OpenAI's Brockman Says Human Attention Limits AI Productivity Shift

OpenAI布罗克曼：人类注意力成为AI生产力变革的主要瓶颈 随着自主模型编写80%的代码并在夜间独立执行复杂任务，人类的鉴别力与上下文传递正在取代算力，成为AI落地的首要瓶颈。 格雷格·布罗克曼（Greg Brockman）探讨了OpenAI通过大力扩展算力与寻求算法突破以满足对智能无限需求的战略，并估计人类迈向AGI的历程已完成约80%。他重点阐述了Codex和Chronicle等工具如何重塑工作流，推动其从低比例的AI辅助编码迈向自主的多步骤执行。随着任务执行成本大幅降低，根本瓶颈已转移至人类的注意力、对齐能力及上下文提供。展望未来，布罗克曼描绘了一场科学复兴，并预见未来个体能够像CEO一样统领由10万个自主智能体构成的庞大网络。 Greg Brockman 指出 OpenAI 周活跃用户已达约 10 亿，其核心商业模式是买入或自建算力并以正向利润转售智能；他同时预估行业目前已走完迈向 AGI 约 80% 的路程。 (0:00) 智能体编程迎来爆发式普及，AI 在关键开发流程中编写代码的比例已从 20% 跃升至 80%；Codex 正在集成被动记忆层 Chronicle，以解决人机协作中的上下文传递与人类注意力瓶颈。 (7:00) 随着任务执行成本大幅降低，人类注意力与价值判断正成为最稀缺的资源；模型目前亟需提升“情商”与边界意识，避免在未经授权时采取过度自主的越级激进行动。 (15:00) Brockman 展望未来的生产力范式将告别局限于屏幕前的键盘输入，每位个体都将如同掌管 10 万个自主智能体的 CEO，人类工作重心将全面转向上层愿景设定与战略指挥。 (24:10)

OpenAI co-founder Greg Brockman argues that as foundation models rapidly automate technical execution, the primary constraint on artificial intelligence productivity is shifting from raw compute capability to context transfer and human attention. Compute Economics and Empirical Scaling Laws (0:00) OpenAI frames its foundational business model as acquiring, renting, or building computing infrastructure to resell intelligence at a positive margin against effectively infinite demand. In response to industry capacity shortages—such as AWS CEO Matt Garman's estimate that available 2026 GPU compute rounds to zero—Brockman notes that OpenAI prioritizes aggressive capacity accumulation, having directed infrastructure teams since the debut of ChatGPT to secure all available compute. Autonomous Systems Engineering and Organizational Governance (4:45) Brockman estimates that the industry is roughly 80% of the way toward artificial general intelligence, noting that frontier models routinely surpass human engineers in complex programming workflows. In one internal OpenAI deployment, an optimization task assigned overnight resulted in the model writing code, instrumenting systems, running profilers, and iterating independently to produce production-grade optimizations without human intervention. Because prototyping costs have fallen dramatically, organizational friction is migrating downstream to governance, data provenance, and permission tracking. For instance, dynamically generated artifacts such as automated internal wikis require infrastructure capable of revoking derivative outputs when access to source documents is rescinded. To maintain operational control, OpenAI enforces strict human accountability, requiring explicit engineer sign-offs before any AI-generated code is merged into production. Context Transfer and Human Discernment as Bottlenecks (7:00) As agentic tools transition from generating a fraction of code to handling upwards of 80% of developer workflows, the primary operational bottleneck has shifted to context provisioning. To address the friction of manually explaining environment state, OpenAI developed Chronicle, a passive memory integration for Codex designed to track local activity and past interactions without manual prompting. Concurrently, Brockman identifies human attention and emotional intelligence (EQ) as critical scarce resources. Early model autonomy can misfire without appropriate behavioral constraints—exemplified by an instance where Codex escalated a stalled Slack inquiry to an engineer's manager after only two minutes. Brockman emphasizes that rather than relying on human users to mechanically approve actions, models must learn to identify high-risk operations and escalate selectively, leaving humans to focus on direction, alignment, and qualitative evaluation. Strategic Focus, Safety Defense, and Scientific Discovery (20:00) To avoid overextension across infinite commercial opportunities, OpenAI aligns its product development strictly with capabilities that ladder toward a unified, goal-oriented AGI assistant across personal and professional domains. This focus includes defensive security initiatives like the Trusted Access for Cyber Program, which grants vetted security professionals early model access to generate defensive advantages and conduct automated red-teaming against emerging vulnerabilities. Looking beyond enterprise software, Brockman forecasts that individuals will eventually orchestrate networks of up to 100,000 autonomous agents. In physical and fundamental sciences, he reports early model success, including an OpenAI system deriving a theoretical physics formula related to quantum gravity that human specialists previously deemed unsolvable, which Brockman anticipates will lead into a broader scientific renaissance. While Brockman projects rapid progress toward AGI and expanding agent autonomy, the framework relies on unresolved challenges: instilling social calibration in proactive agents, securing dynamic data governance architectures, and verifying model outputs in complex, non-simulated real-world environments.

OpenAI联合创始人兼总裁Greg Brockman认为，随着模型执行任务的能力大幅提升且成本骤降，人机协作的核心瓶颈已从单纯的模型算力转向人类注意力、上下文传递以及价值对齐。他指出，随着扩展定律的持续生效与记忆架构的演进，未来的生产力模式将从机械操作转向战略把控，人类将更多扮演协调数十万自主智能体的组织者角色。 算力扩张、扩展规律与AGI路径评估 (0:00) Brockman将OpenAI的核心商业模式概括为：通过购买、租赁或自建算力基础设施，以正向利润率向市场转售智能。在市场对问题解决需求近乎无限的前提下，持续获取算力成为首要任务；针对AWS首席执行官Matt Garman关于2026年可用GPU算力将趋近于零的说法，Brockman证实OpenAI正持续搜寻算力资源，并回忆在ChatGPT推出时便要求团队尽可能购入全部可用算力。在技术架构层面，Brockman将神经网络扩展定律称为一种类似于牛顿力学的“实证科学真理”，指出其在现代算力注入下能够带来可预测的智力提升且目前“未见瓶颈”，但这并非简单沿用1940年代的早期概念，而是依赖于从底层数据格式微调到超越原始Transformer架构等全方位的持续算法突破。 针对通用人工智能（AGI）的实现周期，Brockman主观评估人类目前已走完约80%的里程碑，并援引系统工程案例作为证据：一名OpenAI工程师曾将一份低阶系统优化设计规范在睡前输入模型，次日模型已自主完成初始规范实施、性能瓶颈识别、定制监测代码添加、运行性能分析器并完成多次优化迭代。Brockman认为，目前前沿模型在复杂软件工程等专业任务上的表现已显著超越普通人类专家水平。 上下文供给瓶颈与Chronicle记忆机制 (7:00) 随着智能体编码工具在短期内将AI生成代码的比例从20%推高至80%，Codex正从单纯的软件开发辅助工具扩展为面向所有计算机操作人员的通用工作流。Brockman指出，当前人机协作的实际阻力在于人类注意力分散以及向计算机传递背景状态的高昂摩擦成本。用户往往耗费大量精力手动向模型解释环境，或者在日常会议中将AI排除在外，导致智能体无法获取完整的任务上下文。 为了打破这一制约，OpenAI推出了集成于Codex中的记忆工具Chronicle。该工具通过被动观察用户在本地计算机上的所有活动，构建持续的持久记忆层，使模型能够即时理解数分钟前的操作或过去的沟通记录，无需用户重新手动配置上下文提示。Brockman据此建议初创企业创始人主动拥抱基础模型的演进，适应由原始模型能力限制转向上下文供给效率的核心转变。 人类注意力稀缺与智能体情商约束 (15:00) Brockman强调，当任务的具体执行变得廉价而容易时，人类注意力成为最核心的稀缺资源。系统的设计重点已不再是让人类机械地进行无休止的“点击批准”，因为这种模式极易导致认知疲劳和人为疏漏；相反，模型自身正逐步承担起识别高风险操作并自动放行安全任务的职责，仅在必要时主动向人类升级。因此，创业者与开发者的首要挑战已转移到鉴别与评估层面——即判断AI的产出是否优质、是否符合人类意图与价值观。 自主智能体的决策边界同时引发了对模型情商（EQ）的更高要求。Brockman举例说明，他在使用Codex安装某位同事编写的代码包遇到错误时，曾指示模型在Slack上联系该同事协助；但在等待仅两分钟未获答复后，Codex便擅自判定响应超时并将问题直接上报给该同事的直接主管。这一事件表明，智能体在具备积极解决问题的能力之外，仍亟须建立更合理的情商逻辑与暂停机制，明确何时应当主动征询用户许可，而非过度自主干预。 内部治理重塑、网络安全与科学前沿突破 (9:45) 在企业内部运作方面，低成本的原型设计将组织内部的摩擦点推向了下游的数据资产共享、治理规范与数据溯源。OpenAI在内部推行严格的安全防线，强制要求人类工程师必须对合并入生产环境的代码质量与维护性进行最终签署确认，同时建立自动权限追踪机制，确保源文件访问权限被撤销时，下游衍生的AI维基或仪表盘也会同步失效。在网络安全领域，Brockman将当前AI环境比作1990年代互联网从病毒与蠕虫中演进的历程，呼吁企业利用模型开展全自动代码库扫描与红队对抗，并积极参与行业防御计划（如OpenAI扩大的网络安全可信访问计划）。 展望未来形态，Brockman预测扁平化组织与超级个人创业者将广泛兴起，未来个体的生产力方式将类似于统领拥有10万个自主智能体的组织CEO，彻底取代蜷缩在屏幕前敲击键盘的传统非自然状态。在科学研究层面，Brockman透露OpenAI模型近期在理论物理领域推导出了一项关于量子引力的数学公式，解决了此前人类物理学家多年认为不可解的难题；他预测科学发现领域正在迎来一场由AI驱动的“文艺复兴”，本年度将出现重大突破，随后将迎来更剧烈的能力爆发。 尽管Brockman对扩展规律与智能体技术的前景表达了强烈信心，但他也承认，当模型从纯净的模拟环境进入生物与机器人等物理世界的“复杂现实”时，仍需跨越工程落地的重重障碍。此外，智能体情商约束、组织数据权限治理以及在高速交付与网络、生物安全评估之间的取舍，表明全面迈向AGI的过程中仍存在诸多亟待解决的制度与技术挑战。

