursor --classic. [3]
To make code appear first more often, change Cursor Settings → General → Startup → Window Restoration to Last Used Windows, then close Cursor from Editor Window. The preference lives in Cursor Settings rather than the normal VS Code settings. Cursor staff warns that restoration can still fail in some cases, so cursor --classic is the clearer route when the opening window matters. [3]
The route also works in the other direction. From Editor Window, Open Agents Window opens the agent-first view, and Cursor says both windows can remain open at once. [1] You are choosing where to begin and what to keep visible, not locking a project into one mode.
What did Agents Window change? Tasks now come first
Cursor changed the starting unit of work. Editor Window begins with code to inspect and change, while Agents Window begins with tasks that agents can run across repositories and environments. Cursor introduced the interface on 2 April 2026 and presented parallel agent management as its main purpose. [2] [4]
The distinction is concrete in Cursor’s own descriptions. Its staff calls Editor Window the classic Cursor or VS Code flow, with files, a terminal, extensions, and flexible views. Agents Window is the agent-first interface for running work in parallel. The Cursor 3.0 changelog adds the execution scope: agents can run locally, in worktrees, in the cloud, and through remote SSH across multiple repositories and environments. [2] [4]
The latest staff explanation I found, posted on 25 August 2026, says Agents Window opens on launch when Cursor has no other windows to restore. A restored session, a changed preference, or an explicit classic launch can lead to Editor Window instead. Agents Window can lead on a cold start, but it does not always open first. [3] [5]
One launch point, two supported work surfaces
The first surface changes what receives attention, while documented commands keep both workflows available.
The figure makes the practical distinction visible: Agents Window starts with delegated tasks, while Editor Window starts with the code. Both remain supported.
Why did users think Cursor removed the IDE?
For the users in these threads, Agents Window made code feel secondary even though Cursor still supported Editor Window. Three Reddit discussions from 18 to 22 August 2026 capture that confusion: users objected to the new view, compared Cursor with a chatbot, or could not find the editor setting. Replies pointed back to the retained editor and startup controls. [6] [7] [8]
On 22 August 2026, a post titled “I hate the Agents Window” had about 103 votes when this article was researched. The author objected that code felt secondary and that supervising many parallel agents did not match how they think about tickets. Other replies said they had switched the surface off or set window restoration and moved on. [6]
A separate post from 21 August 2026, captured at about 50 votes, said Cursor used to feel like an IDE and now felt like another chatbot. Replies pushed back by pointing out that both windows still existed and could be open together. [7]
The smaller but more revealing support question arrived on 18 August 2026. Its author simply could not find the setting that would make the IDE open instead of the agent view. [8]
The threads establish confusion and conflicting preferences, while Cursor’s documentation establishes that both windows remain supported. The vote counts are moving snapshots, so they cannot tell us which window most users prefer. That is the evidence boundary that matters here.
My read is that Cursor underestimated how much a default communicates. When an AI editor opens on tasks and agents, files become something you enter after choosing work rather than the thing visible before the task is described. The editor did not disappear, but Cursor made it feel less central.
Cursor’s two windows change what you review first
Editor Window keeps the patch in front of you, while Agents Window keeps delegated tasks in front of you. For my workflow, the practical difference is where review begins and how many completed changes may be waiting when the agents stop.
In Editor Window, the source tree, open files, terminal, extensions, and split views remain visible while an agent works, so you can inspect the implementation as it develops. Cursor documents those capabilities. [1] [4]
In Agents Window, delegated tasks stay in view while agents work across separate repositories and environments, including worktrees and cloud tasks. [1] [2] I find that arrangement easier for independent tasks, but it does not remove the need to read their diffs, run acceptance tests, or decide which changes should be merged.
This is not a choice between manual coding and automation. Both windows use Cursor’s agents, and Agents Window still includes file access. Cursor documents Cmd/Ctrl+P for finding, viewing, and editing a file there, plus Cmd/Ctrl+Shift+F for searching across files. [1] The choice is whether code review stays beside one active change or becomes a later step across several delegated changes.
Worktrees help the second workflow because they keep parallel file states separate, but separate files do not make the patches correct. In practice, ten independent tasks can still leave one developer with ten changes to understand. [2]
Which Cursor window should I open first?
Open Editor Window first when one patch needs close inspection or your work depends on the terminal, extensions, and several file views. Open Agents Window first when independent tasks can run across repositories or environments. My preference is to keep code first until parallel work gives me a clear reason to switch. [1] [2]
A fair performance comparison would give both windows the same clean starting commit, model, settings, prompt, acceptance test, and task scope. I would record task time separately from review time, then count changed files, rejected edits, and whether I could explain the final patch without reopening the agent transcript.
For parallel agents, I would start with tasks whose files and acceptance tests do not overlap, then stop adding concurrency when the review queue grows faster than I can clear it. That decision follows from the limited attention available to any reviewer, not from a Cursor performance claim.
Based on Cursor’s documentation on 27 August 2026, I would set Last Used Windows, close from Editor Window, and keep cursor --classic as the explicit route. Those paths are enough to put the IDE back in front of me without giving up Agents Window when parallel work actually helps.