20VC: Atlassian CEO on Why Agents Won't Replace Interfaces | Why Enterprises Won't Trust Chinese Models | Why The Jury Is Still Out On The Benefits of Specialised Models with Mike Cannon-Brookes

在本期 20VC 节目中，主持人哈里·斯特宾斯（Harry Stebbings）与 Atlassian 创始人兼首席执行官迈克·坎农-布鲁克斯（Mike Cannon-Brooks）展开了一场广泛的讨论，内容涉及企业软件的演变、AI 代理的影响、设计哲学以及在快速变化的技术格局中的企业战略。 对话从挑战一个前提开始：AI 代理将完全取代人类与数据源及传统软件应用的互动。迈克认为，虽然代理将处理日常的功能性任务，但涉及品味、协作和创造性业务工作流程的领域仍然需要人类的指导。他对人工智能的“魔法按钮论”不予认可，并坚信无头软件（headless software）是“无脑的”。相反，在 AI 时代，用户界面和设计变得更加关键。Atlassian 正在大量投资于基础设计挑战，以解决新的范式，例如代理如何在协作平台中出现、管理人类与 AI 的共同创造，以及建立内容溯源和问责制。 在转向战略和技术决策时，讨论涵盖了围绕开放模型的辩论以及地缘政治考量。迈克指出，尽管客户对中国模型存在一些抵触情绪，但企业对开源权重模型仍然非常开放。他对比那些声称从零开始构建专有模型的公司保持怀疑，并暗示许多公司实际上是在微调现有的开源权重选项，以实现更好的性价比。在运营方面，迈克强调了财务纪律和技术成本管理的重要性，重点介绍了 Atlassian 内部的 AI 网关团队及其在模型路由和混合方面的工作，以优化速度、质量和成本。 展望未来，迈克分享了他的观点，即由于大量的资本和人才投入，模型推理的成本将急剧下降。他认为，科技行业正从信息不对称的世界转向想象力不对称的世界，在这里，将 AI 应用于独特业务问题的能力是终极的差异化因素。在反思 Atlassian 的核心战略时，迈克强调了持续维护全球最佳企业上下文图谱的重要性，该图谱融合了一家公司的决策、历史文档、代码和社交图谱，以为人类和代理保持数据的最新性和可操作性。 在节目接近尾声时，讨论触及了作为创始人的个人和心理现实。迈克分享了他的能量来源——与聪明人一起解决难题——并反思了在数字不断加速的世界中保持专注的挑战。他建议创始人“对自己发动和平”，并强调在技术生产快节奏中留出空间进行深入、缓慢思考的重要性。最后，迈克强调了 Atlassian 对持续产品演进和设计工艺的长期承诺，确保公司在未来十年的软件创新中继续保持领先地位。

In this episode of 20VC, host Harry Stebbings sits down with Mike Cannon-Brooks, founder and CEO of Atlassian, for a wide-ranging discussion on the evolution of enterprise software, the impact of AI agents, design philosophy, and corporate strategy in a rapidly changing technological landscape. The conversation begins by challenging the premise that AI agents will entirely replace human interaction with data sources and traditional software applications. While agents will handle pedestrian, functional tasks, Mike argues that areas involving taste, collaboration, and creative business workflows will still require human guidance. He dismisses the "magic button thesis" of AI and expresses strong conviction that headless software is "brainless." Instead, user interface and design are becoming even more critical in the AI era. Atlassian is heavily investing in foundational design challenges to solve new paradigms, such as how agents appear in collaboration platforms, managing human-AI co-creation, and establishing content provenance and accountability. Moving into strategic and technological decisions, the discussion covers the debate surrounding open models and geopolitical considerations. Mike notes that while there is some customer pushback regarding Chinese models, enterprises remain very open to open-weight models. He remains skeptical of companies claiming to build proprietary models from scratch, suggesting that many are actually fine-tuning existing open-weight options to achieve better price-performance ratios. On the operational side, Mike stresses the importance of financial discipline and technological cost management, highlighting Atlassian's internal AI gateway team and their work on model routing and blending to optimize speed, quality, and cost. Looking forward, Mike shares his perspective that the cost of model inference will drop precipitously due to massive capital and talent deployment. He believes the technology industry is shifting from a world of information asymmetry to imagination asymmetry, where the ability to apply AI to unique business problems is the ultimate differentiator. When reflecting on Atlassian's core strategies, Mike emphasizes the ongoing importance of maintaining the best enterprise context graph in the world, which blends a company's decisions, historical documents, code, and social graphs to keep data fresh and actionable for both humans and agents. Towards the end of the episode, the discussion touches on the personal and psychological realities of being a founder. Mike shares his source of energy—working with smart people on hard problems—and reflects on the challenges of maintaining focus in a world of constant digital acceleration. He advises founders to "wage peace on yourself" and stresses the importance of carving out space for deep, slow thinking amidst the fast pace of technological production. Finally, Mike underscores Atlassian's long-term commitment to continuous product evolution and design craft, ensuring the company remains positioned to lead through the next decade of software innovation.