Summary

戴乐’s core bet is not to add an app to a traditional guitar, but to digitize the full chain from strings and motion to sound and build a “third-generation guitar.” Sensors turn every fretting and plucking action into a standardized MIDI signal, while beginners can use LEDs and an easy-play mode to perform and sing within minutes; more advanced players can still practice real fingerwork. 牛亚锋 believes the rebuild could expand what was previously a niche market by 10x to 20x.
AI’s biggest change to smart instruments is turning a passive tool into a private teacher and music partner that understands the user. The system can learn which songs a user repeatedly chooses, how their tastes evolve, which chords they struggle with, how hard they play and even their emotional state, then dynamically recommend lessons, content and creative fragments. That depends on the hardware and App already being data-driven, not on simply adding an AI label. 牛亚锋’s long-term formulation is direct: “It will make smart hardware proactive.”
Models such as Suno have lowered the barriers to generation and stem editing, but pure software still lacks the bodily participation of performing, which is precisely where smart hardware fits. 戴乐 wants to break chord progressions, rhythms, drums and lead melodies into modular, schedulable fragments, allowing AI to supply the material stored in a professional musician’s head while leaving the choices and expression to the user. The end state is not one-click generation of masses of songs that have nothing to do with the listener, but “what you feel is what you play”—the user inputs an emotion and outputs music directly through the instrument.
The same digitized hardware platform can support PMF across regions, but content and functionality must be localized. 牛亚锋 characterizes China, Japan and South Korea as markets shaped by karaoke and singalong culture, while Western users may have broader exposure to instruments and a stronger emphasis on individuality and original work; the former therefore calls for more teaching and song content, while the latter leans toward MIDI, Cubase, DAWs and low-barrier creation. The product strategy can be a common hardware layer with differentiated Apps, or separate SKUs where necessary.
Retention comes not from forcing usage frequency, but from making the product an always-available setting for relaxation. 戴乐 iterates on features and content every 2 weeks; by 牛亚锋’s preliminary estimate, it may already have users in the 100K-plus range and total usage approaching 2M hours. His product test is blunt: “A good product, or a sticky product, must either get users hooked or help them relax.”
The three pillars of a decade in hardware entrepreneurship were surviving first through overseas markets, building Shenzhen supply-chain speed, and then entering a guitar category large enough to matter. The company made overseas markets its first base around 2019-2020, then used Shenzhen to accelerate the integration of 200+ innovative components; 牛亚锋 puts the traditional musical-instrument market at more than $20B, with guitars accounting for nearly half, or roughly $10B, and annual unit sales in the 10M-plus range. “If you’re doing hardware, you still have to come to Shenzhen”—because R&D, mass production and delivery have to be right from the start.
The financing hurdle for AI hardware has fallen, but the commercialization hurdle has not, and capital excitement can even trigger an unproductive race over parameters. 牛亚锋 started with a RMB150K entrepreneurship prize, whereas teams today may secure RMB15M or even US-dollar funding at the outset; his advice is to validate the pain point, PMF and real-world deployment as quickly as possible, because “investors’ money can’t burn forever.” 戴乐’s next phase still prioritizes the second-generation product, App and content, while adding brand expression, global offline experiences and differentiated creative capabilities for third-generation products.

Deep dive

1. A decade of entrepreneurship began with RMB150K—and a refusal to wait

牛亚锋 was born in 1993 in Luoyang, Henan, and comes from an engineering background. As a child, he dismantled every appliance at home that could be taken apart; in his freshman year of college, he joined a lab to study programming and electronics. He wanted to learn guitar but never did, and turned lowering the barrier to guitar playing into a science-and-innovation project.
The actual startup capital came from winning a China-US entrepreneurship competition and its RMB150K prize. As graduation approached, he debated whether to join a large company first and gain experience. His answer to himself was: “A job will still be there, but entrepreneurship can’t wait.” The company was registered, and the venture has now continued for 10 years.
李翔 said the direction initially received little recognition. The team started in a basement with a “student army,” and for the first 3-4 years relied mainly on entering startup competitions around the country to secure limited funding—enough, somehow, to keep R&D and the team running.
The experience ultimately condensed into three words: “real, passion, lasting.” At the end of the episode, 李翔 recalled that 牛亚锋 often played 张震岳’s 《再见》 at competitions during his student years, and 牛亚锋 played and sang it again. From sustaining the startup on competition prizes to selling products in multiple countries, “keep going without looking back” became the footnote to the decade-long journey.

