According to @scottshics, an open-source team released OpenManus to run a Manus-like agent locally, and he refactored inefficient memory organization to optimize the memory stack so Claude 3.7 could run it smoothly, with a local demo shared (source: @scottshics). According to @scottshics, the agent showed broad generality by ordering Uber and UberEats with almost no additional customization and needed limited assistance for flight booking (source: @scottshics). According to @scottshics, the main blocker is authentication, as he had to manually scan a code to authorize the agent to act on his behalf (source: @scottshics). According to @scottshics, these results verify his earlier vision and signal a rapidly improving agentic era (source: @scottshics). For trading relevance, the update identifies two concrete variables to track from this field report: local agent performance enabled by memory optimization and the authentication bottleneck constraining autonomous actions (source: @scottshics).