Summary of Key Points
This article focuses on the critical phase in which the AI industry is transitioning from merely showcasing technical prowess to delivering tangible value. Through in-depth conversations with over 20 CEOs and 27 senior executives during the World Artificial Intelligence Conference (WAIC), Huxiu has identified 42 practical experiences across six key dimensions: Tokens, Robots, Implementation, Endpoints, Realization, and Organization. The core conclusion is that the AI industry has moved from a stage of technical feasibility to one of commercial sustainability. Companies are no longer paying for the capabilities of models but rather for tangible results. Whether AI can bring about clear, cost-effective, and higher-quality improvements for customers will be the decisive factor in its ability to establish a foothold in the market.
1. The Token Aspect: Focus on Results, Not Just the Number of Tokens
Tokens, which can be understood as units of information processed by AI (such as the number of characters generated or pixels recognized in images), were once the primary metric for charging AI services. However, companies are increasingly realizing that tokens do not represent true value; the results are what matter.
- Paying by token is like paying for the number of words in a written article, but customers want to know if the article will generate sales, not just how many words it contains. In the future, payments will likely be based on tasks completed. For example, if an AI writes 10 advertising copy drafts and only 3 are successfully used, the company will pay a fixed fee, regardless of the number of tokens used.
- The same applies to computing power: While cheap computing resources are readily available, those that are both affordable, reliable, and readily accessible are in high demand. Just like mobile data, it’s not enough to have a low price; the service must also provide good signal quality, avoid outages, and ensure seamless performance during usage.
- Companies are reevaluating their cost calculations. They no longer compare the cost per token but rather the total cost of each task. For instance, the true cost of providing customer service includes how many times AI is called, how many retries are needed, and whether the problem is ultimately resolved.
2. The Robot Aspect: Show off Skills at Exhibitions, but Stability Matters in Production
Robots may perform impressive tricks at exhibitions, but in production lines, stability is far more important than occasional highlights.
- Production lines do not reward exceptional moments; robots that operate continuously, even if they are slightly slower than skilled humans, will produce more over time. A 1% failure rate, repeated thousands of times a day, can lead to constant downtime. Therefore, reliability and error-free operation are absolute requirements.
- The learning process for robots is similar to that of students: human demonstrations serve as lessons, simulation environments provide practice opportunities, and real-world feedback helps improve their performance.
- Don’t be fooled by fancy terms like “world models” or “VLA”; customers will only pay for robots that can perform practical tasks accurately, such as screwing screws or moving parts. Even the most humanoid robots must be durable and efficient to meet industrial standards.
3. The Implementation Aspect: The Challenge is Not the Model, but Reengineering Processes and Changing Mindsets
The biggest obstacle for companies adopting AI is not the technology itself but reorganizing existing processes and changing people’s mindsets.
- Overcoming three major barriers is essential: taking months to organize internal knowledge (e.g., converting old employees’ expertise into a format understandable by AI), integrating with existing systems, and setting up proper access controls. This is just the beginning.
- Companies are concerned about the lack of clear cost benchmarks for using AI. Without a defined budget, large firms are reluctant to invest due to uncertainty.
- Organizational changes are necessary. Instead of adding AI tools to existing processes, the focus should be on shifting from a human-centered approach to an AI-centric one, where AI handles repetitive tasks while humans make judgments and decisions. However, many companies only optimize execution, leaving internal inefficiencies (such as unnecessary meetings and reports) unchanged.
4. The Endpoint Aspect: Hardware is Not the Problem; User Adoption Is
For home robots and smart devices, the key is not whether they have AI capabilities but whether users will actually use them daily.
- New household robots may initially perform a few tasks (like cleaning or turning on the air conditioner) with only a 30%-50% success rate. They need to learn from real-life usage scenarios, such as remembering the location of furniture and user habits, to improve their effectiveness.
- Simulation environments are not reliable; robots that perform flawlessly in labs may fail in real-world settings. Selling a device is just the start; collecting user data and continuously improving it is crucial for success.
- End-user AI solutions are not free; while the hardware costs are one-time investments, long-term expenses (chips, memory, electricity, maintenance) must also be considered. Similar to electric vehicles, users need to pay for charging and maintenance.
- Services should be proactive. The value of new endpoints lies in anticipating user needs—automatically turning on lights or adjusting the temperature before the user even asks.
5. The Realization Aspect: Customers Want Clear Results
AI service providers must demonstrate tangible benefits to earn revenue, not just claim they can empower customers.
- A closed-loop approach is essential. An AI-generated ad must be successfully deployed and generate sales to be considered effective. High churn rates, masked by new users, will lead to greater losses over time.
- A sustainable advantage comes from assets that AI cannot replace, such as expert knowledge, industry insights, and tacit skills (e.g., how experienced salespeople interact with customers). These are the core competencies that AI cannot acquire.
- Humans will always be responsible for any mistakes made by AI. When issues arise, someone must take responsibility.
6. The Final Aspect: The True Value of AI Is Yet to Be Determined
The article concludes that the AI industry is moving from assessing technical feasibility to evaluating its commercial viability. The next 1-2 years will shape the future landscape for the industry. Huxiu will continue to explore how tokens can create real value that is recognized, paid for, and repurchased by customers. This is a question that all AI practitioners and companies must address, as only when customers profit can AI services thrive.
In summary, the transition of the AI industry from technology demonstration to practical application is underway, and the next few years will determine its long-term trajectory.