THE SHORT VERSION
- Cisco president Jeetu Patel gave 27-year-old Padmanabhan Krishnamurthy the task of refactoring 98% of Cisco's legacy code within 24 months, with unlimited AI tokens.思科总裁 Jeetu Patel 让 27 岁的 Padmanabhan Krishnamurthy 在 24 个月内重构公司 98% 的遗留代码，并给他无限 AI token。
- Krishnamurthy joined through Cisco's acquisition of EzDubs and had been working five levels below the role Patel says he was suited to.Krishnamurthy 因思科收购 EzDubs 入职，此前在组织中比其经验低五个层级的岗位上工作。
- Patel told him he could go nuts, and said he wants more hires in the same mould who challenge the system and the status quo.Patel 称已告诉他可以「放手干」，并表示希望找到更多这样敢于挑战现状的年轻人。
The assignment这项任务
Cisco's president and chief product officer has given a 27-year-old employee an unusual mandate: refactor 98 percent of the company's legacy code within 24 months, with four weeks as a proof of concept, and unlimited AI tokens to do it. Fortune reported that Jeetu Patel described the hire as an energizer bunny and told him to go nuts.思科总裁兼首席产品官交给一名 27 岁员工一项不寻常的任务：在 24 个月内重构公司 98% 的遗留代码，以四周作为概念验证，并给他无限 AI token 去做。Fortune 报道称，Jeetu Patel 形容这名员工像「永动机」，并告诉他「放手干」（go nuts）。
Tom's Hardware reported the same details and named the employee as Padmanabhan Krishnamurthy, known as Paddy, a former startup founder who joined Cisco through its acquisition of the translation startup EzDubs.Tom's Hardware 报道了同样细节，并点明员工名为 Padmanabhan Krishnamurthy（Paddy），是一名前创业者，因思科收购翻译初创公司 EzDubs 而入职。
How he got the job他怎么拿到这份工作
The assignment came out of a conversation rather than a reorganisation. Tom's Hardware reported that Patel found Krishnamurthy buried away in the organisation doing something unstrategic, five levels below where his experience suggested he belonged, and that in a meeting in Cisco's San Francisco office Krishnamurthy said he could rewrite the company's legacy code with a team of just two people.这项任命来自一次谈话，而不是一次组织调整。Tom's Hardware 报道称，Patel 发现 Krishnamurthy 被埋没在组织中做着「不具战略性」的工作，比他经验所对应的位置低五个层级；在思科旧金山办公室的一次会面中，Krishnamurthy 说，他可以用仅两人的团队重写公司的遗留代码。
That claim is what got him the job. According to the report, Krishnamurthy told Patel that even the most AI-forward people on Patel's own team were not thinking that way. Patel's response was to give him the mandate and the tokens, telling him he could go nuts because he was going to refactor code.正是这句话让他拿到了任务。据该报道，Krishnamurthy 对 Patel 说，连 Patel 团队中最「AI 优先」的人都不是这样想的。Patel 的回应是给他授权和 token，说既然他要重构代码，就可以放手干。
Why legacy code is the real target为什么真正的目标是遗留代码
Legacy code is the part of a software company that nobody wants to own. It works, it is expensive to change, and it holds business rules nobody fully remembers. A 24-month target to refactor 98 percent of it is less a coding exercise than a bet that AI tools can read, test and rewrite code faster than a team of engineers can, without breaking the business in the process.遗留代码是软件公司里没人愿意接手的那部分。它能运行、改动昂贵，而且承载着没人完整记得的业务规则。要在 24 个月内重构其中 98%，与其说是编程练习，不如说是在押注：AI 工具能比一支工程师团队更快读懂、测试并重写代码，同时不在过程中弄坏业务。
The proof-of-concept month matters more than the headline number. If a two-person team can show a working pipeline in four weeks, the project moves from a pitch to a programme; if it cannot, the 98 percent figure is a slogan. The report does not say what happens after the four weeks.概念验证的那一个月比标题数字更重要。如果两人团队能在四周内展示一条可运行的流水线，项目就从提案变成计划；如果做不到，98% 就是一个口号。报道没有说四周之后会发生什么。
What unlimited tokens actually buy无限 token 究竟买到了什么
Unlimited AI tokens are a strange benefit to advertise, and they reveal something about how the work is expected to be done. Token limits are the practical constraint on agentic coding: an agent that reads a codebase, writes tests and iterates burns far more tokens than a person typing into a chat window. Removing the budget makes the tool a utility rather than an expense to justify.「无限 AI token」是一种奇怪的福利宣传，也透露了这项工作将如何完成。token 上限是智能体编程的实际约束：一个读取代码库、编写测试并反复迭代的智能体，消耗远高于一个人在对话框里打字。取消预算，等于把工具变成一种公用设施，而不是需要论证的开支。
