The Vibes of Browser Agents vs IDE Agents

I’ve mostly used IDE-based coding agents like Cline and Copilot, with great success (and frustration). Lately some new entrants in the browser agent space have caught my attention, specifically OpenAI’s Codex and Google’s Jules. While the end result is some code in a PR, the nuances of the experience are quite different.

A surreal nighttime rooftop dotted with glowing microdrones hovering in isolation, casting soft thistle and rose light in a sea of dark blue and black shadows.

IDE Agents

IDE agents are like a having a pair programmer. They are in the mix with you, by your side, writing code with constant guidance and refinement. Hell, Cline still steals your focus and takes over your editor to write code for you, making it impossible to work in parallel on another task. They are very much in the flow of your work, and you can feel their presence as a collaborator.

Browser Agents

Browser agents are more like having a teammate. You delegate unambiguous tasks to them and wait for a PR to review. Meanwhile, you can work on other things. They are not in the flow of your work, but rather a separate entity that you interact with when you need something done. They are more like a remote worker than a pair programmer.

The Vibe Shift

The difference is palpable. I’m finding with browser agents that I am doing way more PM work instead of being in the code. They can do tickets almost as fast I can create them, so the bottleneck is now my ability to create new tasks and review finished PRs instead of how fast Cline can write code.