2. The starting point for a “third-generation guitar” is turning every playing action into data

牛亚锋 describes the evolution of the guitar in three technological generations. Around 1850, wood and steel strings drove the rise of the acoustic guitar; in his account, around the 1920s Les Paul put an electric pickup into the instrument’s body, and effects pedals and amplifiers subsequently expanded its tonal range, supporting more than a century of diverse sounds and rock culture.
戴乐 defines its product as a fully digitized third-generation guitar. One side, aimed at beginners, uses pain-free silicone strings and touch detection; the other retains real strings. Sensors send the player’s actions into a controller, audio-synthesis system and acoustic system.
In quick-start mode, side-mounted LEDs show where to press. Pressing any fret can represent a chord, while plucking any string triggers strumming or fingerstyle; users select a song in the App and can perform and sing within minutes. In learning mode, the system can also determine in real time which chords were played correctly and which were not.
Digitization is not only about simplification. Traditional instruments lack a sensing layer, while 戴乐 can read every press, pluck and change in force. 牛亚锋 summarizes the mission as “reconstructing the way people interact with the world of music,” because the first-principles need is not to master difficult movements but to smoothly play songs one likes and express one’s emotions.

3. It cannot stop at being a toy—and need not reproduce traditional-instrument barriers for a small minority’s last-mile needs

Asked whether it is ultimately an instrument or a toy, 牛亚锋 says the third-generation guitar has moved beyond that binary. It uses a toy-like low barrier to get users started, while retaining real strings, fingerwork and an advanced progression path. The goal is neither to lose beginners 3 minutes in nor to make them practice for 3 months without being able to sing a song.
The product path runs from entry-level play to learning, advancement and creation. Techniques such as slides, tapping and harmonics will be added across generations, while sound quality and range will continue to improve. The second-generation product targets a more professional tonal experience, benchmarked against the sound of Yamaha products priced around RMB7,000-8,000, but delivered in a product accessible to a mass audience.
His trade-off line is roughly 90%-95% of users: raise the level of professionalism while lowering the interaction barrier. The specialized feel demanded above that level is somewhat like photographers insisting on the texture of film; most people recording daily life and travel may find Insta360 sufficient. Digital instruments do not need to recreate a universal barrier for a minority preference.

4. Real smart hardware is not Bluetooth connectivity, but analysis, decisions and proactive interaction

Looking back at “smart hardware” 10 years ago, 牛亚锋 says many products used the label mainly to sell more or sell better. Add Bluetooth, Wi-Fi and App connectivity, and a product was called smart, though in substance it remained closer to ordinary hardware—hence the element of “fake intelligence.”
With AI integrated over the past 2 years, hardware has begun to organize information, analyze it, think through it and make simple decisions. He expects sensor-equipped hardware to become broadly intelligent over the next 10 years, because models can understand the information collected and further determine how a device should act.
In smart instruments, hardware plus App plus traditional controls is still only 1.0. The key to 2.0 is not adding a chatbot, but having the system learn a user’s usual songs, aesthetic preferences and practice status, then proactively recommend, remind and accompany. The relationship gradually shifts from “people control hardware” to continuous interaction between people and devices.

