14 lessons from a week in China

It is a lot of information to absorb in just a few days. I do not think one week is enough to understand China. But it is enough to realize that some of our questions about innovation need to evolve.

Which model is better? Which platform just launched a feature? What can we already generate with a prompt? Those questions still matter. But during the visits, my attention increasingly turned to another question: how does all of this connect to make an operation work?

I organized these reflections into 14 lessons, with a few checkpoints to bring the conversation back to business. These are questions I brought away from the visits: what makes sense for Brazil, who can build it and where do we start?

1. A trip about technology also needs to be about education

One of our first stops in Shanghai was Shanghai Business School. Before visiting the technology companies, we explored hospitality training spaces, with areas dedicated to coffee, tea and wine, and attended classes on innovation and business culture.

Starting there helps us remember the people who will use the technology. Each person has a routine, knows how to do certain things and needs to learn others. If a company wants to change how that work happens, it also needs to take part in that learning process.

This question applies both to those developing a product and to those buying a solution. How will people learn to work with it? An important part of implementation begins with that answer. Notes from Shanghai Business School.

Practical checkpoint: follow an activity from beginning to end. Where does the person get stuck, who do they ask for help and what do they need to learn?

2. Trust is built through the work that continues after a meeting

In the business class, trust, cooperation and long-term relationships emerged as central themes. The discussion involved understanding what the other person needs to achieve and finding opportunities to build something together.

That connects closely with what I believe about community. A connection gains value through exchange, continuity and a willingness to contribute. Meeting someone can open a door. Building a relationship means finding shared interests, doing what you agreed to do and continuing to be useful to the other party.

The question that remains is what happens after an introduction, an event or a conversation with a potential partner. That is when the work of building together begins. The class offered a perspective on this that needs to be understood in its context. Notes from the business culture class.

Practical checkpoint: choose a conversation worth continuing. What can you contribute, and what next step makes sense for both of you?

3. Selling the robot is only part of the work

At WhalesBot, there were kits, robots, drones and programming tools. Alongside the equipment, the offering included content, teacher training, challenges and competitions. Educational robotics was presented as part of a learning journey.

In practice, the product needs to serve the child who is learning, the teacher leading the activity and the school incorporating it into its routine. Each person faces a different difficulty. An engaging activity loses momentum if the teacher cannot prepare it or the school cannot find a way to keep the equipment in use.

This applies to many Brazilian businesses. Sometimes we build an interesting technology and leave everything that enables someone to use it for later. Implementation, support and training are part of the promise to the customer, and they may be what sustains the business. Insights from the WhalesBot visit.

Practical checkpoint: look at what happens after the sale. Who helps the customer get started and solves the first difficulty?

4. Making an AI speak is different from being able to hold a conversation with it

During the visits to iFLYTEK and Agora, language appeared in applications for translation, education, workplace support and voice interaction. The conversation at Agora also connected with the work I am doing on my own clone. Giving AI a voice raises a series of questions about how the conversation will work.

A voice can sound convincing while the experience is still poor. The response might take too long, ignore a correction or keep talking when the person wants to interrupt. These details may seem small in a presentation. In everyday use, they determine whether someone can complete the task and whether they will want to use the product again.

In voice-based customer service, the conversation needs to lead somewhere. Can the person explain what they need, correct a piece of information and continue? When they need help, can someone take over the case? That is the experience the product will be judged on. Notes from iFLYTEK and the Agora visit.

Practical checkpoint: test a conversation that includes an interruption and a correction. Check whether the task is completed and at what point a person is needed.

5. A good model still needs to become a good product

At MiniMax, we discussed models, products and distribution. The visit helped distinguish layers that are often mixed together in the AI debate. There is the model's capability, the experience built on top of it and the path through which someone discovers, tries and incorporates the product into their work.

A technical improvement expands what can be built. But the customer still needs to discover the solution, understand what it is for and be able to use it. That transition between technical capability, product and distribution deserves as much attention as choosing the model.