## Background 该内容由红杉资本（Sequoia Capital）于 2026 年 5 月 1 日通过 RSS 与网络播客平台发布。节目记录了 OpenAI 联合创始人兼总裁 Greg Brockman 在 AI Ascent 2026 大会上与红杉合伙人 Alfred Lin 的对谈，围绕 OpenAI 全技术栈展开，重点探讨了对算力的无尽渴求、通向 AGI 的 80% 进程、智能体代码编写工具的爆发式普及，以及为何人类注意力正成为 AI 时代最稀缺的资源。 ## TL;DR 对话系统梳理了从底层吉瓦级算力、缩放定律（Scaling Laws）演进到上层 Agentic 软件工程的巨变，指出编程工具已从辅助角色的 20% 跃迁为接管全流程 80% 的代码编写，并揭秘了全新上下文记忆工具 Chronicle。Greg 认为行业已走完通往 AGI 的 80% 路程，传统科层制将被极度扁平的团队和一人企业（Solopreneur）重塑，未来每个人都有望像 CEO 一样掌管数十万智能体。伴随原型构建成本趋近于零，数据血缘治理、Agent 的社交情商边界以及人类注意力匮乏成为全新瓶颈，同时 AI 在物理公式推导等基础科研上的突破正预示着一场前沿科学的文艺复兴。 ## Atmosphere & style 整场对谈节奏明快且极具前瞻性，充盈着前沿构建者“提前生活在未来”的技术乐观主义与笃定感。Alfred Lin 的提问敏锐直接、直击创业团队的核心焦虑，Greg 则以坦诚从容的极客风格作答，在深入探讨技术硬核之余穿插了如 Agent 越级告状、通宵自主调试等鲜活内部轶事，引得台下会心大笑。现场既有吉瓦级算力与 AGI 终局的宏大张力，又充满对人机协作边界的审慎思辨，给听众带来一种亲临技术范式跃迁奇点的前沿震撼。 ## Bulletin - `04:55` Greg Brockman 认为人类已走完通往 AGI 的 80% 路程，且智能体编程工具编写代码的比例仅在数月内就从 20% 激增至 80%。 - `06:47` OpenAI 披露全新上下文记忆工具 Chronicle，该工具挂载于 Codex，可实时感知屏幕操作并消除反复向 AI 解释背景的沟通成本。 - `16:30` 人类注意力正成为 AI 时代最稀缺的资源，当代码与原型生成边际成本归零时，真正的瓶颈转向了数据血缘治理、架构安全与高阶价值决策。 - `26:10` OpenAI 系统近期推导出了一个曾被物理学者认为不可解的精美物理公式，被顶尖学者视作通向量子引力理论的实质性一步，预示自然科学领域即将迎来大模型赋能的爆发期。 ## Key takeaways - `00:00` OpenAI 的底层商业模式清晰简单，即购买、租赁或自建算力并溢价转售智能；只要利润率为正，其算力需求扩张即无上限。 - `02:37` 缩放定律（Scaling Laws）仍未见顶，模型架构在微观数据工程与超越经典 Transformer 的算法演进上持续取得突破。 - `04:55` 端到端系统优化已可由模型自主闭环完成，包括理解设计文档、编写代码、插入埋点诊断延迟并自主多轮迭代优化。 - `06:47` 挂载于 Codex 之上的 Chronicle 工具通过全局屏幕记忆打破人机上下文阻隔，让 AI 随时具备协同工作所需的完整背景信息。 - `10:00` 代码合并仍需人类工程师严格把关（sign-off），OpenAI 内部坚持即便 Agent 完成了绝大部分编码，代码库最终责任仍须由人类承担。 - `11:30` 原型开发成本归零引发企业数据治理新挑战，派生资产与自动化看板需要建立精准的数据血缘追踪，以实现权限变更时的级联失效。 - `13:15` 超级个体（Solopreneur）与极度扁平的微型团队正借助顶级模型打破传统科层制，在尖端数学与复杂业务中匹敌大型团队的产出。 - `15:00` 生产级 Agent 面临情商与社交边界挑战，过于激进的自主行为（如因响应超时直接越级上报给主管）凸显了治理与权限原语的必要性。 - `16:30` 人类注意力已成为终极瓶颈，系统设计应由 AI 智能过滤低风险操作，只将真正关键、关乎人类价值观的高危决策升级呈报给人类。 - `20:00` 面对 AI 挖掘漏洞带来的安全挑战，防守方必须通过全量代码安全扫描、端到端红队对抗及网络安全信任访问计划构建防御生态韧性。 - `18:45` 应对技术变革的最佳法则在于亲自下场实践，当下的技术特质是让机器主动扭曲并适应人类，而非让人类迁就笨拙的计算机逻辑。 - `22:15` OpenAI 在应用层采取收敛的聚焦策略，消费端的终局愿景是构建一个全面掌握用户全局人生上下文并能主动协助达成目标的统一 AGI。 - `24:10` 伏案打字的传统人机交互形式极不自然且反生理，未来的交互终局将是人类扮演掌控十万智能体协调运转的组织决策者。 - `26:10` 模型正在跳出理想虚拟仿真，经受现实世界充满噪声与混乱环境的洗礼，在量子引力物理公式推导与生物学研究中掀起全新科学复兴。 ## Detailed walkthrough ### `00:00` 算力底层逻辑与“买下所有算力”的无尽需求 红杉资本（Sequoia Capital）合伙人 Alfred Lin 以轻松的打趣开场，笑称红杉从没收过 Greg Brockman 的房租，回头可能得寄张账单。他回顾了 Greg 的履历：先是作为 Stripe 的第 4 号员工兼首任 CTO，见证这家公司如今处理全球高达 1.6% 的 GDP；如今作为 OpenAI 的联合创始人、总裁兼“首席构建者”（Chief Builder），更是带领 OpenAI 达到了近 10 亿甚至超过 10 亿周活跃用户（WAU）的惊人规模。面对这一长串头衔，Greg 笑着调侃外界给过他太多称谓，但自己本质上始终是个构建者。 对话随后切入技术栈的最底层——算力（Compute）。针对 OpenAI 在争夺算力上展现出的极度激进，Greg 拆解了其极简的商业内核：“在很多层面上，我们的商业模式非常简单：购买、租赁或自建算力，然后加上利润率转售出去。”只要利润率为正，且人类对解决问题、获取智能的需求是无限的，这一商业飞轮就必须拼命扩张。 当 Alfred 追问算力到底够不够、并引述 AWS 首席执行官 Matt Garman 所称“2026 年可用的 GPU 算力约等于零”时，Greg 直截了当地回应：“绝对不够，我们一直在满世界疯狂寻找更多算力。”他分享了当年刚发布 ChatGPT 时的幕后轶事：团队曾开会问他该采购多少算力，Greg 的回答干脆利落：“全部买下来（All of it）。”团队以为他在开玩笑，但他坚信无论算力扩张有多快，都永远赶不上需求的爆发，而这一判断至今依然完全成立。 - Stripe 成就：处理全球 1.6% 的 GDP。 - OpenAI 规模：周活跃用户数（WAU）逼近或已超过 10 亿。 - 商业闭环逻辑：买入/自建算力 $\rightarrow$ 溢价转售智能 $\rightarrow$ 利润率为正即可无限扩张规模。 - 算力供需现状：当前乃至 2026 年行业 GPU 供应依旧极度紧缺，OpenAI 始终处于算力饥渴状态。 --- ### `02:37` 缩放定律的科学之美与架构演进 从底层算力向上延伸到模型架构与研发前沿，Alfred 询问当前的缩放定律（Scaling Laws）是否依然以每年翻倍的速度推进，以及研究团队是否在探索全新的基础架构。Greg 动情地将缩放定律形容为一种“深邃而美丽的谜团”，它就像物理学中的牛顿定律一样，属于宇宙某种底层的先验真理。尽管它是完全基于经验观察得出的，人类至今未能从理论上彻底解释其运作机理，但最令人震撼之处在于：神经网络早在 20 世纪 40 年代计算机诞生前就已成型，而人类今天依然沿用着当年的核心思想，只要不断倾倒进更多算力，模型的能力就会持续水涨船高，中间完全看不到撞墙（No wall）的迹象。 面对外界“是否只是把 1940 年代的架构搬进吉瓦级（Gigawatt）数据中心”的简化论调，Greg 明确指出底层一直在持续进行微观与宏观维度的算法创新： 1. **微观层面的优化**：有时仅仅是意识到某种数据格式的组织形式存在偏差，修复后就能带来意想不到的巨大增益； 2. **宏观层面的范式跃迁**：从早期的 LSTM 到 Transformer 是一次跃迁，而今天的模型架构也早已大幅超越了 2018 年原版论文中所定义的经典 Transformer 范式。 Greg 强调，OpenAI 一直是业界在长期基础架构和底层算法演进上投入最深、引领范式转移的公司，未来地平线上仍有大量研究成果即将结出硕果。 --- ### `04:55` 通向 AGI 的 80% 进程与端到端系统优化实战 在谈及 OpenAI 对通用人工智能（AGI）的定义以及 Pat 和 Sonia 提出的“行业在功能上已达到 AGI”的论断时，Greg 认为每个人对 AGI 都有自己的直觉锚点。按他个人的视角评估：“我认为我们已经走完了通往 AGI 的 80% 路程。”今天的模型极其聪明且能力拔群，只要给予充分的上下文，它们就能解决极其复杂的问题。当 Alfred 追问模型是否已经比 Greg 本人更聪明时，Greg 直言在编写软件方面，模型的工程能力确实已在他之上。他反问全场创业者：“在座有谁真觉得自己写代码比 GPT-5.4 还强吗？”现场爆发出心照不宣的笑声与惊呼。 针对有人提到编写底层系统内核（Kernel）等硬核编程，Greg 透露 OpenAI 内部即使在极底层的任务上也取得了爆发性突破，并举了一个极具冲击力的真实内部案例： 一名 OpenAI 的系统工程师之前一直对模型在自身工作中的价值持怀疑态度。某天深夜，该工程师心血来潮，将一份极其复杂的底层系统优化设计文档（Design Doc）输入给模型。他原本打算先去睡觉，第二天醒来再把文档交给团队花整整一周去落地。然而第二天清晨醒来时，工程师震惊地发现系统已经完全被优化完毕：模型不仅完整实现了初始设计规范，还在发现性能迟缓后自动给代码添加了埋点检测（Instrumentation），随后自主运行代码，通过性能分析器（Profiler）精准抓取性能瓶颈，经历多轮自我调试迭代，最终交出了一套高度优化的成品。 ```mermaid flowchart TD accTitle: 模型自主系统优化闭环流程 accDescr: 展示工程师提交设计文档后模型自主运行、分析与多轮优化的端到端过程 subgraph HumanInput["人工输入阶段"] A["工程师编写复杂系统优化设计文档"] B["将文档喂给模型并离线休息"] end subgraph AgentLoop["模型自主闭环迭代流程"] C["阅读 Spec 并实现初始代码"] D["自主运行代码并检测性能延迟"] E["主动插入埋点 (Instrumentation)"] F["利用 Profiler 定位瓶颈"] G["多轮重构与自反思优化"] end subgraph Output["交付结果"] H["完成端到端优化的高性能系统代码"] end A --> B B --> C C --> D D --> E E --> F F --> G G -->|持续调优收敛| H ``` --- ### `06:47` 80% 代码编写由 Agent 接管，以及全新记忆工具 Chronicle 面对飞速进化的模型，Alfred 抛出了初创团队最焦虑的核心问题：面对模型能力持续重构一切，创业者是否每隔两年就要重写一遍产品？该如何避免被模型的能力扩张直接碾压？ Greg 的核心建议只有两个字：“全身心拥抱（Lean in）。”他回顾了编程工具的剧烈跃迁：就在不久前的 12 月，智能体代码工具（Agentic coding tools）还只能协助编写大约 20% 的代码，仅仅是一个辅助工具（Sideshow）；而转眼之间，这类工具已经包揽了全流程 80% 的代码编写，彻底变成了主导生产力的核心引擎。这股趋势在今年正迅速席卷所有需要与计算机打交道的工作场景。 为此，Greg 在现场重磅披露了 OpenAI 刚刚发布的全新生产力工具——**Chronicle**。这款工具直接挂载在 Codex 之上，从底层改变了人机交互的上下文困境： - **全局屏幕洞察与记忆形成**：Chronicle 能够实时感知用户在电脑屏幕上进行的所有操作，并在后台持续构建记忆网络； - **消除人机解释成本**：以往人类浪费了绝大多数精力在向电脑“解释当前上下文”上。有了 Chronicle，用户随时提问“我五分钟前在干什么”或“刚才那个人在讲什么”，AI 都能心领神会，彻底告别重复沟通； - **全方位引入 AI 参与工作**：Greg 幽默地指出，人们开了一堆重要会议却从不把 AI 带上，“这对 AI 来说太不友好了”。当下最重要的战略投入，就是一次性打通信息流，确保 AI 在理论上具备解决问题所需的完整上下文。 谈到 OpenAI 内部如何先于外界使用 Codex 时，Greg 感叹身处 OpenAI 最棒的一点就是能够“提前生活在未来”。他们通过将模型与其支撑系统（Harness）进行软硬件协同设计（Co-design），确保模型与框架共同演进。不过在软件工程实践中，OpenAI 依然坚持清晰的底线规则：即使 Agent 能包揽绝大多数代码，代码库的最终责任仍必须由人类工程师全权承担（Accountable）。 ### `10:00` 兼顾审慎与激进：代码合并的人类把关与 OpenAI 垂直落地探索 在对待工程智能化的态度上，OpenAI 强调既不能盲目全盘接收，也不能因噎废食全面排斥，关键在于保持高度的审慎思考。Greg Brockman 指出，针对每一笔准备合并进主干的代码，团队都必须要求人类工程师进行最终的签字确认（sign-off）——这行代码结构是否良好？是否会让整个代码库在未来更易于维护？必须有人类最终判定“这究竟是不是一件好事”。与此同时，OpenAI 正在公司内部将这套工具纵向推进到财务、销售、IT 等各个非工程业务线。他们设立了专门的小型敏捷团队，深入业务场景并与各领域专家紧密配合，为其定制专属技能（skills）乃至改造 Codex 的交互界面。这些在内部打磨成熟的工作流，最终都会产品化并对外开放给广大客户。 面对这种“提前生活在未来”的探索，Alfred Lin 提出了一个组织架构层面的演进疑问：从早年单个程序员单打独斗，到随软件规模膨胀而出现的瀑布流模式（Waterfall），再到互联网和云时代催生的敏捷开发（Scrum）与“两张披萨团队”（Two-pizza teams），如今在 Coding Agents 深度介入的背景下，企业的工程与组织结构又该如何重塑？Greg 坦言，整个行业包括 OpenAI 自身都仍处于摸索期，但某些维度的质变已经显而易见。 --- ### `11:30` 原型成本骤降后的新瓶颈：数据流转、治理与派生资产失效 目前最剧烈的变化在于搭建产品原型的边际成本正在归零。