What the author actually reported
Shah says he was bouncing between the editor, an AI chat, a terminal and other panels, and the friction pushed him to experiment. He credits VS Code's new Agents window but calls the experience an afterthought. He felt it was an agent workflow squeezed into a traditional editor. That is his opinion, not an established product fact.
He tried three tools:
- Cursor: he praised its polish and its dedicated agents window.
- Google Antigravity: he liked the 2.0 update, which he describes as a standalone, agent-first app.
- Devin Desktop: this became his daily driver, mainly because of its Kanban-style Agent Command Center. It lets him see agent tasks by stage, open a session, inspect changes and return to the board.
He gives an example of the shift. He could hand an agent a whole job, such as redesigning a dashboard, updating its navigation and fixing related layout problems. He no longer directed every small step. He stresses that he still had to review the generated code and verify the results. He also says outright that VS Code isn't obsolete.
Two caveats on the story itself
The duration is inconsistent. The headline and opening say a month. The closing section refers to "an entire week" away from VS Code. Both come from the author, so treat the exact length as unresolved. The experience is real enough, but it may be shorter than the headline suggests.
Devin Desktop isn't a brand-new editor. Devin's own FAQ says Devin Desktop is the new name for Windsurf, with the same IDE, editor and features unified under the Devin brand. The company says it arrived as a standard over-the-air update, with plan, pricing, extensions and settings carrying over automatically. Coverage dates the change to June 2, 2026. Web Developer reports that Cascade, the older agent, reaches end-of-life on July 1, 2026, replaced by the Rust-based Devin Local with sub-agents. That's reported by one outlet, so check Devin's docs before relying on it for planning.
How the Devin Desktop command center works
Devin's documentation describes the Agent Command Center as a surface for managing every local and cloud agent in one place, organized as a Kanban board grouped by status so you can see what's working, blocked or ready for review. The board is the default view, but the full IDE is still there. Devin's FAQ states that nothing is getting removed. That matches Shah's note that he can still edit files and debug manually.
One reported extra is support for the Agent Client Protocol. According to CodingWithAI, any ACP-compatible agent can run inside Devin Desktop at launch, including Codex, Claude Agent and OpenCode. Shah doesn't mention this, and I haven't verified it independently. It is a notable difference from a single-vendor agent setup.
Where VS Code stands
Microsoft's current documentation shows VS Code isn't standing still, and the split between its two surfaces is deliberate:
- Agents window: for delegating tasks across several projects, tracking progress and reviewing results in one place.
- Chat view beside the editor: for work that needs frequent editing, debugging or testing in one project.
- Session handoff: Microsoft says you can continue supported sessions between the two interfaces.
Microsoft's docs also list limits that matter if you're comparing tools:
- Copilot Cloud sessions work only for GitHub-backed repositories.
- Multi-root sessions aren't yet supported in the Agents window.
- By default, the window requires GitHub authentication.
So "manage agents across projects" doesn't mean every agent type or repository layout is supported equally. That's worth checking before you assume a switch or a stay will fit your setup.
The other two tools, briefly
Cursor. Cursor's 2.0 changelog from October 29, 2025 describes a multi-agent interface. It says users can run up to eight agents in parallel on a single prompt, with git worktrees or remote machines preventing file conflicts. It also says each agent works in an isolated copy of the codebase. That's a dated snapshot, not a guarantee of today's behavior.
Google Antigravity 2.0. Google's May 19, 2026 announcement describes a standalone desktop application for macOS, Linux and Windows, where there is no IDE. Google says it groups conversations by project, which can span multiple folders with their own permissions, rather than by repository. Google also recommends using it alongside an IDE of your choice. That undercuts any reading of the product as a straight VS Code replacement. In Google's framing, it's a companion, and Shah's "agent-first" description is accurate.
What to take from this
The useful distinction isn't VS Code versus the rest. It's two working styles:
- Code-first with AI help. You stay in a project, in an editor, with chat alongside. Microsoft's docs still point this way for focused work.
- Task delegation. You assign work, monitor several agents and review the results. Shah's tools put this up front.
The two aren't exclusive. Microsoft's docs warn that AI can produce incorrect code or misunderstand your intent, and that you remain responsible for which changes reach your codebase. They also say a Git worktree keeps code changes separate but isn't a security boundary. That is Microsoft's guidance for VS Code, and the same caution is reasonable for any agent tool, though I haven't verified each vendor's isolation model.
If you want to try this yourself, treat it as a time-boxed experiment, not a verdict:
- Pick one real project and one agent tool, and decide in advance what "better" means for you (fewer context switches, faster reviews, fewer regressions).
- Check that it supports your repository host and project layout.
- Look at how parallel tasks are isolated and how you review diffs before merging.
- Keep your normal editor installed for manual debugging.
Shah's account is a credible signal that agent-management interfaces are becoming a real product category. It isn't evidence that they're more productive or safer than a well-configured VS Code setup. That would need controlled testing, and this piece doesn't provide it.