大公司实验很多，缺的是押上全部：Jeetu Patel 如何让 Cisco 全面转向 AI

Lenny opens by noting that Cisco's AI summit had guests including Jensen Huang, Sam Altman, Marc Andreessen, Fei-Fei Li and the CEOs of Intel and AWS. Patel says the team worried about human absorption limits across a 12-hour run of fireside chats, but attendees stayed engaged. His first takeaway is that the capabilities overhang is real: science is solving amazing problems while enterprises say they struggle with adoption.

His second takeaway is that AI is much easier to apply to obvious cases like coding, where Cisco expects a product to be 100% written with AI within a couple of weeks, than to other business functions, which need nuance and an understanding of how each business works. The third, prompted by Marc Andreessen and Kevin Scott, concerns falling birth rates and an aging population. Patel says that if a large share of people need care and there aren't enough people to give it, the survival of humanity depends on a successful AI.

With 90,000 employees, Cisco has to make innovation a choice, Patel says, one you can make every hour of every day. His first lever was clarity about what is and isn't up for debate, because in a large company someone can always say no, a kind of pocket veto. He adds that big companies do experiment, but when something works they keep hedging, whereas Cisco decided to go all in on AI, and reassured people that their job relevance depended on becoming dextrous with it.

The second move was changing how success was defined. Executives wanted to be general managers with their own sales, marketing, product and engineering silos, which would split a roughly $40 billion product business into many small ones. Patel wanted a platform company rather than a holding company of acquisitions, where customers feel the same reliability, trust and design everywhere, loosely coupled but tightly integrated.

The third was a mental-model shift about five and a half years ago: operate in an open ecosystem rather than a walled garden, even partnering with competitors. If a customer picks two vendors, he says, you owe it to the customer to invest in the other vendor's success too, because customer success flows back to you. Together with building great products and being AI-first from the top down, these principles shaped the transformation.

Patel calls Cisco a critical infrastructure company for the AI era and names three things that could hold AI back. There is an infrastructure constraint in power, compute and network bandwidth; a trust deficit, since models are non-deterministic and hallucination is a bug when you need predictable systems; and a data gap, as public human-generated data runs out and growth shifts to enterprise, synthetic and machine data, where Cisco can play a large role.

On the infrastructure side, he explains that models outgrew single GPUs, then servers, then racks, so the operative word is "connected". Nvidia and AMD make GPUs while Cisco connects them, and data centers hundreds of kilometers apart must now run as one coherent cluster, fully in sync during a training run. Cisco supplies networking, optics, safety and security, observability and a data platform for this.

Asked what isn't priced in, Patel recalls telling Ray Kurzweil about a world of indefinite lifespans and being told that exponentiality happens across several dimensions at once. He argues that today's view of AI as a productivity and aggregation tool is the tiny tip of the iceberg, since AI will generate original insights that don't exist in the human corpus. An OpenAI forward-deployed engineer told his team to stop thinking of Codex as a tool and think of it as a teammate, which changed how they used it.

He insists the safety and security risks are real and that guardrails and checks and balances matter because AI is in service of humans. He rejects the polarized narrative that we will either have nothing to do or that AI is useless, and asks instead how to make life better, solve diseases and eradicate poverty. He mentions Elon Musk's "house cat" best-case image and praises the depth of thinking at Musk's company despite critiques.

On raising his daughter, Patel says they did not keep technology away from her, which he calls a passive and perhaps lazy choice that worked out. At almost 15 she told him she had a core set of convictions she'd hold even if everyone disagreed. His takeaway is to expose kids to technology while focusing on durable values like kindness, humility and work ethic. Lenny notes that Anthropic's constitution for Claude shows a similar parallel between raising a person and steering an AI.

Patel learned the idea of "permission to play" from Box CEO Aaron Levie. Before entering a new category, ask whether it is logical for your company to build it and whether you have a route to mass distribution. Without that, product and sales blame each other, while with it the dollars spent on product get outsized returns. For Cisco, networking GPUs and building a trusted system for AI is a natural extension of 40 years of doing it for other infrastructure.

He also credits Levie with the lesson that persistence and stamina beat intellect. Lenny argues that as code gets cheap and everyone launches, the question of why you will win matters more, and Patel agrees that human judgment and intuition about which problems to solve matter more as AI slop floods the market.

At Cisco's size, ideas for new markets arrive constantly and Patel says he declines 90 to 99 percent of them, because focused expenditure of calories gives outsized results. His example is consumer tech: Cisco can do it, but lacks a distribution channel in its DNA, so he prefers the areas where it can operate from strength.

Patel names Cisco CEO Chuck Robbins as one of his favorite humans and a model of confidence. When a press article gave Patel too much credit, Robbins told him that if you don't care who gets the credit, you go a lot farther. Patel draws from this the importance of knowing what you're good at and assembling a team around what you're not.