The visit connects with an idea I have been advocating: technology and business need to be considered together. A tool can do many things and still lack a path for someone to recognize its value. Those who know the customer and those who understand the technology have work to do together. Insights from the MiniMax visit.

Practical checkpoint: explain how the customer finds the solution, what they can do the first time they use it and why they would return.

6. An alert gains value only when it leads to a decision

In Hangzhou, Hikvision brought AI closer to the physical world. We saw demonstrations involving images, sensors, acoustic perception and automation. An industrial component could be observed to support an inspection. A signal picked up in the environment could inform an alert.

This broadens the conversation about artificial intelligence. It can help identify a failure, assess quality and organize a physical process. Its value also depends on what happens next: who receives the information, what decision they need to make and how that decision reaches the operation.

After the alert, who does what? The difficulty might lie in detecting a failure. Or the information might already exist but take too long to reach someone who can act. These two situations open up different opportunities for those who know that operation. Notes from the Hikvision visit.

Practical checkpoint: follow a real alert through to resolution. Where does the case sit waiting, and who can move it forward?

7. AI is starting to take on a sequence of work

At Alibaba, execution appeared in commercial routines. One example connected a request received by email to supplier research, proposal preparation and a response to the customer. It was a sequence with a goal, information and tools that needed to work together.

What interests me about this application is the possibility of delivering a more complete task. A good answer can help while still leaving someone copying information between systems and chasing the next response. When AI participates in that sequence, the product begins taking responsibility for a larger part of the delivery.

This also changes the question for those building it: which part of this work can my product handle? Answering requires knowing the systems, the exceptions and who can authorize each step. Notes from the Alibaba visit.

Practical checkpoint: choose a routine between the request and delivery. Which part can you handle, including review and rework?

8. Building a business alone is still a networked effort

It was in this context that we discussed OPC, or One Person Company. One person coordinating a greater capacity for execution with the support of agents, platforms and services. To me, this is one of the most interesting discussions in entrepreneurship: expanding the possibilities of starting and building with a smaller organization of your own.

But that person still depends on suppliers, infrastructure, distribution and business relationships. The conversation at Sandbox, later in Shenzhen, brought this point to hardware. Design, manufacturing, quality, logistics and customer service remain necessary, even when tools make certain tasks more accessible.

I see a major opportunity for more people to get started. Someone who understands a problem can gain the ability to build, test and reach customers. As the business grows, understanding where other people are needed and how to work with them remains part of the journey. The conversation about agents and OPC at Alibaba and notes from Sandbox.

Practical checkpoint: what can you already build with AI, and where do you still need a partner? Look for the capability you are missing for the next step.

9. Good infrastructure also helps you learn

At WeBank in Shenzhen, we discussed data, architecture, credit decisions and operating at scale. A wall of patent certificates made some of the bank's accumulated technical knowledge visible. One particularly interesting point was realizing how infrastructure can also create the conditions to try a change, track the results and decide whether it makes sense to expand it.

The conversation about bringing lessons from this model to Brazil came with an important caveat: local data, customers and conditions would need to be studied. That care applies to the entire trip. Discovering a solution in another country broadens our frame of reference. Applying it requires understanding why it works and what changes in our context.

Before expanding a change, we need to be able to learn from it. Who takes part in the test, which results will we track and what would make us change direction? Infrastructure helps when it allows us to answer these questions with data about what happened. Insights from the WeBank visit.

Practical checkpoint: before a test, write down what you want to discover and how you will recognize an improvement.

10. A new product begins long before its launch

At the BYD museum, a sign bearing the name “BYD Battery Co., Ltd.” helped tell a story that began before cars. The journey connected batteries, components, research and expansion into the automotive sector. In another installation, the parts of a vehicle were separated and suspended, making it possible to see the collection of systems behind the finished product.

It is a useful image for thinking about innovation. We see the launch. We do not always see the years of learning, the capabilities and the decisions that made that launch possible. Even when technology accelerates one stage, there is still work to do in engineering, coordination, testing and improvement.

This perspective helps us look at our own companies. Accumulated knowledge can open an opportunity that is not yet obvious. It is worth understanding what we already know how to do and where it might meet a new problem to solve. Notes from the BYD museum visit.