5. Data makes AI a genuinely personalized music teacher for the first time

A traditional acoustic guitar cannot know that a user has been repeatedly practicing an F chord, while a digital instrument can record chords practiced, weak points, song sections and performance scores. AI can therefore identify weaknesses and reassemble lessons and content, rather than showing everyone the same fixed path.
The next layer is emotional data: which sections excite the user, how hard they play, what they sing and what they want to create may all leave signals in the hardware and App. 牛亚锋 believes that the songs someone likes today versus 1 year ago can reflect different understandings of life and different stages of a person’s life; the system must update as those preferences change.
李翔 asks how music, which represents emotion, should coexist with AI, which is more rational. 牛亚锋’s answer is not for AI to replace emotion, but to act as a private teacher or music companion. If the system knows a user meditates, the guitar sitting nearby might proactively offer meditation music, gradually moving beyond the role of a simple tool.

6. Suno lowers the generation barrier; hardware returns creation to the body

牛亚锋 has recently been spending time with Suno. He observes that it has moved from generating an entire song end to end to supporting stems: relatively sophisticated users can open individual tracks, trim a drum section they find too loud and even ask the model to replace it with another style.
This will allow professional creators to produce large volumes of content at lower cost. The episode emphasized a key distinction: pure software generation lacks the user’s physical sensation, movements and performance feedback. “Driving in a simulator” is not the same experience as driving on the road, and hearing music is entirely different from playing it with one’s own hands.
戴乐 therefore breaks chord progressions, rhythm patterns, drums, piano and lead melodies into fragments. Users do not need to adjust every musical element from scratch; AI can schedule material based on their tastes and skill level, while users still combine it through the App and hardware, injecting their own choices and personality into the result.
牛亚锋 imagines an integrated state in which a user, in a particular setting and emotional state, improvises music that matches the moment. In the past, only highly skilled musicians with a large internal library of fragments could do this; in the future, beginners could use AI to access that library. “They only need to input the feeling, and the output is music”—in other words, “what you feel is what you play.”

7. MIDI, content slicing and behavior logs are more fundamental AI infrastructure than a chat interface

Every press and pluck at the hardware layer produces a small MIDI signal, and MIDI is a standard instrument protocol. At the App layer, the system continuously accumulates fragmented content. 牛亚锋 calls these AI’s “ingredients”: “It needs the ingredients before it can cook.”
The simplest application is product-knowledge Q&A. The next step is reassembling lessons based on the user’s level and recommending preferred content such as folk songs. Because the model knows both the hardware actions and App behavior, it may schedule the next lesson and content recommendations based on actual playing performance.
The platform could also connect people: matching users by interest, organizing bands and extending existing offline band functions online, with AI potentially acting as the “band captain.” 牛亚锋 does not rule out opening up some creative and AI capabilities so users can combine them into new formats their teams had not imagined, though he continues to describe this direction as a possibility.

8. Product maturity is not about adding more features, but knowing what to remove

The earliest prototype was a pair of gloves that let users “play” in the air. Engineers thought they were cool, but users could not understand them. The team then turned them into a wristband and kept adding features such as a time display and step counter; after nearly 1 year of work, the demo still left testers confused: “Is this a wristband or a music product?”
牛亚锋 consequently changed his engineer’s instinct to keep adding. Excess features become cost, weight, battery drain and cognitive load. A good product should keep subtracting until the core function generates the greatest experience in the most precise setting; “what to keep and what to give up” tests a product team more than simply piling on materials.
The first-generation drum product, defined around 2020, also took a wrong turn. It added lights so the drumsticks glowed when the drums were struck and was sold mainly overseas; real drummers kept asking for something closer to wood and an unvarnished timber feel. The second generation therefore shifted to a natural-wood style, because the target user, ultimately, should decide “what is cool.”
Over 10 years, 牛亚锋 moved from a purely engineering mindset to “engineer plus artist.” Technology solves the structure; an artist’s understanding reaches human nature, individuality and emotional value. Internally, there may be thousands of components and complex scenarios, but once the product is enclosed and placed in a user’s hands, it must be simple and fluid. That is also what he learned from Apple’s products.

