I dismissed Cursor the first time I tried it, somewhere around January. It felt like a poorly-copied VSCode — awkward, derivative, trying too hard. Everyone I knew praised it, but I didn't see it. Whatever they were building, it wasn't for me.
Then Cursor 2.0 dropped.
The SMS That Said Everything
I hadn't logged into my Cursor account in months. When I finally did, they sent me a verification code via SMS. The sender? Not "Cursor" or "Cursor IDE."
Work OS.
That's when I realized: they're not trying to build a better code editor. They're trying to replace how we work entirely.
Files Are Out, Agents Are In
Cursor 2.0 doesn't care about files anymore. It cares about agents.
The experience feels like managing employees via chat. You spawn a new agent, give it a task, click to the next one. Each agent has its own chat window. You fire off tasks to different agents and wait for any of them to respond when they're done.
It's mindblowing. Like running multiple Cline tabs or multiple Claude Code terminals in parallel — except Cursor redesigned their entire UI around this experience.
They even built their own proprietary model for this: Composer 1, which they claim is fast and close to frontier-level quality. That deserves its own post after I test it more thoroughly, but the signal is clear: they're betting everything on multi-agent workflows.
The Parallelization Paradox
Here's the thing: running multiple coding tasks in parallel on the same repo hasn't proven that useful for me — at least not for code.
It can work when you're fixing unrelated bugs that are "distanced" from each other in the codebase. But here's the catch: the coordination overhead kills the benefit. By the time you've analyzed which bugs can safely run in parallel without stepping on each other, you could have already fixed them one by one with a single agent.
But for other tasks? Like investigating telemetry across multiple services — something I do constantly — this is dramatic. Spawn one agent to check error rates in prod, another to analyze logs, another to query metrics. Suddenly you're not bottlenecked by sequential execution anymore.
Human-Orchestrated vs. AI-Orchestrated
This is where Cursor 2.0 diverges sharply from something like Claude Code.
With Cursor, you are the manager. You spawn four agents to handle planning, implementation, testing, and code review. You coordinate. You decide.
With Claude Code, the main agent is the manager. You give it a goal, and it decides whether to spawn sub-agents — maybe one for UI, another for backend. The main agent orchestrates: it spawns a code reviewer agent, which reports back findings, then the main agent applies fixes or delegates them to the sub-agents that wrote the code. It's iterative orchestration, all happening within a single conversation.
Two completely different orchestration models. One gives you control. The other gives you delegation.
Neither is strictly better — they're optimized for different mental models. Cursor assumes you know how to break down the work. Claude assumes the AI should figure that out.
Intent Over Implementation
Ever since I discovered these coding agents at the start of the year, I've believed we're entering an era where we care less about how code is written and more about what it's supposed to do.
Cursor 2.0 takes us a few steps closer to that future.
The file explorer still exists — you can open it if you really want to — but the default view is the Agents Explorer. The interface is built around agents and tasks, not files and folders. A short, concise review experience lets you see what code changed across all files — not a tedious scroll through diffs in each individual file.
The interface is designed around the assumption that you want to think about outcomes, not file structures.
Not There Yet
To be clear: we're not all the way there. The illusion breaks. You still need to review what the agents actually wrote. You still need to understand the codebase. The dream of "just describe intent, get perfect code" is still a dream.
But the direction is right. And more importantly, Cursor is designing for that future, not for the past.
They're not building a better VSCode. They're building something else entirely.
The "Work OS" branding isn't just marketing. It's a signal. They're not competing with editors anymore. They're competing with how we think about work itself.
