高手怎么用 AI？普通人怎么学 AI？投资人如何投 AI？｜对谈课代表立正
Summary
This Shizilukou Crossing episode with Kedaibiao Lizheng argues that advanced AI use is less about better prompts and more about rebuilding work around Agentic Workflow, durable Context Engineering, and reusable AI Skills. The discussion contrasts chat-style AI use with tool-using products such as Codex, Claude Code, and Cursor, then extends the same logic to enterprise deployment through Forward Deployed Engineer roles. The second half surveys AI investing themes, including AI For Science, Embodied AI, World Models, Everything Agent, Video Models, Voice Interaction, and the possibility of WeChat as a high-context agent platform.
Key Claims
- Advanced AI users do not merely ask better questions; they build scaffolding around AI, including tools, preferences, documents, process memory, and judgment standards.
- Chat interfaces have limited compounding value because context must be restated, outputs do not naturally accumulate, and tool access is constrained.
- Agentic Workflow tools such as Codex, Claude Code, and Cursor matter because they can work across files, tools, and persistent project context.
- AI Skills are valuable when they encode implicit knowledge into explicit boundaries, procedures, standards, and required context; simple style imitation is a weak version of the idea.
- Context Engineering may become a durable differentiator as models and base tools converge.
- AI entrepreneurship is getting harder for conventional SaaS patterns because users and competitors can recreate lightweight tools quickly with agents.
- The episode frames Forward Deployed Engineer work as a likely enterprise bottleneck: AI needs people who can map processes, culture, databases, and organizational context before deployment.
- The investment discussion suggests two broad escape routes from foundation-model pressure: moonshot domains such as AI For Science and Embodied AI, or deep industry work with hard operational detail.
- The host’s AI investment watchlist centers on Everything Agent, Embodied AI, World Models, Video Models, software workflow rebundling, and Voice Interaction.
Key Quotes
“从聊天到 agent” — the episode’s recurring framing for the shift from casual AI use to tool-using work systems.
“脚手架” — the guest’s metaphor for the tools, context, preferences, and standards that make AI more useful over time.
“skill as a service” — the guest’s commercial framing for reusable AI workflows that may be easier to sell as outcomes than as raw skill files.
“旧商业模式的石头不见了” — the guest’s description of how familiar startup assumptions are weakening in the agent era.
“第二次文艺复兴” — the phrase used for a possible creative and generalist resurgence caused by AI reducing routine tool work.
Connections
- Kedaibiao Lizheng — guest whose AI learning advice emphasizes agents, skills, context, and workflow redesign.
- Shizilukou Crossing — podcast/media context for the episode.
- Codex, Claude Code, and Cursor — concrete tools used to illustrate the move from chat to agents.
- OpenAI and Anthropic — frontier AI companies referenced in discussions of agent products, market structure, and enterprise deployment.
- Context Engineering, AI Skills, and Agentic Workflow — the episode’s core practical framework for becoming more effective with AI.
- Forward Deployed Engineer — role presented as an enterprise implementation layer for AI.
- AI For Science, Embodied AI, and World Models — frontier investment directions framed as harder to commoditize.
- Everything Agent, Video Models, and Voice Interaction — application and interaction directions the host is watching.
- WeChat — discussed as a possible agent platform because of its user base, social graph, context, and mini-program ecosystem.
- ZhenFund — entity connected to the host’s investment work and AI founder network.
Contradictions
- None identified. This is the first ingested source in the wiki, so there is no prior wiki content to conflict with.