在一个由五到七人组成的团队中，只要有一名 AI “超级用户”，就能使整个团队的 AI 采用率和使用率提升 56%，Atlassian 的 Avani Prabhakar 告诉 TNW。她说，Atlassian 将每周使用 AI 最复杂功能超过 40 次以改变工作流程的人定义为超级用户，并会追踪他们是谁。
One AI “super user” in a team of five to seven people lifts the whole team’s AI adoption and usage by 56%, Atlassian’s Avani Prabhakar told TNW. Atlassian counts someone as a super user if they use AI’s most sophisticated features more than 40 times a week to change a workflow, she said. It tracks who they are.
Prabhakar 是 Atlassian 的首席人员与 AI 赋能官，该职位于 4 月宣布设立，此前她担任首席人员官。她已在公司工作七年半，常驻悉尼。她 2001 年在美国通用电气 (GE) 开始职业生涯，当时从事业务而非人力资源工作。后来她在英国航空工作了七年，也曾在富士施乐任职。
Prabhakar is Atlassian’s chief people and AI enablement officer, a role it announced in April, after she served as chief people officer. She joined the company seven and a half years ago and is based in Sydney. She started her career at GE in the US in 2001, in the business rather than in HR. She later spent seven years at British Airways and worked at Fuji Xerox.
她在阿姆斯特丹举行的 HumanX 大会上接受了 Cristian Dina 的采访，录制了 TNW Podcast 节目，此前该播客已推出过 Boomi 的 Steve Lucas 和 Remote 的 Job van der Voort 两期节目。完整对话见下方视频。
She spoke to Cristian Dina at HumanX in Amsterdam for the TNW Podcast, after episodes with Boomi’s Steve Lucas and Remote’s Job van der Voort. The full conversation is in the video below.
是人的变革，而非技术项目 Prabhakar 表示，企业应将 AI 视为一场人的变革。部署 AI 是容易的部分，让人们喜欢上它也很容易。大公司在安全与治理、以及成本方面容易陷入麻烦。
A people change, not a tech project Companies should treat AI as a people transformation, Prabhakar said. Deploying it is the easy part, and so is getting people to like it. Big companies get into trouble on security and governance, and on cost.
因此，Atlassian 的每个团队都会获得一份所谓的 “AI 工作协议”。协议规定了在该岗位上 AI 可以在哪些场景使用、哪些场景不得使用。例如，产品研究人员不得将公司信息输入 AI 工具；招聘人员会被明确告知哪些决策由自己做出、哪些由 AI 做出。
So every team at Atlassian gets what it calls an AI working agreement. It sets out where AI can be used in that job and where it cannot. A product researcher cannot put company information into an AI tool, for example. A recruiter is told which decisions are theirs and which are the AI’s.
钱包制取代 Token 无限制用法在成本控制上，她表示，每位员工都会获得一个 “AI 钱包”，即根据其角色分配的月度额度。一旦用完额度，申请追加额度需要审批，并须说明用途及其改进的工作流程。TNW 曾于 7 月报道过工程师的 AI 预算。
Wallets instead of token maxing On cost, every employee gets an AI wallet, a monthly allowance that depends on their role, she said. Once they hit the limit, they need approval for more and must explain the usage and the workflow it improves. TNW reported on the engineers’ AI budget in July.
“不要奖励 Token 使用量，只需监控 Token 使用量，因为那会导致非常错误的行为。”Atlassian 的 Avani Prabhakar 表示。她说，更大的节省来自 Teamwork Graph（团队协作图谱），即 Atlassian 的内部上下文层。它将知识、对话、人员、资产和代码汇聚到一个图谱中。让 Claude 仅在该图谱中搜索，使 Token 使用量减少了 48%，准确率提升了 44%。
“Do not reward token usage, just monitor token usage, because that drives very wrong behavior.”Avani Prabhakar, Atlassian The bigger saving came from the Teamwork Graph, Atlassian’s internal context layer, she said. It pulls knowledge, conversations, people, assets and code into one graph. Sending Claude to search only that graph cut token use by 48%, and accuracy rose by 44%, she said.
不要强制推行 将 AI 使用作为与绩效挂钩的强制指标是“非常糟糕的开端”，Prabhakar 说，因为人们会开始钻空子。企业也不应像过去设立数字化转型办公室那样，在一旁单独组建一个首席 AI 官团队。在任何人尝试 AI 之前要求提供商业案例，会扼杀实验。
No mandates Making AI use a mandate tied to performance is “a very bad start,” Prabhakar said, because people start gaming it. Nor should companies build a separate chief AI officer team on the side, the way they once set up digital transformation offices. And asking for a business case before anyone can try AI kills experimentation.
“ROI 是结果，不是入场费。”Avani Prabhakar先任务、后技能、再角色 Atlassian 按任务、技能、角色的顺序推进工作。她说，目前大家看到的 AI 收益都在任务和工作流层面。任何声称 AI 已取代某个角色的人，产出的都是“大量 AI 垃圾”。在技能方面，Prabhakar 领导一个约 13 人的行为科学家团队，Atlassian 还与斯坦福大学合作开展实验。一项实验发现，毕业生基于 AI 做决策的有效性比资深管理者高出 19%。
“ROI is an outcome. It’s not an entry fee.”Avani Prabhakar Tasks, skills, then roles Atlassian works through tasks, skills and roles, in that order. For now, everyone sees AI’s benefits at the level of tasks and workflows, she said. Anyone claiming AI has replaced a role is producing “a lot of AI slop.”On skills, Prabhakar runs a team of about 13 behavioural scientists, and Atlassian has partnered with Stanford on experiments. One found that graduates make decisions based on AI 19% more effectively than senior managers, she said.
Atlassian 还设立了约 30 个绿地项目，由五六人组成的团队运作。在这些团队中，工程师构建原型，设计师提交代码。在角色层面，公司新设了“设计技术专家”职系，面向因 AI 而工作方式改变的设计师。
Atlassian has also set up about 30 greenfield projects run by teams of five or six. In those teams, engineers build prototypes and designers commit code. On roles, it has created a new job family, the design technologist, for designers whose work has changed with AI.
开放文化 Atlassian 拥有约 1.2 万名员工，分布在 3000 多个城市，且无办公室强制令。员工可居家办公、在办公室办公或在任何地点办公。公司多写文档、少做演示，发送 Loom 录像而非总开会。她说，这种书面化、异步化的习惯为 AI 提供了上下文，使其能更好地读写。
Open culture Atlassian has about 12,000 employees in more than 3,000 cities, and no office mandate. Staff can work from home, from an office or from anywhere. It writes more and presents less, and it sends Loom recordings rather than always holding meetings. That written, asynchronous habit gives AI the context to read and write better, she said.
客户零号 Prabhakar 现在将 60% 到 70% 的时间用于与客户交流。她的任务是打造一份“客户零号”故事：在 Atlassian 内部开展实验、编写行动手册，并将其推向大型企业。
Customer zero Prabhakar now spends 60% to 70% of her time with customers. Her brief is to build a “customer zero” story: run the experiments inside Atlassian, write the playbook, and take it to large companies.
她给毕业生的建议源于与女儿的交谈，女儿想攻读法律。她说，追随让你感到快乐的事物，并在其同时学习 AI。
Her advice to graduates comes from talks with her daughter, who wants to study law. Follow what gives you joy, she said, and learn AI alongside it.
“不要出于恐惧做职业选择。” Avani Prabhakar
“Don’t make your career choices out of fear.”Avani Prabhakar