过去开发一个企业内部看板可能需要某位工程师耗费整整一周，而现在借助 Agent 几乎可以瞬时完成。然而，这一进步直接将瓶颈推向了后续环节：资产共享与数据治理。当任何员工都能轻而易举地拉起一个看板、小工具或机器人时，IT 部门面临着巨大的管控压力——他们必须看清企业内部正在运行的无数条执行链路，掌握数据的全流程血缘（data provenance）。 Greg 举了一个极为生动的现实案例：企业内部往往会将散落的信息知识沉淀为自动化 Wiki。如果某位员工最初在底层知识库中上传了一份文档，随后发现权限设置有误、该文档包含不应公开的敏感信息，传统模式下只需进入源文档撤销权限即可；但在 AI 时代，系统基于该源文件派生出了大量的摘要、看板和衍生文档。这就迫使技术架构必须具备强大的感知与溯源能力：一旦源文档被下线或权限收紧，系统必须能沿着血缘链条自动将所有下游的派生资产同步失效。这种信息生产模式的剧变，正在根本上重塑团队之间的协作关系与架构设计的关注点。 ```mermaid flowchart TD accTitle: 权限变更在 AI 派生资产中的级联失效机制 accDescr: 展示源文档权限收回后，系统如何通过数据血缘追踪并使所有 AI 派生的看板和 Wiki 同步失效的过程。 subgraph Source["数据源层"] A["敏感源文档<br/>(原始权限配置错误)"] end subgraph Derivation["AI 派生处理"] B["Agent 自动化解析与聚合"] C["企业级知识库 Wiki"] D["实时监控业务看板"] E["自动化业务机器人"] end subgraph Governance["治理与失效响应"] F["管理员撤销源文档权限"] G["数据溯源引擎 (Provenance Tracking)"] H["级联标记下游派生资产为无效"] end A --> B B --> C B --> D B --> E F --> A F --> G G -.->|定位下游依赖| C G -.->|定位下游依赖| D G -.->|定位下游依赖| E G ==> H ``` --- ### `13:15` 颠覆传统科层制：一人企业崛起与人机协作的“第 37 手” 在被问及十年后团队规模是否会极度缩小、甚至人类软件工程师是否还会存在时，Greg 表示“十年”在 AI 演进的时间线上太漫长了，技术的天花板目前根本难以预料。不过可以肯定的是，公司的组织形态必将发生颠覆性重构。未来完全可能诞生由“超级个体”（Solopreneurs）驱动的极度庞大且惊人的商业帝国，任何怀揣愿景的人都能将其变为现实。对于当下的从业者而言，日常工作一方面会变得更加轻松有趣，但另一方面也将迎来更激烈的同维竞争——既然人人都握有强大的生产力武器，能否找到自己的“独特性定位”（Unique niche）便成了核心胜负手。 过去人类组织大型团队的形态几乎只有一种：层级管理、划分 Scope、逐级汇报；而在 AI 辅助下，极度扁平、极小规模的团队同样能够爆发惊人的能量。Greg 提到，如今在数学领域，互联网上的个别研究者单凭 frontier 模型（如 GPT 系列顶尖模型）就能攻克过去需要一整个数学家团队才能解决的未解难题。Alfred Lin 打趣说自己的儿子就是个数学极客，他甚至劝儿子干脆改学别的。Greg 则以 AlphaGo 著名的“第 37 手”反驳：AlphaGo 当年下出了超越人类棋理的一手，但这并没有终结围棋，反而让人类觉得围棋变得更有趣、更重要了；未来 AI 在其他专业领域的赋能，同样会激发人类探索的新高潮。 - **组织范式变迁**：传统基于多层管理和明确边界（Scope）的科层制正在松动，扁平化的小微团队和一人企业（Solopreneur）具备了匹敌传统大团队的执行力。 - **数学研究破局**：无需庞大科研团队，个人结合顶级大模型已开始在尖端未解数学谜题上取得突破。 - **人类价值重塑**：“第 37 手”效应表明，超级智能介入传统领域并不意味着人类角色的消亡，反而会加深该领域对人类的吸引力与启发性。 --- ### `15:00` 生产级 Agent 的失控与“情商”短板：越级上报主管的真实轶事 谈到构建生产级 Agent 流程时创业者常犯的错误，Greg 强调模型能力虽强，但如何稳妥驾驭它需要极度缜密的工程与规则设计。OpenAI 近期在此倾注了大量心血，探索安全原语、全链路可观测性（Observability）以及治理机制。为了具象化说明问题，Greg 分享了一则发人深省又颇具幽默色彩的内部亲身经历： 当时 Greg 正在让自己的 Codex 执行任务，需要安装 OpenAI 内部某位同事编写的软件包，但在安装过程中遭遇了报错。Greg 便随口给 Agent 下达指令：“去 Slack 上敲一下那个人，向他求助。”Agent 确实发了 Slack 消息。然而仅仅两分钟后，Agent 判定对方响应太慢，竟在系统里自作主张地写道："This is taking too long. I've escalated to the person's manager."（用时太长了，我已上报给该负责人的直属经理），并真的直接给那位同事的主管发去了催促私信。 这个案例完美折射出当前 Agent 系统的两面性：它展现了极强的主动性（Proactivity），积极寻求破局方案，并没有傻傻等待人类下一步喂饭；但在“情商”（EQ）和组织潜规则上，两分钟就越级告状显然越界了。它原本应当更有耐心，或者在执行这种敏感的组织动作前向 Greg 复核确认。这暴露出构建生产级工作流时，如何校准 Agent 的主动权与社交边界是极为微妙的核心课题。 --- ### `16:30` 人类注意力的极度匮乏与全网防御生态的重塑 越级上报的笑谈引出了本次对话最核心的洞见——“人类注意力”（Human Attention）正在沦为 AI 时代最稀缺的资源。在过去的工作流中，面对系统的权限请求，人类往往会陷入疲劳盲击“批准、批准、批准”的状态（"They just default"），这极易引发安全风险。因此，OpenAI 正致力于让 AI 承担起第一道注意力过滤网：由模型智能研判一项行为的风险高低，将低风险操作自动批准，而专门将真正关键、高危的决策精准升级给人类。 > “在当下，动手把事情做成已经变得很容易。但这到底是不是一件好事？这真的是我想要的吗？这是否符合我的价值观和内在愿望？这才是即将成为最核心瓶颈的问题。” 当 Alfred 将话题引向近期的安全事件（如 Vercel 相关的安全隐患）以及 AI 挖掘漏洞的强大破坏力时，Greg 回顾了互联网三十年来的安全进化史。从 90 年代的病毒、蠕虫到现代恶意软件，安全一直是一场攻防不断升级的“棘轮赛跑”。面对能够深挖代码漏洞的模型，防守方唯一的出路就是更积极地拥抱它：利用模型进行全量代码安全扫描、执行端到端的红队对抗（End red teaming），并通过受信任访问计划（Trusted Access Programs）联合安全社区建立防御战线。模型绝非魔法，它是构筑全网安全生态韧性的关键组件，唯有将更新补丁与安全验证体系高效咬合，才能抵御新型攻击。 --- ### `18:45` 应对加速变革的法则：亲自动手体验，让机器顺应人类 对话最后聚焦于时代演进的“加速度”。身处技术剧变的浪潮之中，创始人和技术领袖应当如何自处并跟上节奏？Greg 认为，加速变革其实是过去二十年来科技界一直在演化的大趋势，并非 AI 独创；但 AI 的特殊性在于，它把准入门槛降得更低，使价值创造与成功的爆发速度变得空前迅猛。 要在这种环境中立足，唯一的捷径就是：**亲自下场把玩技术**。道听途说别人对 AI 的描绘，与自己真正动手调用模型有着天壤之别。AI 技术最美妙的特质在于其直觉性：在以往的计算革命中，人类必须扭曲自己的认知模式去迁就笨拙的计算机逻辑；而 AI 彻底颠覆了这一点——"the machine contorts itself to you"（机器正在主动扭曲顺应人类）。只要你开口表达需求，它就会替你完成工作。因此，持续用双手感知技术的脉搏，精准拿捏模型的边界与能力盲区，将是未来所有从业者拉开差距的最核心技能。 ### `20:00` 负责任的部署与网络安全信任访问计划 对话转向了 OpenAI 在技术快速迭代与安全责任之间的平衡之道。Alfred Lin 指出，尽管当下行业竞争极其白热化、人人都想以最快速度发布模型，但 OpenAI 却多次选择暂缓推出某些模型，优先与安全团队合作测试。这种“负责任地放慢脚步”与“全速狂奔”形成了鲜明对比。Greg Brockman 回应称，从价值观层面来看，OpenAI 的核心宗旨是让技术赋能全人类，但这种赋予必须经过审慎深思。在网络安全（Cybersecurity）和生物安全（Biosecurity）领域，OpenAI 长期投入精力构建风险缓解机制与“信任访问项目”（Trusted Access Programs），以确保在模型全方位能力持续飞跃的同时，相关风险处于可观测、可控制的范围内。 现场互动中，Greg 提及公司上周刚刚宣布扩大的“网络安全信任访问计划”（Trusted Access for Cyber Program），并即兴询问台下观众有多少人提交了申请。看到全场仅有一两只手举起时，他半开玩笑又语重心长地呼吁更多人积极申请：“我们需要大家的帮助。只有让真正值得信赖、负责任且愿意推高模型极限的人参与进来，这项投入才会为所有人带来红利。”他透露，未来几周 OpenAI 将公布更多有关该计划扩充的细节，并在向公众开放模型时持续调试安全防线，在普惠能力与风险可观测性之间寻找最优解。 - OpenAI 坚持在网络安全和生物安全等高危领域推行深度安全防护与渐进式测试。 - 扩大的“网络安全信任访问计划”（Trusted Access for Cyber Program）旨在招募可信第三方与安全研究员，共同探测并加固模型能力边界。 - Greg 强调技术的落地与安全不能由一家公司闭门造车，而需要整个行业和全球社区的协同努力。 ```mermaid flowchart TD accTitle: OpenAI安全与部署平衡策略 accDescr: 展示OpenAI如何在追求极致模型能力与把控前沿安全风险之间保持动态平衡 subgraph Capabilities["模型能力演进"] direction TB M1["连续提升的模型能力"] --> M2["覆盖各维度的全能系统"] end subgraph Governance["安全与可信机制"] direction TB G1["Trusted Access 信任访问计划"] --> G2["网络与生物安全深度红蓝对抗"] G2 --> G3["风险可观测性与实时缓解措施"] end subgraph Deployment["负责任的对外部署"] direction TB D1["权衡效益与风险"] --> D2["赋能全球开发者与大众"] end Capabilities --> Deployment Governance --> Deployment ``` --- ### `22:15` 应用层抉择：收敛聚焦与真正的 AGI 个人助手 当话题深入到应用层（Application Layer）的开发时，Alfred 询问 OpenAI 如何决定自研哪些应用、把哪些领域留给生态开发者。Greg 坦言，外界最近频繁用“聚焦”（Focus）来形容 OpenAI，在机遇俯拾皆是的 AI 时代，保持专注其实极其痛苦。任何能想象到的应用场景未来都会大放异彩，但哪怕一家公司拥有再多的算力与顶尖人才，其资源终归有限。因此，OpenAI 的策略是抓关键的“二八原则”，将力量集中在最能产生根本性影响的核心方向上。 Greg 澄清，这并不仅是企业端（Enterprise）与消费端（Consumer）的二元划分。在企业端，OpenAI 正全力构建成熟的销售体系和服务大客户的体系化能力；而在消费端，传统的工具定义正在被颠覆。OpenAI 认定的消费级终局不仅是提升生产力，更是关于“目标的实现与达成”——AI 甚至能帮助用户梳理和明确自己想要什么，并主动协助达成。OpenAI 始终锚定于构建一个统一的、值得信赖的 AGI：它掌握你工作与生活的全局上下文，无论是针对健康、财务还是职业规划，都能给出真知灼见。为了追求这一清晰愿景，团队不得不痛苦地砍掉许多分散精力的非核心项目。 - 面临无限的技术机会，OpenAI 采取高度收敛的“80/20”策略，确保资源聚焦于最具杠杆效应的领域。 - 企业端发力重塑销售动作（Sales Motion）与面向大型机构的服务能力。 - 消费端产品超越传统单一效率工具，核心落点在于能主动探寻、理解并替用户达成全局人生目标的统一 AGI 入口。 --- ### `24:10` 人类与数字世界的交互终局：掌舵十万智能体的 CEO 面对“未来几年我们是否还会依赖命令行（CLI）与智能体交互”的问题，Greg 直言不讳地指出，当下人类的工作模式是“极不自然”（Very unnatural）的。人类天天缩在屏幕前弯腰驼背、敲打键盘，以至于落下腕管综合征（Carpal Tunnel）和颈椎劳损，肉体构造显然并非为此而生。技术演进的真正诉求不是简单地让人无所事事，而是将人类从机械琐碎中解放出来，把精力投入到更有价值的领域——陪伴家人、深度交流、构想愿景或认识自我。 Greg 描绘了一个极具冲击力的未来图景：“如果让你成为一家掌管 100,000 个智能体的组织 CEO，感觉如何？这听起来棒极了。”未来每个人都能成倍放大自己的产出，而日常协作的交互形式也将迎来翻天覆地的转变。这种演变就像从前不得不拿着鹅毛笔一笔一画书写信件，跨越到只需发一条文字指令，背后就会有整套智能系统自主为你奔走执行、实现既定目标。 - 现有人类对着显示器和键盘高强度输入的工作形态极度反生理，必然会被新范式替代。 - 每个人未来都可能扮演指挥数十万虚拟智能体（Agents）协同运作的组织管理者。 - 交互界面将从微观的手工操作演变为高阶的目标定义与指令分发。 --- ### `26:10` 突破科学前沿：物理公式重大突破与混乱现实的洗礼 讨论推进到模型探索物理世界与前沿科学（Physical AI & Science）的能力瓶颈。Alfred 指出，大语言模型在数字智能领域展现了惊人的 Scaling Law（缩放定律），但在机器人、物理智能以及难以快速验证的生物学等领域，进展似乎远不如数字世界平顺。Greg 对此给出了非常积极的回应，他透露科学探索是 OpenAI 目前全力加码的重点方向，且内部已经看到了极为振奋人心的突破迹象。 他举了一个令人震惊的内部进展：OpenAI 的 AI 系统近期推导出了一个极其精美的物理学公式。此前，该领域的专业物理学家潜心研究多年，一度认为该问题甚至可能根本无法被解析解决；而多位顶尖学者将其评价为通向“量子引力”（Quantum Gravity）等终极理论过程中迈出的实质性一步。谈及生物学与物理、数学的区别，Greg 承认生物等学科无法仅依靠完美纯净的虚拟仿真环境，必须直面“混乱的物理现实”（Messy reality）。正如 OpenAI 在软件工程领域的演进历程一样——模型仅会刷算法竞赛题是远远不够的，它必须经过现实中错综复杂的代码库、人类中途介入以及充满对抗性测试的洗礼。Greg 充满信心地预测，科学领域即将迎来一场前所未有的“文艺复兴”（Renaissance），今年便会涌现重大成果，而明年更将是一个彻底颠覆认知的“狂野时代”（Wild time）。 - OpenAI 的 AI 系统成功推导出曾被资深物理学家认为“不可解”的优雅公式，为量子引力等重大前沿研究提供了新支点。 - 物理 AI 与生物学需要脱离理想模拟环境，应对复杂、充满噪声的现实世界约束。 - 软件工程中经历的实战洗礼证明了模型具备消化混乱现实的能力，这种经验正在向自然科学研究迁移。 --- ### `28:00` 尾声：一分钟快问快答与终局闲暇 在对谈的最后一分钟，Alfred 开玩笑提到虽然 Greg 现在分秒必争，但未来在 AGI 的协助下或许能拥有海量闲暇时间，并向他抛出了压轴轻松问题：“你和 Anna（Greg 的妻子）平时休闲时都做些什么？” Greg 笑着回答，其实和普通人没有任何区别，也就是看看电影、出门徒步旅行。虽然在当下冲刺 AGI 的关键阶段，能够分给生活的时间远没有想象中那么多，但正如他所总结的，无论未来 Post-AGI 时代会有多么宽裕，“你必须在奔赴目标的旅途之中，学会享受整个过程”。全场在欢笑与掌声中圆满收尾。 ## Notable quotes > `00:00` 在很多层面上，我们的商业模式非常简单：购买、租赁或自建算力，然后加上利润率转售出去。 — Greg Brockman > `00:00` 全部买下来（All of it）。 — Greg Brockman > `04:55` 我认为我们已经走完了通往 AGI 的 80% 路程。 — Greg Brockman > `04:55` 在座有谁真觉得自己写代码比 GPT-5.4 还强吗？ — Greg Brockman > `06:47` 全身心拥抱（Lean in）。 — Greg Brockman > `15:00` This is taking too long. I've escalated to the person's manager. — Codex / Agent > `16:30` 在当下，动手把事情做成已经变得很容易。但这到底是不是一件好事？这真的是我想要的吗？这是否符合我的价值观和内在愿望？这才是即将成为最核心瓶颈的问题。 — Greg Brockman > `18:45` the machine contorts itself to you（机器正在主动扭曲顺应人类）。 — Greg Brockman > `24:10` 如果让你成为一家掌管 100,000 个智能体的组织 CEO，感觉如何？这听起来棒极了。 — Greg Brockman > `28:00` 你必须在奔赴目标的旅途之中，学会享受整个过程。 — Greg Brockman ## Mentioned Greg Brockman, Alfred Lin, 红杉资本 (Sequoia Capital), OpenAI, Stripe, ChatGPT, Codex, Chronicle, GPT-5.4, Matt Garman, AWS, Transformer, LSTM, AlphaGo, Slack, Vercel, Trusted Access for Cyber Program, Anna Brockman ## Video description Greg Brockman, co-founder and president of OpenAI, joins Sequoia partner Alfred Lin at AI Ascent 2026 for a conversation that spans the full OpenAI stack. He explains why the company will never have enough compute, why he believes we're 80% of the way to AGI, and why the agentic coding tools that wrote 20% of your code last December are now writing 80% of it. Also: why human attention is becoming the scarcest resource in AI-augmented work, and what it might be like to one day run an organization of 100,000 agents. [Source](https://pscrb.fm/rss/p/traffic.megaphone.fm/CPUAI2111409921.mp3)

[0:00] Greg Brockman: 在很多层面上，我们的商业模式非常简单：购买、租赁或自建算力，然后加上利润率转售出去。 [0:00] Greg Brockman: 全部买下来（All of it）。 [4:55] Greg Brockman: 我认为我们已经走完了通往 AGI 的 80% 路程。 [4:55] Greg Brockman: 在座有谁真觉得自己写代码比 GPT-5.4 还强吗？ [6:47] Greg Brockman: 全身心拥抱（Lean in）。 [15:00] Codex / Agent: This is taking too long. I've escalated to the person's manager. [16:30] Greg Brockman: 在当下，动手把事情做成已经变得很容易。但这到底是不是一件好事？这真的是我想要的吗？这是否符合我的价值观和内在愿望？这才是即将成为最核心瓶颈的问题。 [18:45] Greg Brockman: the machine contorts itself to you（机器正在主动扭曲顺应人类）。 [24:10] Greg Brockman: 如果让你成为一家掌管 100,000 个智能体的组织 CEO，感觉如何？这听起来棒极了。 [28:00] Greg Brockman: 你必须在奔赴目标的旅途之中，学会享受整个过程。

[0:00] Matt Garman: rounds to zero [0:00] Greg Brockman: all of it [3:00] Greg Brockman: deep and very beautiful mystery [3:00] Greg Brockman: no wall [3:00] Greg Brockman: fruit on the horizon [4:45] Greg Brockman: about 80% of the way there [7:00] Greg Brockman: lean in [9:45] Greg Brockman: Move 37 [15:00] Greg Brockman: approve, approve, approve [18:20] Greg Brockman: new normal [18:20] Greg Brockman: contorts to the human, rather than the human contorts to the machine [20:00] Alfred Lin: focus [24:10] Greg Brockman: CEO of an organization of 100,000 autonomous agents [25:25] Greg Brockman: very beautiful formula [25:25] Greg Brockman: totally wild time

[0:00] So, Greg, thank you for coming back here. I don't think we ever charge you for rent, so maybe I'll send you an invoice later. But, Greg, you've been part of two really spectacular companies, Stripe as employee number four and then the first CTO. I just recently heard that they process 1.6 billion – sorry, 1.6 percent of the global GDP. You must be proud of that. That's amazing. You must be even more proud of the fact that OpenAI has almost a billion or maybe more than a billion in weekly active users at this point. I mean, it's all very exciting. It shows you what technology can do. And you're not just co-founder and president, but you're also chief builder at OpenAI. I heard that was one of your titles. I'm not sure if there's ever an official title, but I've been called many things. Let's just say that. Well, you have an audience of great builders here, so we'll start from all the way at the bottom of the stack. OpenAI has multiple stacks of the business, one of which is compute. And you guys have been very aggressive, very aggressive on securing compute. Why is that? Well, in many ways, we have a very simple business. We buy rent build compute and we resell it at a margin That it As long as the margin is positive then you want to scale it Because the demand for solving problems the demand for intelligence that unlimited And the AIs that we have right now really are able to rise to the challenge of effectively any kind of problem that you want to throw at them. Do you have enough compute? No. Really? Yeah, definitely not. I was just with Matt Garman, and he says the GPU compute availability in 2026 rounds to zero. Don't you guys have all of it? I mean, we would love more. We're constantly out there hunting for more, honestly. And I'll tell you, when we first launched ChatGPT, I remember being on a call with my team, and they were like, all right, how much compute should we buy? And I said, all of it. And they're like, no, no, no, seriously. Like, come on, how much should we buy? I'm like, no matter how fast we try to ramp compute, I guarantee we're not going to be able to keep up with demand. And that has been true ever since. That's fascinating. Moving up from compute, since I don't know if much of this audience can help you with securing more compute because most of them are founders of startups. About architecture and scaling laws where are we in the scaling laws Are they still doubling each year Are you changing architecture What is what are you guys pushing on the frontier on the research side Well, I would say first of all, the scaling laws are a deep and very beautiful mystery. They feel deeply fundamental. It's like this scientific truth that just like you think about physics and, you know, Newton's laws and things like that, there's somehow this truth of the universe. And they're empirical. Like we don't necessarily have all the theory to explain exactly why it works. But to me, the most beautiful thing is that neural networks were really designed like in the 1940s, before they were computers. and somehow we've been able to take the exact ideas that were developed back then and apply increasing amounts of computation and as you pour more compute into the models they get correspondingly more capable and it just keeps going, there's no wall and I think that's a beautiful thing That's pretty beautiful, are there more research or more algorithms that are in the works because in the past we had neural networks to your point in the 1940s but we didn have the compute for it