Practical checkpoint: what capability has your company already built that it could use to solve a new problem?

11. Knowing how to manufacture and knowing how to sell need to come together

At Sandbox in Shenzhen, the conversation brought together AI hardware, brand building and international operations. Wapitee offered the perspective of connecting products with consumers in other markets. The ability to produce needed to meet a proposition that would make sense for someone to buy and use.

The meeting showed how engineering and the market need to learn together. Positioning, user experience, communication and distribution help define what is worth building. The discussion about crowdfunding added an important distinction: a campaign's visibility alone does not establish that manufacturing, delivering and sustaining the product will be economically viable.

Those who know the customer, those who can produce and those who understand distribution each have parts of this problem to solve. Bringing these people together is an opportunity for entrepreneurial communities in Brazil. The exchange takes on another dimension when it helps someone build and bring a product to market. Notes from the Sandbox visit.

Practical checkpoint: beyond the interest in buying, what do you need to check to be able to manufacture, deliver and sustain the product?

12. A team of agents also needs to be organized

At Zhuiyi Technology, the presentation placed a team of specialist agents at the center of the proposition. Customer service, employee assistance, decision support and system operation appeared as complementary functions. One segment described a simulation: one agent plays a difficult customer, another performs the task and an observer evaluates the interaction.

This design helps make the work visible. What does each agent receive? What decisions can it make? When does it need to pass the case on? Simulation offers situations to learn from, while real use will bring other behaviors to monitor.

The conversation also addressed people's work. In its answer about employment, the team acknowledged reductions or restructuring in some operations and mentioned reassignment. Those implementing this change need to take care of how the team will work, learn and take on new responsibilities. Notes from the Zhuiyi visit.

Practical checkpoint: rehearse an exception. Who notices the problem, can intervene and helps the team work with the new process?

13. AI glasses need to help in a real situation

At INMO, AI took the form of glasses that place information in the field of view. Translation, guidance and presentation support came into the conversation. The product families had different propositions and architectures, and a question about navigation led to discussion of a future update with Google Maps.

Trying the devices brought the discussion closer to the person's activity. Following a conversation, consulting a script and receiving directions are different needs. Usefulness depends on when information appears and the effort required to use it. Comfort, battery life, connectivity and the environment belong in the same assessment.

This is one way to look at the glasses: in what situation does having the information there, in your field of view, help? The answer depends on the task, the device and the features already available. Trying them in that context is the next step in understanding where the product makes sense. Notes from the INMO visit.

Practical checkpoint: choose a task where the glasses might help. Check the model, the available software and how the experience changes without a connection.

14. AI and robotics will increasingly come together

My bet is that artificial intelligence and robotics will move ever closer together. Systems capable of interpreting information, coordinating tasks and, in certain contexts, guiding actions in the physical world. That perspective appeared in different ways across the education, industrial perception, agents and interfaces we encountered.

This can expand what one person or a small team can achieve. For those who understand a problem in education, industry, services or commerce in Brazil, there is room to build. Knowledge of that work helps identify where technology can fit in and what needs to work for someone to use it.

Along with the ability to build come decisions about safety, data, maintenance and responsibility. Teachers, operators, developers, customers and partners have different knowledge to contribute. Bringing these people into the conversation can help turn a technical possibility into something that works in everyday life.

Practical checkpoint: which problem in your sector have you started seeing differently? Talk to the people who live with it and choose a hypothesis to test.

I return from these visits with better questions about what we can build in Brazil. To start, I would choose one checkpoint and one concrete routine. I would put someone in charge, set a deadline and establish a way to track whether the test helped.

I want to keep discovering new technologies. But above all, I want to understand which ones we can turn into better businesses, learning opportunities and solutions that make a difference in people's lives in Brazil.

To me, this is a direction for NewHack's work: bringing knowledge, people and opportunities together so that more entrepreneurs can experiment and put their ideas to work. I brought together the visit notes and photos from the trip to continue this conversation.