“AGI 来了？我用了一周，头皮发麻“｜对谈张昊然：Moxt 联合创始人

Summary

This Shizilukou Crossing episode interviews Zhang Haoran, co-founder of Moxt, about why he feels “AGI has arrived” inside daily digital work. The discussion frames Moxt as an AI-Native Workspace where AI Coworkers live inside Organizational Context, read and write AI-friendly artifacts, participate in collaboration, and generate work surfaces from current context. The episode extends the wiki’s agent thread from personal OS context toward organization-level context, while adding a human-centered boundary: Moxt wants agents to amplify people rather than market them as cheaper replacements.

Key Claims

• Moxt is positioned as an AI-native workspace where users create and train a team of AI coworkers, starting from the default assistant Momo.
• Zhang Haoran uses a digital-work definition of AGI: in many industries, most people’s work can increasingly be handled by AI systems when the context is available in a usable form.
• The episode’s central product claim is “more context”: AI performs better when documents, meetings, data, project plans, code changes, and communication state live in an agent-readable workspace.
• AI-Native Workspace design differs from adding agents onto older human-first tools because Moxt treats Markdown, CSV, JSON, HTML, and file-system-like structure as the default work substrate.
• Organizational Context is not only a knowledge base; it includes current project state, PRDs, data definitions, meetings, ownership, and daily team progress.
• AI Coworkers in Moxt are goal-oriented roles with memory and skills, not just named prompts. Zhang Haoran describes sales, project-sync, thinking, creative, and critic agents around his own workflow.
• Moxt can turn existing context into Generated Work Interfaces, such as project boards, business dashboards, analyses, and HTML visualizations, which weakens the need for fixed Jira-like screens.
• The team reduced some synchronous meetings because AI could preserve and surface work state, but the episode still treats human ritual, emotion, taste, judgment, and value choices as important.
• The Slack discussion shows a friction point in current Human-Agent Collaboration: an IM built for human-to-human messages may not naturally support many AI agents and personal assistants in the same workflow.
• Token cost is presented as another management target: a user can make spending limits part of an agent’s goal rather than treating cost control as a separate back-office setting.
• Moxt’s “Agents MD” document is described as a high-authority value and instruction layer, similar to a space constitution, with “amplify people” and privacy/security as core boundaries.
• Zhang Haoran sees incumbent tools such as Notion, Feishu, DingTalk, Jira, Slack, and Manus as relevant competition, but argues that legacy products face an innovator’s-dilemma problem if the workspace itself must be rebuilt around AI.

Key Quotes

“AGI 已经来了，只是我们打开它的方式还不对” — Zhang Haoran’s opening provocation about digital-work AGI.

“Markdown、CSV 和 HTML 可以被看作新时代的 Doc、Excel 和 PPT” — Moxt’s AI-readable work-format thesis.

“放大人” — Moxt’s stated value boundary against replacement-first AI employee marketing.

Connections

• Moxt — central product and company context.
• Zhang Haoran — guest and co-founder explaining the product thesis.
• Momo — default Moxt assistant used as the starting AI coworker.
• AI-Native Workspace, Organizational Context, AI Coworkers, and Generated Work Interfaces — core new concepts added by the source.
• Context Engineering, Agentic Workflow, Persistent Agent Memory, and Agent Harness — existing agent concepts reinforced by Moxt’s more-context and memory/skill claims.
• Human-Agent Collaboration, Digital Employees, and AI Organization Design — human, role, and organization-design themes sharpened by Moxt’s AI-coworker framing.
• Agent-Facing Interfaces and Headless Software — adjacent interface themes around agent-readable artifacts and generated UI.
• Slack and Manus — reference products used to explain current collaboration friction and competitive pressure.
• Human Judgment Under AI and AI Workforce Monitoring — human-value and monitoring boundary connected to Moxt’s “amplify people” posture.

Contradictions

• No direct contradiction with prior wiki content. The source reinforces the existing view that agent usefulness depends on context, memory, tools, permissions, and workflow redesign rather than model calls alone.
• The source adds a tension with replacement-oriented Digital Employees marketing: Moxt uses AI coworker language but explicitly rejects positioning AI as a cheaper substitute for human employees.
• The source also shifts the context discussion from personal OS-Level Context and Intent Context toward shared Organizational Context, suggesting that some agent products may win by moving users into a new workspace rather than staying entirely inside existing surfaces.