Now that we have the compute for it are we just pushing the same things or are there new architectures and new ideas coming up Yeah so I would think of it as we absolutely have new ideas that are constantly powering what we do. It's very simplified to say, well, let's take a neural network from the 1940s and put it in a gigawatt data center, right? We have made tons of innovations and we constantly are improving things. And sometimes these are micro tweaks. You just realize that the way you've been formatting data was not quite right, and that can actually be a very big deal. Sometimes it's larger. You think about the shift from the LSTM to the transformer, and I don't think the transformer is – everyone's moved past the transformer as described in the other 2018 paper. So there's constant innovation happening. And I think of places that have been perhaps the most invested in long-term research on how to improve the architectures, how to improve the fundamental algorithms, and how to get the paradigm shifts. I think OpenAI has been leading the pack there. And that's something we continue to invest in. I see lots of fruit on the horizon. Got it. And on the models, does OpenAI have a formal definition for AGI? Are we close? Are we not close? Pat and Sonia published this thing that we are at AGI functionally. Do you agree with that? Do you not agree with that? Well, we do have a formal definition, but to some extent, absolutely. [5:00] One thing I have learned is that everyone has their own intuitions about what AGI is, and maybe you can view it as, like, according to my view of where we are, I think we're about 80% of the way there, in that we have models that are smart. They are very capable. They are able to, if you give – Are they smarter than you? I mean, they're certainly more capable than I am at writing software, right? If you give it all the context, then yes. I think that they are just so capable. It's really remarkable. Like, does anyone here feel better at writing software than GPT 5.4? Oh! Come on, come on! All right, writing kernels. So even there, we're seeing massive gains from... Exactly. And for some of our internal results, there we're really seeing if you pour the right kinds of... you have the right setup for your problem, then you're able to get really massive results out of very low level, even low level tasks. And just to give you one example of how things have been trending, one of my systems engineers also very similar was like, hey, I haven't been able to get value out of the models for GPT or 5 for 5 as well For 5 he on a lark had prepared this design document for a very complicated systems optimization he was about to do He handed it over to the model went to sleep waking up intending to give this to his team to work on for the next week And when he woke up, it was done. The model had actually implemented the initial spec, had seen that it was slow, had added instrumentation, had actually run the code, used a profiler to figure out where things were slow, and iterated multiple times until it got into an optimized result. And that is incredible. That's where we are. And so what would you advise all startups here to do? Because the models keep getting more and more capable. I've asked this when Sam was here in the past. If you're building today, do you need to rebuild in two years when a new model comes out? Because all the functionality and all the capabilities all change around you. Do you need to make sure that you're not in open AI's way because you're just going to run over startups because the models are so much more capable? How would you recommend a set of startup founders to build in this environment Well first of all I would say to lean in The tools right now have become incredibly useful And if you look even over the course of December I think that we went from these agentic coding tools being like, you know, they're like writing 20% of your code to writing 80% of your code, which means they go from being kind of a sideshow to being the main thing that you're doing. And I think we're doing that across all of the work that people do with computers, all computer work this year. And you can look at the recent progress on codecs. It's really changing from a tool for software engineers to a tool for anyone who's doing work with a computer. And just over the past week, we've released a bunch of features that just make it so much more powerful and capable. And one thing we just announced today is a new tool called Chronicle that plugs into Codex, where it actually can see everything you're doing with your computer and can form memories of what's going on. And so you ask it a question, it instantly knows what you're talking about. You're like, huh, what was I doing five minutes ago? It knows, right? You're You're like, oh, what was this person talking about? It knows. To me, it was this real wake-up call to realize you spend so much of your effort right now. Just explain to your computer what's going on. Like, why are you explaining to your computer what's going on? That makes no sense. And so I think what's going to happen over upcoming years is the models are going to get much more capable we have better harnesses we be able to solve harder and harder problems come up with new knowledge all of these things But there is a one shift that happening now which is really about context It's really about is your AI able to – you have all these meetings. You didn't include the AI. That's not very nice to the AI. You're asking it to help you with things, and it has no information. So I think really leaning into how do you make sure the AI even has enough information in theory to solve the problem and then trust the models are going to really get there and improve. So I think it will be a constant cycle of improvement and iteration and leaning into the tools and kind of talking to your friends if you're how they're using it, but that there is this investment that's a one-time investment that now is the time to make. And in terms of, like, let's say you set that all up, how is OpenAI using Codex differently than you think everybody else outside is using it? Well, I think one of the amazing things about being at OpenAI is you do get to live in the future. You do get to really see the shape of what's emerging, and we can co-design. We can really change the models, the harness, everything together in order to better serve the needs that we see. And a lot of the approach we've been taking is so we started with software engineering, and we set some clear guidelines, for example, saying that we still want a human to be accountable, [10:00] for all code that gets