Curosr: Expanding our long-running agents research preview (cursor.com)

Cursor has expanded its long-running agents research preview, now accessible to Ultra, Teams, and Enterprise users at cursor.com/agents. This development follows prior research revealing that frontier models struggled with complex, long-term tasks. To address these limitations, Cursor created a custom harness, enabling agents to autonomously manage larger projects. Initial results are promising, demonstrating that these long-running agents can handle intricate tasks like building chat platforms and refactoring systems, with merge rates on par with other models. The significance of this advancement lies in its potential to transform software development workflows. By emphasizing strategic planning before execution, these agents can align on task requirements upfront, thereby reducing errors and improving outcomes. The harness utilized allows for collaboration among multiple agents that work together to complete complex tasks, achieving results faster and with higher quality. As developers find themselves able to initiate massive coding tasks that run continuously without supervision, the end vision of self-driving codebases becomes increasingly attainable, paving the way for more efficient development processes and fewer human interventions in the future.