Tony Fadell：1.0 产品要靠有见识的品味拍板，别把判断交给 AI

Fadell says the debate over a physical keyboard on the iPhone was the most heated and longest-running argument inside Apple. One camp wanted to win over BlackBerry users. The other pointed out that only 1% or 2% of mobile phone users had a BlackBerry at the time, so the real question was what the other 98% would want.

Having worked on virtual keyboards since General Magic in the 1990s, Fadell knew the limits of single-touch screens. Multi-touch had only existed on a big table, so there was no consumer-ready form to user test. His team set up tests of typing speed and error rates on hardware and virtual keyboards, and spent months iterating across hardware and software.

By the end he convinced himself the virtual keyboard was not going to be a hardware problem. It was not as good as a hardware keyboard, but it was good enough, and others came around. The data showed pros and cons on both sides, and Fadell says it ended as a classic case of data versus opinion, with Steve Jobs telling holdouts to get on board or work on another project.

Fadell says a new category product has few analogs, so most decisions have to be opinion-based, and a very small group must own them. Relying only on data either gives you an undifferentiated product or just a pile of data. He calls it a benevolent dictatorship: the team states the vision and accepts that nobody knows what they don't know until they ship.

The problem is hardest in consumer products, because consumers judge the whole experience, from marketing and discovery to key features and usability, and they will not give real feedback until they can spend their own money. Fadell notes that many corporations hire consultants and run user studies that lack this full context, which he describes as covering oneself with data. A good product manager, he says, accepts the risk and takes the heat if the call is wrong.

Fadell defends micromanagement in a specific sense. Early in his career he thought everything mattered and drove people nuts, but he later learned to micromanage only the few details that matter to the customer, manufacturing or long-term vision. He means micromanaging the decision and the data behind it, not the day-to-day operations, and sometimes a crisis or a system-level change across many layers requires it.

Fadell is visibly pained that the Nest smoke alarm has been discontinued. He calls it one of the toughest products his team ever made, with so many constraints, and says it was the best product in its space for a decade. He thinks it became an orphan, or stepchild, inside Google, because it takes a great deal of care to maintain something that crystalline and nobody wanted to put in the effort.

He argues Nest would be central to a home AI assistant today, since AI needs context and well-placed sensors can pick up comings and goings and audio without invading privacy. Fadell says AI was a foundational piece of Nest from 2010, but they could not call it the Nest AI thermostat in 2011 because people would have freaked out. He believes the vision was simply about 15 years early, and says he is now receiving business plans for a Nest 2.0.

Asked how he decides what is worth building, Fadell says he always starts from pain: pain people have now or will soon have. Such pains often come from an unintended consequence or a technology limitation that evolved but never got revolutionized. For the thermostat, the pain was choosing between comfort and savings, since heating and cooling is about half of an energy bill, and programmable thermostats were too arcane to use.

His answer was a thermostat that learns, using AI, when you are home and what temperatures you like. It was a much better-looking device that cost five to six times more, at $249, but he says the pitch was that it could save $800 to $1,200 a year and pay for itself in a year or two. Fadell stresses that the product was also a reinvented system, including how it was installed and sold through installers, just as the iPod was the iPod plus iTunes and the iTunes music store.

The second half of the formula is why now. The iPod came when portable mass storage, new lithium batteries, digital music and low-power ARM processors converged. The iPhone came with multi-touch, faster processors, Wi-Fi everywhere, 3G on the horizon, digital cameras and YouTube, which together made it very different from a BlackBerry, which he calls a texting machine.

Fadell says the iPod was not big enough at first. The first two generations only sold to Mac loyalists, who were under 1% of the market, and sales spiked and then died each quarter. It took the third generation, with Windows support and the iTunes music store, for it to take off.

Jobs had said Windows support would happen over his dead body, since the iPod was supposed to help sell Macs, so a skunk works effort ran behind the scenes. Fadell says the same happened with the stylus, which Jobs never wanted on the iPhone or iPad. When it was added, it became a major feature for certain professionals, artists and hobbyists. He argues you should keep working on things you can see on the horizon even if the opinionated leader dislikes them.

His rule is to make the product, fix the product after customer feedback, then fix the business by getting the margins right. The first iPods and iPhones did not make money, and Nest needed two generations of smoke detectors and several thermostat generations. Fadell quotes the idea that you only fail if you stop, and that otherwise it is called learning. He adds that without Windows connectivity the iPod would have effectively cost $3,000 once you bought a Mac, and that Apple, nearly bankrupt in 2001, might not have survived without the iPod.

Fadell says builders live in the context of their own product, but customers do not know the product exists. Your marketing, website, ads and earned and owned media must put the product in their context, with visuals and words that make them feel understood before they ever try it. Word of mouth from trusted early adopters is the best driver, and each wave of adopters needs different language, as in his updated Crossing the Chasm graphic in the book.

He gives the example of pushing the iPod into Europe around the fourth generation. Apple ran the same marketing as in the US, and sales did not work, because the messages did not resonate and there was no installed base for word of mouth. Fadell says new regions often require rethinking the marketing and starting that word of mouth deliberately, and even in software the awareness cannot be compressed by distribution alone.

Asked about the "1,000 songs in your pocket" tagline, Fadell says it was genius when he first heard it, and that at the time marketing was separate from engineering with Jobs acting as the hub. The iPod shipped in about five to six months, starting in April and shipping at the end of October. He contrasts that with the iPhone, which had a two-and-a-half-year development and naming and trademark issues with Cisco's iPhone.