It also shifts the constraint elsewhere. If tokens are free at the margin, the limits become review capacity, test coverage and the cost of a mistake in code that other systems depend on. A company can buy as much model usage as it wants; it cannot buy back a broken release.它同时把约束转移到了别处。如果 token 在边际上免费，限制就变成评审能力、测试覆盖，以及一行代码出错时其他系统所依赖的代价。公司可以买下任意多的模型用量，却买不回一次坏掉的发布。
The hiring argument underneath背后的招聘论点
Patel's bet is also a comment on hiring. Fortune framed the story against a backdrop of youth unemployment rising as AI takes over entry-level work, with companies saying they plan to hire fewer graduates because AI can do the job more cheaply. The Cisco example runs the other way: a junior employee is given one of the largest mandates in the company.Patel 的这次押注也是对招聘的一种表态。Fortune 把故事放在一个背景里：随着 AI 接手入门级工作，青年失业率上升，企业称计划少招毕业生，因为 AI 做同样的事更便宜。思科的例子方向相反：一名初级员工被赋予了公司里最大的任务之一。
Patel said he wants more hires in that mould, people who challenge the system and the status quo and who energise the teams around them. That is a hard profile to interview for, and the story does not say how many such people Cisco expects to find internally or externally. It is one data point, chosen because it is vivid.Patel 说，他希望找到更多这类人：敢于挑战体系与现状、并能带动周围团队的人。这样的画像很难通过面试筛出来，报道也没有说明思科预期在公司内外找到多少这样的人。它只是一个被选中来讲的、鲜明的样本。
What a 98 percent refactor actually risks重构 98% 究竟在冒什么风险
Refactoring almost all of a large company's legacy code is not a single project; it is a series of migrations whose failure modes differ. Some systems are internal and forgiving, others carry billing, authentication or network control, and a rewrite that changes behaviour in one of those is an outage rather than a bug. The 98 percent figure is only meaningful if the remaining 2 percent contains the parts nobody dares touch.重构一家大公司几乎全部遗留代码，不是一个项目，而是一连串失败模式各不相同的迁移。有些系统是内部的、容错高；另一些承载计费、认证或网络控制，任何行为改变都是事故而不是缺陷。98% 这个数字只有在剩下的 2% 恰好是没人敢碰的部分时才有意义。
That is also why the two-person claim is the interesting part. Two engineers plus models can produce a lot of rewritten code, but they cannot personally hold the domain knowledge of every service. The test of the project is whether the AI pipeline can extract that knowledge from tests, logs and the people who still remember, which is a different capability from writing code quickly.这也是「两个人」这个说法值得注意的原因。两名工程师加上模型可以产出大量重写代码，但他们无法亲自掌握每个服务的领域知识。项目的真正考验是：这套 AI 流水线能否从测试、日志和仍记得来龙去脉的人那里提取这些知识——这与快速写代码是两种能力。
What is established, and what is not哪些已经确定，哪些还没有
Established: a 27-year-old employee, Padmanabhan Krishnamurthy, was given the task of refactoring 98 percent of Cisco's legacy code within 24 months; he joined through the EzDubs acquisition; Patel said he had unlimited tokens and told him to go nuts; four weeks were set as a proof of concept; Patel wants more hires of the same type. Not established: whether the proof of concept succeeded, what the 98 percent target covers technically, and how many other employees get similar mandates.已经确定的：27 岁的 Padmanabhan Krishnamurthy 被要求在 24 个月内重构思科 98% 的遗留代码；他因 EzDubs 收购入职；Patel 称给他无限 token 并让他放手干；四周被设为概念验证；Patel 希望招到更多同类人。尚未确定的：概念验证是否成功、98% 在技术上覆盖哪些范围，以及有多少其他员工获得类似授权。
For readers, the useful signal is that AI coding tools are now being used to change the maintenance backlog rather than to write new features. This site has covered a former GitHub chief executive building an AI coding forge, OpenAI's enterprise workflow partnership with ServiceNow and the scale of AI spending commitments. All three describe companies buying capacity to rework what they already run.对读者而言，有用的信号是：AI 编程工具正被用来改变维护积压，而不只是写新功能。本站报道过一位前 GitHub CEO 打造 AI 编程工坊、OpenAI 与 ServiceNow 的企业工作流合作与AI 支出承诺的规模。三者都描述了企业在购买能力，去改造自己已经在运行的东西。
Sources & reporting limits来源与核查范围
Follow Industry to find related stories on your next visit.关注产业动态，下次访问时阅读相关文章。
