He Guang/Protocol Architect 人机共生时代的关系与边界设计者

Re-architecting organizations for the AI-Native era — where execution breaks long before AI does.

@Remotion @satyanadella

@Remotion 生成一支完整内容视频

@OpenAI If that's true, the future architecture may look like: Knowledge Work (domain object) ↓ Runtime (orchestration) ↓ Loop (reasoning) ↓ Capabilities (tools / skills / MCP) That feels like a natural evolution of AI-native systems.

ChatGPT Work reflects a shift in how people are using AI, moving beyond just answering questions to getting real work done across web, mobile, and desktop. You can ask ChatGPT Work to take on entire workflows with a single request. It will understand your goals, use context from selected apps and files, create polished documents, decks, analyses, sites, and reports, and keep work moving while you stay in control. openai.com/index/chatgpt-…

Introducing ChatGPT Work, a new agent in ChatGPT powered by Codex and GPT-5.6. It can take action across your apps and files, stay with a project for hours if needed, and turn a goal into finished work. It’s a whole new way to get work done.

@OpenAI One question we're exploring: Should the primary runtime object still be Task? Or should it become Knowledge Work—where goals, drafts, plans, execution and artifacts are simply different stages of the same evolving object?

@OpenAI This feels less like another agent feature and more like a shift in the unit of computation. The system is no longer optimizing for responses. It's optimizing for completed work.

@addyosmani Owning the outer loop is powerful—but eventually we'll also need to answer: • Which work is admitted into the loop? • Who authorizes it? • How is responsibility anchored across autonomous execution? Execution is becoming abundant. Governance may become the real bottleneck.

@addyosmani Really like the framing. It feels like we're watching the abstraction layer move upward: Prompt → Agent → Loop. The next shift may be from Loop to Knowledge Work Runtime.

@dotey @cnyzgkc @zarazhangrui @op7418 非常帮

@cnyzgkc @zarazhangrui @op7418 大佬试试我写的 Skill，可以配合 Agent 内置浏览器编辑，可以导出可编辑的 PPTX，可以 AI 配图 x.com/dotey/status/2…

用AI 做PPT，我非常推荐两个人的skill 1. 张咋啦 @zarazhangrui frontend-slides:：github.com/zarazhangrui/f… 2.归藏 @op7418 的 guizang-ppt-skill：github.com/op7418/guizang… 使用方式很简单： 将上述两个地址发给Codex，然后跟agent说安装skill 做PPT的流程，建议先生成文字稿后，再使用skill进行改造

英文视频→双语字幕+结构化摘要，一键搞定 本地Faster-Whisper转写(免费)+LLM翻译(几分钱/视频) 6步全自动，断点续传，多引擎切换 MIT开源：github.com/jie955/video-t…

Thinking Session: When AI Begins to Reveal Thought Itself

AI's next decade isn't about being smarter — it's about making thought visible. ChatGPT made knowledge conversational. Agents made tasks executable. Thinking Session attempts the next step: making thinking itself observable, iterable, compilable. #ThinkingSession #AIDesign

We built AI as an answer machine. But the real use case isn't asking questions — it's thinking. Thinking Session proposes a new primitive: AI as the structural revelation mechanism within the thinking process itself. Not smarter answers. Visible thought.

Three trends converging: (1) LLMs now possess structural understanding, (2) long context enables continuity of thought, (3) user behavior shifting from "asking" to "expressive thinking." When all three mature simultaneously, thinking runtime becomes viable.

Industrial evolution: the history of AI is the history of "core system objects." Chat era → Message. Agent era → Task. Thinking Session era → Thought Stream. For the first time, computers stop merely processing human outputs and start engaging human cognition itself.

Compile: the phase transition. Before Compile, everything is provisional — all structures are candidates, thinking remains open. After Compile, Thinking → Making. System switches to execution: PRDs, ADRs, decisions. "Thinking is done" becomes an explicit event, not an assumption

UserMarker: the sovereignty mechanism. AI may propose structures, organize logic, expose contradictions. But AI may never define the final conclusion. Users pin, reject, rewrite. AI is the observer and facilitator. Humans remain the sole sovereign of their own thinking.

a PM struggling with SaaS redesign. She could only say "it's too heavy, changing it is troublesome." The dual-stream revealed: she was using experience vocabulary ("heavy") not architecture vocabulary ("coupling"). The real problem wasn't click count — it was architectural risk.

Conversation Stream sustains thinking fluidity — follow-ups, reflections, revisions, not rapid closure. Thought Stream runs silently in background — detecting motifs, tensions, drifts, thesis candidates. It observes how thinking generates, without interfering.