9. AI is moving down from the application layer into the driver layer, but heat is not a product rationale

On the Open Cloud mentioned in the episode, 牛亚锋 first offered an honest qualification: “To be honest, I haven’t used it.” He does not rush into something simply because it suddenly becomes hot; turning new technology into a real product in daily life takes time to validate, and personal interest cannot be conflated with commercial value.
He nevertheless sees a clear direction: models are moving from NLP and simple recommendations toward higher-dimensional reasoning, decision-making and Agent collaboration. Previously, users entered information and systems returned information; now, more ambiguous and sophisticated tasks can be broken down among different Agents to perform sequences of actions on a personal computer.
The next step could take AI close to the hardware layer. Engineers previously wrote drivers and upper-layer applications separately; in the future, that code may be generated by AI. If devices get their entry protocols and standards right, language-based requirements could become code in real time, allowing hardware such as cameras to call existing modules instead of relying forever on fixed functions preloaded in firmware.

10. New hardware opportunities sit at the two ends of sensing and operating the world

Asked which categories could become the phenomenon-level products of the next 3-5 years, 牛亚锋 declined to name any with certainty: “It’s actually quite hard to predict which products will take off.” He would rather bet on the underlying conditions. AI itself is virtual; it cannot independently sense the environment or directly operate the world, so it needs sensors and actuators to fill those two ends.
Cameras offer a model for perception. Previously they were simply connected devices; combined with AI, they might identify whether the owner is happy or unhappy upon returning home and abstract raw footage into a higher-dimensional state. Air conditioners and other actuators could then turn on or respond accordingly, moving smart homes from remote control toward genuinely proactive responses.
This combination does not mean “putting AI into everything.” 牛亚锋 stresses that commercialization still depends on an excellent product manager finding a clear pain point, then deciding how sensors can help the model understand the environment and how actuators can address a real need. Adding more hardware does not automatically create value.
李翔 jokes that when devices handle every task at home, users will have the time to play guitar. The joke points to the division of labor in consumer hardware: AI eliminates low-level operations, while music products compete for the mental time that has been freed up.

11. Globalization is not translating a domestic product; it is using one underlying platform to carry different music cultures

牛亚锋 characterizes China’s musical starting point as “starting with singing.” Many users have had little exposure to instruments, so singalong play, teaching and familiar songs matter more; Japan, South Korea and China together form an Asian extension of karaoke culture, where lowering the barrier to first use is the priority.
Western markets are different. In his view, instruments and rock culture became widespread earlier, and many people have had at least some exposure to music theory or an instrument, especially guitar. Users are not satisfied with playing other people’s songs; they want to express individuality, develop a hobby and create their own content.
A fully digitized guitar can connect via standard MIDI to Cubase, DAWs and other professional creation software. Combined with sliced material and AI, it lets experienced users go deeper while giving ordinary users a low-barrier entry into creation. That creates room for the same product to move from a “learning device” to a “creation device.”
The product can use a common hardware layer with different App functionality and content for different regions and stages. If the trade-offs cannot be reconciled, different SKUs are also possible. 牛亚锋 does not treat standardization as an absolute; he weighs hardware reuse, clarity of positioning and actual needs in each market together.

12. Algorithms can drive traffic, but only real emotion makes branded content stick

A technical background does not preclude an understanding of social media. Douyin and TikTok are built on recommendation algorithms, and structures around the first 3 seconds, the middle and the ending have gradually become standardized. 戴乐 operated its own accounts in 2021-2022, and some technically trained team members quickly learned the logic of traffic.
李翔 asks how content can hit users emotionally if everything is industrialized and data-driven. 牛亚锋 concedes that algorithms alone produce “a lot of garbage.” Content has to return to real situations, such as the joy felt by someone who has never touched a guitar when they complete their first singalong performance; that experience is what can “hit the heart directly.”
The team therefore brought in both technical and music talent. Around 30-40 members with music backgrounds are distributed across product, operations, livestreaming, testing, arranging and project roles. One guitarist at the factory also brought in a dozen-plus colleagues; their former income from live performances was unstable, but now playing every day is both their job and a sustainable hobby.
This also explains “passion” in the brand culture. Commercialization gives musicians who were previously difficult to gather and employ stable positions, and they in turn influence new users through testing, content and service. 牛亚锋’s formulation is that “passion is contagious,” so brand communication cannot talk only about parameters.