He keeps mentors and former colleagues close: he has had dinner every six weeks for six years with Aaron Levie and Jeff Queisser, talks to Robbins several times a day, and still sees his first California boss, Rick Devenuti, who remains his coach, every two weeks. He also bought a place next to friend Jeremy Burton. He says such relationships need to be treasured.

Patel says that the absence of scale is harder than scale, since a three-person startup needing five more people must raise money while a 30,000-person organization can reallocate dollars. What scale does bring is lossy communication, the telephone game. When he took the job, a board member advised him to treat the company's story as the most foundational exercise, not marketing, and to remain its custodian rather than letting it cascade through layers.

The hidden benefit was that it simplified Cisco's business. If he could not explain the story well enough, he asks, how could 20,000 sellers tell it to the market? He argues that the story is the reason you build the product, with products serving as evidence. Lenny cites Rippling's Matt MacInnis on keeping the intensity of an idea from dropping at each layer, and Patel uses a networking analogy: you don't want packet loss between you and the front line.

To operationalize it without overload, he says clarity of thought produces clarity of communication, so spend time with the team sweating the details and the "why". He recommends reducing asymmetry between the top and the bottom of the organization, giving as much context as legally allowed, and treating people like adults, including saying plainly when the company screwed up.

Patel says he fundamentally disagrees with the management-book rule of praising in public and criticizing in private. Once there is enough trust, he says, a team can critique and debate openly and focus on problem-solving rather than posturing. In private he uses the time to build trust, tell people he has their back, and not be stingy with words; constant pleasantries make dashboards look green while the business grows only 1.5%.

He cautions to watch your tone, a thing he says he still works on, since conflict is a necessary condition of business but productive conflict depends on established trust. Compliments given only as part of a feedback sandwich can feel hollow, he adds, referencing Ben Horowitz.

Coming from the apps world, he learned that in infrastructure you don't always get the glory but you always get the blame. A healthcare customer told him the reason it doubled down on Cisco is that when infrastructure fails, people die, for example missing dialysis or surgery. Patel says infrastructure leaders must orient on their ecosystem's success and be comfortable with others getting credit, which echoes Robbins' lesson.

Patel says he spends little time with people whose energy he doesn't vibe with, however successful, and that ego manifesting as insecurity is off-putting. Life, he says, is more fun when you love the people around you, and he is friends with most of the people at the AI summit.

He tells the story of his mother, who died two and a half years ago after eight weeks in the hospital. Having grown up with a difficult childhood and an abusive father, he had become a parent to her. At 1 a.m. she woke to see him crying and said she had no idea he loved her so much. His lesson: people don't know what's in your mind, so don't be stingy with words, in life or at work, while keeping praise genuine.

For people outside Silicon Valley, Patel's first piece of advice is that the platform you choose and the problems you pick largely determine your path. Harder problems attract better people, and business is a team sport, so a hard, important problem draws the best team. His second is that you can teach and learn many things but not hunger, so find what you are intrinsically hungry about, accepting that about 30% of any job is stuff you will hate.

He tells of a guide named Raj at the Taj Mahal in 2017 who spoke 12 to 14 languages and tried to learn a new one every year to honor visitors. Patel thought the man was smarter than everyone on his executive team yet earned about $10 a day, because he lacked access to a platform. Luck plays a part, so seek out platforms, be prepared when luck arrives and don't be intellectually lazy.

He recommends building a community of people vested in your success, not carrying the world alone, and asking for help non-transactionally by adding value first for a long time. For young people entering a hard job market he says to stay persistent and have stamina. Lenny adds Arnold Schwarzenegger's book title, Be Useful, as the simplest advice for success.

Patel shares a six-part framework, in descending order of importance, and says that if you don't have all six you don't win. Timing is the most important and the one you control least; market comes next, since you must be able to go after a large market a chunk at a time; then team, meaning a well-rounded group whose members complement each other. A great market with a mediocre team pulls you up, while a bad market with a great team drags you down.

Product is the soul of a company, and he thinks it is unethical to sell a mediocre one. Brand is fifth: a mentor warned him never to join a company that has lost its brand, because lost trust rarely returns. Distribution is sixth, since building it does not mean they will come. He uses the list to ask whether it's the right time to double down or whether a trend might not manifest for seven years.

To tell a mega trend from a hype cycle, Patel asks whether most people can understand what it does in its ultimate form without a PhD. He treats AI as a mega trend and cites web3 as a hype cycle whose use case he couldn't grasp. When timing is wrong, he says, you put the idea on ice, as Steve Jobs did with the iPad. With AI moving so fast, he tells his team to imagine six months ahead and prepare for that world.

He warns that experience can bias you, so pair it with inexperience, and calls it foolish to stop hiring entry-level people because it shuts out fresh ideas. He recommends Clayton Christensen's The Innovator's Dilemma and The Innovator's Solution and Ben Horowitz's The Hard Thing About Hard Things. He says AI's learning tools were essential to his taking on the Cisco product role, and his motto is that stamina trumps intellect.