I work as a freelancer, and in the last 4 months since I started using Cursor, I delivered 3 projects for some large companies and 2 others smaller projects for some startups. - production ready launched products.
Only a File Explorer and a Chat window - the power of simply having AI integrated into the IDE is so great.
I used no fancy multi-parallel-agents-herd and I have never felt the need. Actually I would say it was so much more productive to fully focus at issues one at a time, and just take advantage of finishing something that would have taken 1 day in 1h. - then move to the next task.
On the other side I have seen Codex and Glass now, where the direction is to have all of your projects ever opened all in the same Agents sidebar, to run 10 Agents in parallel, manage them, signalling which one needs your attention just not to miss something.
So my question is, what are you guys working on? What are you actually building that instead of having one productive focused session - there is the need to manage 10 agents working in parallel?
It’s a good question, like you I was a skeptic (and still am to some extent)
I mainly use Glass (or my own version of it) for projects where I really don’t care about the code. I wouldn’t use it for professional projects like work
At the minute I have the luxury of time to play with a bunch of side projects such as my own Agentic IDE, a markdown/screenshot editor and a Developer Interview preparation web app.
Most of the time with these projects I don’t really have a plan - I’ll start off with GPT 5.4, test it and have a bunch of ideas which I feed back into agents as I go in a trial and error phase.
Doing this in a traditional way i.e. via a traditional IDE is really time consuming. You get lost in agent tabs with the chat popout, you get distracted by the code and end up juggling multiple windows and terminals. It’s really about removing any friction or distraction to building stuff and getting comfortable not caring about the code
Tldr Id say find something that annoys or slows you down. Build an app for it. Get comfortable not knowing or caring about the code and you’ll start to see the appeal of these so called agent orchestrators. You start to get into a flow - spin off a bunch of asynchronous individual agents, test as you go, and approve their outlet or give them followups.
I really recommend building your own agent orchestrator / vibe coding tool, I made a thread about it and open sourced it powered by the Cursor CLI. It’s awesome to make your own software, with the features you need and no more and exact keyboard shortcuts to fit your exact need.
I haven’t opened cursor in a while. Anyway from someone who was skeptical maybe that helps somehow !