Anthropic Engineering
· Frontier Labs
Effective harnesses for long-running agents
Agents still face challenges working across many context windows. We looked to human engineers for inspiration in creating a more effective harness for long-running agents.