Cognition CEO Scott Wu 探讨他们 15 人的工程师团队如何每人使用五个 AI Devon，目前四分之一的 PR 由 AI 提交，预计年底将超过一半。
Cognition CEO Scott Wu discusses how their 15-engineer team uses five AI Devons each, with a quarter of PRs already AI-committed, expecting over half by year-end.
要点 · TL;DR
AI 将指数级增长，成为最大的技术变革，并将工程师转变为架构师。 AI will grow exponentially, becoming the biggest technology shift and turning engineers into architects.
将 Devon 用作初级工程师，处理明确定义的异步任务，并配以人工监督。 Use Devon as a junior engineer for well-defined async tasks with human oversight.
未来的工程将涉及多个 AI 代理并行工作，从而增加对程序员的需求。 Future engineering involves multiple AI agents in parallel, increasing the demand for programmers.
核心观点 · Key points
AI将是我们一生中最大的技术变革，没有硬件分发限制，呈指数级增长。 AI will be the biggest technology shift of our lives, growing exponentially without hardware distribution constraints.
工程师的角色将从砌砖工转变为架构师，专注于高层次的方向和问题定义。 The role of engineers will shift from brick layer to architect, focusing on high-level direction and problem definition.
随着AI变得更强大，编程将变得更加重要，使更多人能够构建软件。 Programming will become more important as AI gets more powerful, enabling more people to build software.
Devon最适合作为初级工程师使用，异步处理定义明确的任务，由人类提供监督。 Devon is best used as a junior engineer, handling well-defined tasks asynchronously, with humans providing oversight.
软件工程的未来涉及与多个AI智能体协作，每个智能体并行处理不同的任务。 The future of software engineering involves working with multiple AI agents, each handling different tasks in parallel.
对于AI产品，粘性而非护城河是关键，通过积累知识和融入工作流程来构建。 Stickiness, not moats, is key for AI products, built through accumulated knowledge and integration into workflows.
反共识 · Contrarian takes
由于杰文斯悖论和软件需求的增加，AI将导致更多程序员，而不是更少。 AI will lead to more programmers, not fewer, due to Jevons paradox and increased demand for software.
基础智能已经足够；重点应放在教授模型现实世界工程的特性上。 Base intelligence is already sufficient; the focus should be on teaching models real-world engineering idiosyncrasies.
最大的挑战不是模型能力，而是设计人机协作的产品界面。 The biggest challenge is not model capabilities but designing the product interface for human-agent collaboration.
像‘快速行动’和‘雇佣优秀人才’这样的创业建议是陈词滥调，但在执行深度上常被低估。 Startup advice like 'move fast' and 'hire great people' is cliché but often underestimated in execution depth.
最反直觉的教训是，将基础工作做到极致，比如招聘，可以成为差异化因素。 The most counterintuitive learning is that doing the basics extremely well, like hiring, can be a differentiator.
AI编程的未来可能根本不需要查看代码，而是专注于产品层面的交互。 The future of AI coding may not require looking at code at all, focusing on product-level interaction.