13. Retention comes from emotional value users can repeatedly enter; mental consumption is expanding the demand pool

戴乐’s hardware, courses, content and App features are iterated roughly once every 2 weeks, with user suggestions able to enter the feedback channel directly. 牛亚锋 does not describe long-term use as the result of aggressive operations; he believes that once the product is placed at home or taken outdoors, every time users think of it and pick it up, continuous improvements naturally increase the time they invest.
By his preliminary estimate, the company “may” already have users in the 100K-plus range, with cumulative usage “possibly close to” 2M hours. Typical situations include playing favorite songs, learning or recording the day’s emotions for 30 minutes to 1 hour after a dull, oppressive workday. The key line is: “It must either get users hooked or help them relax.”
李翔 offers a concrete example: after learning the instrument, one colleague taught his son and father. The three generations were in different locations but could play the same song; language can be blocked by a generation gap, while music lets people from different eras take different roles and becomes a shared interface for family emotion.
牛亚锋 takes the change in demand over a longer cycle. Humans have always used music to summon and express emotions, but over the past several decades many people in China focused more on food, clothing, housing and transportation. After the pandemic, running, the outdoors, health and mental pursuits expanded rapidly; users became less fixated on pure utility and price-performance, and began asking, “How can I make these 30 minutes relaxing and enjoyable?”

14. Overseas markets, Shenzhen and a large category form the growth path; capital can accelerate PMF but cannot replace it

The first major inflection came around 2019-2020, when the company shifted overseas and established its first foothold in relatively high-quality markets. 牛亚锋 says the domestic market is “impossible if it doesn’t get hot, but hard to hold onto once it does,” while hardware R&D, supply chain, mass production and delivery form a long chain. Surviving first matters more than chasing heat in the early stage.
The second inflection was the move to Shenzhen. A product containing 200+ innovative components would be difficult to integrate and iterate in Beijing, while Shenzhen materially increases speed. Hardware cannot be changed at will after launch like an App can; it often has to be right from the beginning, making the distance to the industrial chain directly relevant to the efficiency of trial and error.
The third inflection was choosing the guitar category: the path ran from a student idea and competition project to a product, a global consumer good and, ultimately, the opening of a new category. 牛亚锋 puts the traditional musical-instrument market at more than $20B, with guitars accounting for nearly half, or about $10B, and annual sales in the 10M-plus range. Choosing a category large enough, then innovating at the underlying layer, was what allowed the company’s accumulated work to scale.
Financing for AI hardware is easier today. 牛亚锋 started with RMB150K; some teams now may receive RMB15M or even US-dollar funding at the outset. But he warns founders to respect the difficulty of hardware. Even areas such as embodied intelligence may have enormous potential, yet companies still need a pain point that can be commercialized today; capital confidence fluctuates, and parameters that cannot create lasting value leave only a bubble when the excitement fades.
In a new financing round, early investors often ask whether the pain point is real; once sales are validated, they ask about the ceiling of the category. 牛亚锋 believes fundraising is not simply a cash transaction. Founders should also ask investors, “Why are you investing in me?” and look for people who truly understand the project and are willing to go through both the highs and lows with it.
李翔 says the company now has close to 200 employees. The core priorities for the new year remain delivering the heavily invested second-generation product and improving its functionality, content and App; the company will also build out the brand-content team, expand global offline experiences and use the third generation and additional product lines to address creative demand in Europe and the US. Using capital when conditions are favorable while remaining healthy when the tide goes out is what “lasting longer” should mean.

Building Agents at Home: Homeschooling, Parenting and More | The a16z Show Ferrari: What happens when you staple a luxury brand to a sports team? (Audio)