Cang Shifu

Cang Shifu is the guest identified as 硅藏藏师傅 in Agent 元年第 500 天：什么在消失，什么在诞生——为什么我们不该再投资 GUI 思维的软件？. He describes agent use from a hands-on builder perspective, including maintaining large codebases, automating content work, using CLI tools, and creating skills.

1 人公司，扛 5 个人的活，还要管 50 个 Agents？｜S10E18 appears to use 归藏 for the same builder context: a Copilot author, PPT skill creator, and Code Pilot builder. In this source, Cang Shifu is the counterweight to Yu Yi’s partner-like agent framing. He treats agents more as controllable tools and workflow components, arguing that long unattended runs can compound errors and drift away from product principles, content principles, or aesthetic standards.

Position In The Episode

• Argues that AI expanded his practical capability, especially for software maintenance and automation.
• Uses FFmpeg-style command-line work to explain why Agent-Facing Interfaces can make hard tools easier for people through agents.
• Recommends a PPT-oriented AI Skills workflow and describes skill commercialization as still early.
• Suggests Chinese agent progress is slower partly because long-task agent loops depend on frontier model quality.
• Introduces Code Pilot as a localized, open coding-agent project that keeps skills, memory, CLI, and harness elements closer to the user.
• In S10E18, argues that non-technical AI builders still need to learn requirements, testing, and result judgment.
• Uses a bounded parallel-work pattern: two or three AI tasks can run at once, but they need review rather than hours of unsupervised autonomy.

Connections

• Tianjie Jack — co-guest discussing investment and product-design implications.
• Agentic Workflow, AI Skills, and Context Engineering — concepts reflected in Cang Shifu’s workflow examples.
• Code Pilot and Open Cloud — agent projects discussed in the episode.
• Human-Agent Collaboration, Agent Permission Boundaries, and AI Use Pacing — S10E18 themes around reliability, review cadence, and agent autonomy.