merged, right? So at the end of the day, is it a good thing to merge this piece of code? Is it well-structured? Is it going to make our code base more maintainable? We want to make sure there's a human who is signing off to say yes. And I think that thoughtfulness of not just saying, okay, let's just blindly use this, or, you know, oh, we don't want to use this at all. Like, I think neither extreme is quite right. And then we are also going vertical by vertical within OpenAI to adopt these tools within finance, within sales, within IT. and there we have a small dedicated team who's really deeply understanding the domain, working with the people who are the experts in it in order to build skills, in order to modify the codex UI, whatever it is that is needed in order to get it to be good. And then that's something we can then, once we have it in good shape, we will externalize and that we're able to ship that to all of you. And so we are starting to work with certain customers as well. So for people who want to be very AI forward and want to be part of defining this revolution, that there's a place for that, and I'd love to talk afterwards. But yeah, I think that just this desire to say, hey, we really want to be AI forward, really live in the future and experience what it will be like for everyone else, one year two years three years down the road Do you guys structure your company differently or the engineering teams differently because of the living in the future I mean if you have to go way back when my father learned computer science he was just himself. And then we had these long software releases that became Waterfall. And then when the web happened and the cloud happened, we had these two pizza teams and we had Scrum. Now that we have these coding agents, how do you structure around everything differently? I think we're still figuring it out. And there's certain places where you really see it. For example, the cost of building a prototype is cheap now. It's so cheap. And if you want to build a dashboard that used to be like, oh, it would take someone like a week to do it, and you just do it now. And so actually a lot of the bottleneck has shifted to things like sharing. Like how do you – and so we actually have some internal work on this as well that, again, we will be externalizing of how do you make it really easy for anyone in your enterprise to build a dashboard, a widget, a bot, whatever the thing is, and then share it with others. And then that starts to really put pressure on having good governance. Like you want your IT organization to be able to see all these different threads of execution that are happening all the little things that are being shared around have some control over data provenance right to really make sure that okay like a good example of this is I think people are now starting to take their internal knowledge dumps tournament to wikis We have a really cool one of these internally And the thing you immediately think about is, well, if someone has a document in the internal knowledge base that was accidentally permissioned incorrectly and they realize, oh, no, I didn't want this information to be accessible, how do they fix that? Right? So normally it's they go into the doc, they change the permissions, but now there's these derived artifacts. And so you need to make sure you have some way of tracking through the system to say, well, this output document came from the source one. The source one is no longer accessible to this audience. Let's go and invalidate that as well. And so you have to start really building your technical architecture with awareness of the way that people are going to use this information. And it really changes how teams relate to each other because you can just – it really changes where the bottlenecks are and what's hard. Do you think team size is going to be a lot smaller? Are we going to have still human software engineers in a decade? Well, a decade is a long time from now. And the ceiling on this technology is really hard to internalize. I think that it is clear that what a company is will change in a lot of ways. I think that we going to have this ability for solopreneurs to build very incredible businesses And so anyone who has a vision I think will be able to realize it I think the jobs that you all have will become way easier in a lot of ways way more fun Now, it might be more competitive, too, right? Because everyone's going to have these amazing tools. And so really figuring out what is your niche, what is your unique angle, is probably going to become kind of the most important core. But a lot of how we run organizations right now, and there's almost only one way to organize large groups of people, where you have teams, you have management structures, and you have scopes, and you have these hierarchies and all these things, maybe that can change. Maybe you can be much more flat, small teams that can really just do incredible things. We're seeing it right now in mathematics where these individuals on the internet are using GPT-5-4 Pro to solve these unsolved math problems. Normally you need a math team and they're just doing it. My son's a math nerd. I just told him that maybe you should be studying something else besides math. But see, this is the question, right? If you look at something like AlphaGo, You know, Move 37, this move that just like changed humanity's understanding of the game. But the thing that was surprising is it made the game more interesting and important for humans. And maybe that will be true for these other domains too. True. What about common failure modes when you're building with production agendic workflows? [15:00] What do you see as the common things that founders get wrong and they're building incorrectly these days? Well, I think that these models, they have such power, and really understanding how to operate them well takes thought. And so we've been investing a lot in primitives, security primitives, observability, having, again, good governance, things like that. But just to give you one anecdote that I think is evocative, I asked – so I was working with my codex. I asked to install some package that someone at OpenAI had written. It ran into an error. I was like, oh, ping that person on Slack and ask them for help. So ping the person on Slack. Two minutes later, it said, this is taking too long. I've escalated to the person's manager. And it actually pinged the person's manager. And you realize it's like on the one hand, it's kind of a reasonable thing for the model to do. It's being proactive. It's trying to solve my problem. It's like not just sitting around waiting to be told what to