鸭哥
鸭哥 is a guest in 20 个问题，搞懂 OpenClaw：爆红机制、本质变化、创业机会, discussing Open Claw from an agent-builder and product-design perspective. He frames chat AI as a “consultant” and OpenClaw as an “intern” because OpenClaw can execute, inspect results, and iterate through Agentic Workflow rather than only answer questions.
Source Position
- He emphasizes feedback loops, Local Agent Execution, context, memory, and instruction following over treating OpenClaw as a model-only breakthrough.
- He built related personal-agent experiments, including a heartbeat mechanism and smart-home control, shifting the developer target from human-facing front ends toward AI-readable OpenAPI descriptions.
- He is skeptical of excessive anthropomorphism for engineer-facing tools, preferring agents that talk less and complete more work.
- He treats Agent Harness, memory, context compaction, orchestration, and AI Skills as the infrastructure that makes agent products useful.
- He highlights the safety tradeoff: local execution is valuable because it has context and tools, but too much permission creates Agent Permission Boundaries risk.
Connections
- Open Claw — central project discussed in the source.
- 豪大 — fellow guest in the episode.
- IM Agent Interfaces and Local Agent Execution — product mechanisms he helps explain.
- Persistent Agent Memory, Agent Harness, AI Skills, and Agent Permission Boundaries — concepts reinforced by his comments.