AI Secret’s scale testing revealed a clear difference between two agent platforms, Hermes Agent and OpenClaw, when handling long-running, multi-step tasks. In these tests Hermes repeatedly continued work across turns, while OpenClaw frequently stopped after completing a slice, asked another question, or waited for a new user prompt.
What happened
- Test runs used tasks that require multiple consecutive steps to complete.
- Hermes’s Persistent Goals mechanism carried goals automatically across turns, allowing the agent to keep executing without explicit user prompts.
- OpenClaw’s Goals track the overall objective but, in the testers’ experience, do not maintain continuous progress: the agent often paused and required further instructions or user intervention.
Why this matters in practice
Testers argue that requiring users to type “continue” or to select detached tasks or configure TaskFlow in order to prevent work from stalling is effectively configuration masquerading as autonomy. If the human must keep prompting the system, the workflow has not been truly automated—management has simply been shifted onto the user. While OpenClaw may offer more plumbing and configuration options, the tests showed Hermes produced the better outcome for sustained task execution.
Conclusion
Based on the AI Secret scale tests so far, the agent that needs the least babysitting delivers the better result; today that is Hermes Agent. OpenClaw’s behavior during long tasks often resulted in interruptions, whereas Hermes’s persistent-goal approach enabled continuous progress without repeated user prompts.