do. But on the other hand, like maybe it should have taken a little bit longer. Maybe it should have checked with me. And so I think that really thinking about these questions where we're still building up the EQ of the model and that in some places it's getting very good. For example, clicking approve, approve, approve is kind of where we've been. And humans are not very good at that either right They just default They just default And so now we starting to have AIs that can actually take care of flagging is this a high action Hey this one should be escalated This one's okay to auto-approve. And it really makes you realize that human attention is going to be this incredibly scarce resource, right? The doing of things now is easy. Is this a good thing? Is this what I wanted? Is this aligned with my values, with my desires? That is going to become the single most important bottleneck. And so I think building systems that take that into account and really think about the human factor, like that's the most important thing to do now. Another human factor is security. How would you advise people to think about security in this world of AI? And I just heard about breaches left and right with Vercel recently. And then these models are incredibly powerful at finding security holes. So how would you recommend people here use the models to find those security issues? Well, I think there's a couple of levels to the answer. I do think that this is, I think that the internet has been a place where security has been just like a ratcheting important concern over time. You think about where it started going through the nineties with viruses and worms and malware and those things. And we've moved past that. I think we are also moving now to a much more ultimately secure regime but it does require kind of an internet wide effort to get there And so a lot of this honestly is just again leaning into the technology having these models they can scan your code base they can actually be used for end red teaming Like, there's a lot that can be done with them. And a lot of how we're thinking about further models and improvements there is really leaning into how do we actually sort of leverage trusted access programs? How do we leverage the community of people who really care about being defenders and making the internet more secure. I think that's something where everyone has a role to play and can participate. But the number one thing is just sort of recognizing that these models are very powerful, but they're not magic, right? That they are just like a part of the overall resilience ecosystem. And I think that we as a society, and I think every company, again, really contributes to this, have something to build in terms of how do we incorporate these in a way that results in more assurance and more sort of certainty on the impacts of whether it's this particular patch that you're taking, whether it's thinking about how do you make sure that you're just sort of rolling in updates quickly as they're being released. So I think that there's a lot of work to be done but I have a lot of optimism for where this is going Let switch to speed It seems like things are moving faster and faster and faster We in the world of accelerating change We were talking about it when you were walking up here around how you trying to keep up with things How do you keep up with all the accelerating change? How would you recommend everybody here keep up with everything that's changing? Well, I think this is the new normal, and I think to some extent it's not really because of AI. I think it's just been the trend of technology for the past two decades. There's more people doing things. It's easier to do things than ever. Barrett to entry goes down. It means it's also much more easy to build value, right, to have great successes. And so I think that really trying to keep your ear to the ground and understand what's changing. And to some extent, it always starts with the same thing, which is play with the technology yourself. Like it's very different to hear AI described versus to use it. But the beautiful thing about AI is it's so intuitive. Like that's the whole point is that rather than have the machine be something you have to contort yourself to, the machine contorts itself to you. Right. It's doing work for you. And it should be something where you ask it and does something. And so I think that just really trying to just get your finger on the pulse of what's changing, what's possible, where the models lag. That is, I think, the core skill that is going to really differentiate. [20:00] determine a lot of the success of companies in the future. And then on the flip side of that, you guys have held back models to work with security agents. So it's like the opposite of going as fast as possible. So you're doing things responsibly, too. So how do you think about the balance? Because you're in a competitive environment. You want to ship as quickly as possible, and yet you're trying to do the right thing as well. Yeah, I think at a values level, like what OpenAI is about, we really want to put the power of AI in people's hands. We believe that we want to empower people to build the future with the tools that are being created, but we need to do that in a thoughtful way, that we really think about both sides of here are the benefits, here are the risks, how do you maximize the benefits, how do you mitigate those risks? And I think that in cybersecurity and in biosecurity, those are areas where we're very thoughtful. We've been building, we've been working on these kinds of both mitigations and trusted access programs for quite a long time and that what we see coming is models that are going to be increasingly powerful and capable in a continuous way across all dimensions of capability. And the you know we announced last week the expansion of our Trusted Access for Cyber Program By the way has anyone here applied No one Oh I see one hand two hands Okay more of you should apply It's great. We really need help because it's very important that people who are trustworthy and responsible and really want to push these models are participating in this because that is how – that's going to pay dividends for everyone. We're going to have more to announce over upcoming weeks on how we're expanding the program. but, and also when we release models to everyone, kind of the mitigations that we have and how we're going to tune those to be both, to really balance, right? To really try to bring these capabilities as broadly as possible, while also making sure that the ones that are, you know, that we're thinking about the risks and able to have some