What the Team Behind Cursor Knows About the Future of Code

Coverage of Every’s first Cursor Camp, featuring Cursor developer education lead and engineer. The central claim is that the traditional IDE paradigm is breaking down as the center of gravity shifts from typing code to managing AI agents that write it.

Key takeaways from Cursor’s team: developers now spend more time directing AI agents than editing code directly; model selection matters far more than prompting tricks (gimmicks no longer work); and agent handoff between local and remote environments is coming. The most striking data point: Cursor’s research team built a working web browser from scratch using AI agents over multiple days, producing 3 million lines of code at a cost of approximately $80,000 in tokens.

The article distinguishes between the model (the AI brain) and the harness (the scaffolding that feeds context, handles errors, and keeps agents on track). A well-tuned harness dramatically improves the same model’s output.

RDCO mapping: Validates the compound-tool thesis — the value is increasingly in orchestration infrastructure rather than raw model capability. The harness concept directly parallels RDCO’s agent architecture where skill definitions and context management determine output quality.