Fadell argues that OpenAI has been a technology demo that went viral and only later started building product teams and product marketing. He contrasts that with Anthropic, which he says is now valued higher with higher revenue, and says thinking about marketing early forces you to think about the product. He prefers writing the press release first because a customer only has room for three or four key features, and cutting two of three tentpole features means you can no longer sell the product.

On the evolution of the product management role, Fadell says product management sits between marketing, sales, distribution, manufacturing, engineering and support, stitching them together. He disagrees that a single prompt can replace understanding those functions, because each represents a clear point of view the customer cares about. Roles may shift, but someone must interpret across them.

He uses the Claude source code leak as an example, as he describes it. At the time, he says, Dario Amodei said 90 to 100% of Anthropic's code was written by Claude, but when real software architects looked at the leaked main loop, Fadell says they were horrified by how brittle and unreadable it was. His point is that an agent can write code that works and passes tests, but it also has to be secure, maintainable and possible to roll back, so you still need humans in the loop.

Fadell says that you get short-term gain for very long-term loss, which is technical debt, and the same applies to product management if you lack good marketers, architects and manufacturing experts. The better approach, he says, is to architect first, lock it in and have the AI work on limited sub-segments, and to do more prototypes to build an informed gut. He compares fast software to H&M versus a luxury brand and says Flighty is luxury software whose original version could not have been vibe coded, because highly differentiated version-one products have no model to copy.

Fadell says storytelling is how humans pass on information and commit to doing things, and compares comedian Dave Chappelle, who can weave a 20-minute story to a punchline, with punchline-only comics. Great teachers also explain why a subject matters, which is what makes it meaningful. When the story comes from the depths of the product itself, like the care in Nest Protect, people want more from that brand.

He learned selling from his father, a Levi's salesman, who sometimes talked customers out of buying something because another product, even a competitor's, was a better fit. Fadell cites Jobs' line that the best marketing just tells the truth. He says Jobs honed the iPhone story every day for two and a half years, pitching it to smart friends until, on stage, he had done it 10,000 or 100,000 times.

Fadell says technology-led teams talk about the what and skip the why, and the why is where the story lives. He notes that Nest would set up a virus of doubt by asking people what they spend on heating and cooling. He also suggests studying infomercials for their psychological and emotional techniques, such as showing pain points, then dialing the hype back and using them with truth. The host jokes about creating the infomercial first as a version of working backwards.

Asked what the next iPhone with AI could be, Fadell separates the long term from the near term. He thinks we will still need a display unless we get brain-computer interfaces or lasers into the retina, and that the best visual display is a smartphone-like slab, possibly foldable. He says the Humane approach of projecting onto your hand does not help, since it is just a different screen, not a better one.

Fadell wants to flip the order of interaction. Today it goes from tapping and swiping, to keyboard, to voice as the tertiary option. Long term, he says voice should be primary, then keyboard, then touch, which is what he always wanted to do with Nest. Voice has always been added at the end, as with cars, so nobody really uses it, but good voice input with real intelligence and memory behind it could change that.

In the near term, he expects devices to look much like today's smartphone, because people do not yet trust AI enough to hand over app interfaces. He compares this to General Magic and notes agents that have deleted people's source code. He also doubts consumers will pay $20 or $200 a month unless the product is incredible, and compares it to paying for full self-driving and still waiting 15 years later.

Fadell says he was building hardware when it was out of fashion in 1995 and 1996, and people in the Valley told him it was all about the internet. He argues that the next level of software requires the next level of hardware, and that AI needs data centers and edge compute. He says that now investors say software-only companies are worthless because anyone can vibe code them, and that they only fund companies with atoms in the plan, to which he says, "Duh."

He points to Waymo as an electric car with many sensors that is an incredible hardware and software platform. The host says Snap's Evan Spiegel made a similar point about needing hardware to survive in software. Fadell says he has seen enough cycles, from the bottom falling out in 1999 onward, to expect Gen 1 and Gen 2 companies, and a shift to those who understand product instead of selling platforms for people to figure out.

Fadell lists portfolio bets at Build Collective: Simbe Robotics, which does retail inventory, Great Parrot, which uses AI and cameras for recycling, AI for spotting textile defects, Recursion on drug design, and a clean agricultural fuel company in Central America. He also says he invested early in Groq and Cerebras. He says he invests in deep technologies that fundamentally change the product and unseat incumbents, advises on product, operations and marketing, and is the inaugural designer in residence at MIT.

Fadell urges product designers to be grounded in real principles, and not to try to addict users or chase money by tearing at the fabric of society. He says that as people have families, they think about the benefit to society as a whole, and that users will feel and reward that, as with Apple's privacy stance, though he calls it a double-edged sword. He recalls Jobs shutting down a suggestion to add porn to the iTunes video store, and says he wishes more leaders were that clear, and objects to companies turning personal connection into a chatbot product.

On the iPhone's impact, Fadell says social media was an unintended consequence, and compares digital consumption to junk food, which needs nutrition labels, warnings and regulation. He says platform companies such as Apple and Google could do much more with consumption tools and information, and hopes AI assistants help people take control. The iPhone, he says, is like a refrigerator: you can put in good or bad food, and still open it every five seconds.

Fadell closes by pointing people to buildc.com and urging builders to read, hone their craft and make great products that make him say he loves them, such as Flighty. His final message is that people can use AI tools, but should not cognitively surrender to the machine. He asks builders to make better stuff than he or his teams can, since they now have better tools.