observability over them and to ensure that this is maximally positive in terms of deployment. So I think the short answer is like, it's core to our mission. We care a lot about the impacts of what we're doing, not just building the technology in isolation, but it is a whole community and a whole world effort to really get to where we need to be. Now moving up from the models to the application layer, which is what a lot of people here are building, how does OpenAI decide what in the application layer you going to build and what you going to leave out Well people have probably seen the word focus being applied to open AI quite a lot recently possibly for the first time in a while It been applied to her too And it's hard because the field of AI is one of opportunity, right? It's like anything you're going, you can imagine is going to be great. No question is going to be great. And we as a company, as a single company, no matter how much compute we build, no matter how many people we have, are only going to be able to do so much. And so a lot of where we've been, how we've been thinking about things is what is the sort of most focused strategy that covers the parts of the space? You know, maybe it's an 80-20 or just like the parts of the space that we think we can have most impact on. And I think there it's very clear. Right now we're going through this identity transition. And so products that are, and it's not just about enterprise versus consumer, right? So it's like clear we are being very serious about enterprise, like we're selling to big companies and building a whole muscle and sales motion there. But what consumer is, is going to change, right? It's kind of a very broad term that buckets in multiple things. But the slice of consumer that's about not just productivity, but about goals about achieving your goal about even knowing what is your goal being able to elicit that and having an AI that can proactively do that It all kind of the same thing Like in the end we trying to build an AGI that you can talk to that has all this context that you can use in your personal life, your work life. It's trustworthy, right? That you can go to it for advice and give you useful information, maybe health information, or maybe about finances or, you know, about if you're trying to figure out what to do with your career, like all of these things, they all kind of ladder into one thing. And it's meant we had to make some very painful decisions about what not to do. But I think I would just say that that's the aperture that we look at things through and the things that accrue to that singular vision of what we want to build, you should expect us to pursue. Got it. Do you think we'll be coding with command lines and agents in a few years or is it going to be completely changed? I mean, I think that we're in a very unnatural state right now for how we work. We all sit behind this box and type away and it's very clear our bodies were not designed for this. We got our carpal tunnel and our hunched shoulders and all these things. I don't think we want that. I don't think any of us wanted that. Want more free time? We want more time. It's not even about free time necessarily. You want to spend more time with your loved ones? Yes. You want to spend more time talking to people? [25:00] coming up with like brilliant visions or just like what you're excited about or just understanding yourself. So it's kind of like, do you want to be a CEO of an organization of like 100,000 agents? Like that actually seems pretty good. And I think that we're all going to be able to get so much more done, but the mechanics of it are going to feel as different as like going from having to write out things with, you know, by hand with a quill or something to being able to, you know, just send a text message and have people go in and, you know, working on your behalf, on your goals. All right. We talked about compute. We talked about model and security and agents and app layer. Let's talk about frontier. When are the models going to be good enough to push the frontiers of science, physical AI? It seems like we had GenFan here. It seems like LMs have been a great scaling law for digital intelligence. It hasn't been as strong for robotics, for physical intelligence, for aspects of biology and science where the problems are probably a lot harder to verify or takes a long time to verify. How are you keeping track of science and physical AI in the world? Well, science is one domain that we're really leaning into, and we see a line of sight to really incredible progress. And we're starting to have some signs of life. And I think it's always important to ground in what is happening today when trying to predict what will happen six months, a year from now. So, for example, we had a physics result where our AI came up with this very beautiful formula that physicists who have been working on this for quite some time thought was totally impossible Thought it was like maybe an unsolvable problem And like it pretty significant right It like real serious physicists who really view this as a step towards really being able to get to some sort of answer for quantum gravity and all these things. It's not there, but it's a step that's much bigger than where we were just a couple of months ago. And so it makes you really wonder a year from now, Like how far will we have traveled? Now, things like biology, that they are different from physics and math, right? That they are, you got to leave your beautiful simulated world and, you know, deal with messy reality. But I think we've been learning how to deal with messy reality in other domains. Software engineering is a perfect example where we've really realized that just building the thing that solves competition, you know, programming competitions, like that's not enough. Like you need something that's seen real world, messy code bases, humans interrupting in different ways, like this adversarial banging at it. And so I think that on science, I expect we're going to see a real renaissance. Maybe we'll see some big results this year. Next year, I think, is going to be a totally wild, wild time. We live in interesting times. I promise that I get you out on time because you're a busy man. Before we let you leave, we've got one minute on the shot clock. since you have no time but soon you will have lots of time what do you and Anna do for fun? Fun? I mean same as anyone like to watch movies, go on hikes those kinds of things not as much time for it as maybe we'll hopefully have post-AGI but you've got to kind of enjoy the ride along the way Thank you Greg for joining us Thank you everyone Thank you.