Comment by Lasang

The concept of long-running background agents sounds appealing, but the real challenge tends to be reliability and task definition rather than raw model capability.

If an agent runs unattended for hours, small errors compound quickly. Even simple misunderstandings about file structure or instructions can derail the whole process.