Open Claw had the viral moment earlier this year. The momentum has moved to a rival called Hermes, and the reason matters more than the leaderboard.
Hermes is open-source software that runs AI agents on a user's own device to send emails, write code and answer messages. Over the past month it has passed Open Claw in GitHub contributions.
The feature developers actually want
The draw is a capability called skill-writing. A skill is a set of instructions telling an agent how to perform a task.
Most agents make users write those instructions or download them. Hermes writes its own. It takes notes on how it completed a job, then reuses them, which removes the manual work that slowed earlier tools.
Why the "sleep" idea matters
The part that catches the eye happens when the agent is idle. Hermes reflects on its skills, simplifying and combining them to work better next time.
Its founders compare the process to human sleep, when the brain commits what it learned to memory. The agent reads back through its skills, finds weak spots and tidies them without anyone asking.
Close to recursive self-improvement, not the same
This sits near a concept the industry treats with caution. Recursive self-improvement describes AI that trains itself to get better, the scenario that worries safety researchers.
Hermes does not do that. The underlying model stays fixed. What improves is the library of skills around it, a softer version of the same idea. The distinction is real, but the direction of travel is the point.
Open Claw's problem
The contrast with Open Claw is sharp. After its breakout, Open Claw is now wrestling with bugs and maintenance, and the early excitement has cooled.
Reliability is the issue. Big businesses will not trust an agent to message and email on their behalf if it cannot be depended on. Open Claw's creators may recover, but the doubt has set in.
Who wins if trust is the gate
The deeper barrier for every agent is trust. Users hesitate to let software act for them.
That favours established names. Products from OpenAI and Anthropic, such as Codex, may gain ground as general-purpose agents because users trust the brand behind them. Anthropic, whose Claude models sit beneath Open Claw, could benefit either way.
Nvidia is building its own version, Nemo Claw, though adoption remains thin.
Hermes has the developers and the cleverer design. Whether that converts into the trust needed for real-world use is the test it has not yet faced.
