Over 200 AI assistants officially launched

2024-05-14 10:42:14

On April 18th, DingTalk announced a significant innovation: the official launch of the AI Assistant Market (AI Agent Store), featuring over 200 initial AI assistants. This model greatly lowers the barriers to creation, enabling users from various industries to have their own exclusive AI assistants. Official data reveals that, by the end of March 2024,

On April 18th, DingTalk announced a significant innovation: the official launch of the AI Assistant Market (AI Agent Store), featuring over 200 initial AI assistants. This model greatly lowers the barriers to creation, enabling users from various industries to have their own exclusive AI assistants.

Official data reveals that, by the end of March 2024, more than 2.2 million enterprises had utilized DingTalk’s AI technology, with over 1.7 million active enterprises per month.

Following OpenAI’s introduction of the GPT Store, insights suggest that AI large models are entering a new era—akin to an “App Store,” which also gives rise to new business models. Subsequently, not only well-known tech companies but also emerging startups have joined this wave, launching their own “stores,” indicating that we will see a multitude of AI app stores emerge in the next few years.

Although technology has always existed as a differentiator for software companies, in the AI era, which champions the notion that “everyone can be a developer,” these technological differences are diminishing, and the moats of the AI era are rapidly shrinking.

In addressing the topic of DingTalk’s launch of the Agent Store, its differentiation from other AI app stores, and the possible future changes in the moats, DingTalk CTO Cheng Cao Hong (nickname: Babu) had an in-depth discussion.

Regarding the perception and evolution of DingTalk products, Cheng Cao Hong (Babu) recalled:

“DingTalk’s precursor was ‘Laiwang’, which initially targeted a broad market and overlapped functionally with WeChat. However, in the fiercely competitive social media market, this initial positioning did not yield the expected success. Later on, DingTalk quickly dedicated itself to the work scenario, providing users with more efficient communication tools and meeting the clock-in attendance needs of many Chinese office workers, thus achieving initial success.”

He continued: “In China, there is a significant demand for efficient office tools in knowledge-intensive and high-tech industries, as well as a vast manufacturing market. In such an environment, DingTalk seized market opportunities and achieved rapid growth, especially during the pandemic when the growth was explosive. I joined the company after DingTalk’s rapid growth, and our challenge was how to not overly rely on previous strategies and achieve new breakthroughs.”

In the process of exploring ways to enhance market customers’ understanding of products, we recognized that we must address multiple aspects of the product ecosystem. Therefore, we established two key strategic positions: one is to serve as a collaborative office platform, focusing on enhancing work efficiency throughout the office environment; and the other as an application development platform, supporting in-depth enterprise business scenarios.

The role of the collaborative office platform is not only limited to simplifying document transfer, document sending, daily chatting, or OA approval processes and other basic office activities. The real challenge lies in playing a role in deeper levels of work, such as organizational governance, business operations, talent management, and the procurement and sales of materials. These are often overlooked needs, like the tip of an iceberg.

Positioning the application development platform as an infrastructure, we are committed to building more in-depth business scenarios with partners. Digital transformation encompasses not only office digitization but can gradually extend to the overall digitization of the organization. Business digitization is also an area that the application development platform can steadily cover and is a key part of an organization’s comprehensive digital transformation.

Based on this philosophy, DingTalk has proposed a PaaS First strategy, which focuses on creating an infrastructure for an open ecosystem. This means more resources will be invested in building underlying technical capabilities rather than developing scenario-based application products. DingTalk will address the challenges and difficult tasks within the ecosystem, including those that are not widely welcomed or are difficult to achieve.

Regarding end-user support, SaaS companies need to face huge investment demands. When accessing large enterprise customers, the first consideration is the support capabilities for multi-platform systems, such as Android, iOS, Pad, and the consistency of the product experience. Data security has become a high-priority requirement, and data privacy protection has become an important factor that businesses must carefully consider. This includes security capabilities, connectivity, the stability and scalability of instant messaging (IM) systems. For example, being able to update the read/unread status of every message in real-time in a large group is a significant test of system performance.

SaaS companies must lay a solid foundation in IM system security and stability, multi-platform adaptation, and open expansion. Companies like Belle have succeeded in meeting the needs for deep customization by developing applications in-house and deploying them to groups. Integrating business collaboration into the instant messaging system, the combination of the office collaboration platform and the application development platform has opened up new possibilities. Functions such as audio-video, IM, and document collaboration, which are part of the office collaboration platform, are gradually becoming key carriers of business digitization. Taking DingTalk documents as an example, its emphasis on multi-person collaborative editing is becoming a new stage for business digitization, indicating that they are not isolated existences.

Finally, the audio and video features widely used in scenarios such as recruitment also offer more potential for the integration of office collaboration and application development platforms.

With the continuous advancement of technology, ecological partners are expected to create innovative interview and recruitment applications based on audio and video capabilities. Combining artificial intelligence technology, possible pre-interview functions will undoubtedly enhance the efficiency of the recruitment process further. This marks an important evolution of products in their lifecycle.

Faced with daily challenges, I often feel a certain kind of entanglement. Ultimately, our understanding of the concepts of “user” and “customer” is deeply meaningful. The key is that we must focus on the real experiencers; whether they are company employees or individual users. It is these users’ experiences that determine the key to the success of our product, and these experiences often do not align with what the boss expects.

As operators, we also need to ensure that our speech is aligned with commercial considerations, ensuring that our products meet the demands of the boss, management, and the entire organization’s development. In this process, we advocate the core values of the “3C” principle. The first C refers to communication, which is our foundation. We must optimize communication and collaboration to ensure an unobstructed flow of information.

While ensuring smooth communication, we also believe that control is crucial. From a management perspective, “control” actually refers to a broader concept of governance. The maturity of a company largely depends on its top-down governance capabilities in various aspects such as finance, organizational goals, and cultural advocacy. This aspect of control involves the governance processes and mechanisms of key business departments like marketing. How to effectively deliver governance decisions and ensure their execution is critical in digital transformation.

On the other hand, we need to focus on the awakening and integration of individual employees. The employees’ perception of power and their experience with work tools must adapt to the control measures by management. What we are exploring is the seamless integration of governance measures with context after establishing an efficient communication platform. This is the problem we are trying to solve in management philosophy, to avoid merely catering to employees or the boss, instead of achieving a win-win situation.

We are not so arrogant as to think we have all the capabilities to handle these issues but rather to explore innovative paths together with our customers. For example, the emergence of deeply customized cases demonstrates the potential for organizational structure change. When the collaboration of a group reaches a certain frequency, we can consider transforming this group into an actual organization. This way of thinking is very interesting, prompting us to consider how technology affects products, and in turn, the entire ecosystem and the underlying management philosophy. This is also the journey we are exploring.

When talking about the balance point between customers and users, it is actually a delicate art. There was a time when some employees were fearful at the mention of DingTalk, but what is the balance between control and context now?

In fact, our goal is not to weaken the control of the management layer but, on the contrary, to enhance it. For example, using widespread data and the assistance of artificial intelligence, leaders can understand the operations of the enterprise more precisely, rather than simply driving employees to work. What we are pursuing is to stimulate the vitality of the employees, allowing managers to make wiser decisions based on data. In fact, these two directions are complementary. At the employee level, we sincerely put people first, for example, by providing experiences like off-duty mode and do-not-disturb functions. For the management layer, they may question why they cannot disturb employees. But essentially, this is actually to improve the efficiency of the team’s goals and productivity. If data transparency is higher and team collaboration more efficient, then true do-not-disturb becomes possible. This makes our goals clearer and our priorities more defined in phases. Through basic communication, combined with AI and data, the convenience upgrade we talk about is also possible. At the product level, everyone is in the process of striving in both directions.

About the topic of AIGC, everyone is very concerned. Speaking of the DingTalk product, we have a wealth of scenarized applications, and the value and role AIGC brings to DingTalk is crucial. Has it brought about disruptive changes, or as we often say, assistance and cost reduction and efficiency improvement?

To answer this question, we need to analyze in detail. Speaking of product changes, we can say they have undergone three stages or patterns, and we are currently promoting the development of these stages. Firstly, one of the impacts brought by artificial intelligence is what everyone first thinks of – applying AI to products, just like installing an engine on a horse carriage in the past. Although this can help popularize the cognition of AI, it does not really solve the problem. The second method is like introducing an assisted driving system “copilot,” which allows us to feel the value of AI more practically. For example, in terms of scheduling and attendance rules, traditional methods may require learning complex scheduling rules. If software training is insufficient, then using it can be very troublesome. But now, with the copilot mode, you just need to describe your needs in natural language, for example, the factory needs to set up three shifts, and the software can help you complete the setting. I think such progress is obvious. The same logic also applies to our YiDa tools, which have shifted from a professional development mode to low-code development, which indeed has greatly reduced the difficulty. For those business personnel without any coding experience, they may need to learn some technical terms such as formula editing, headers and footers, data sets, and may involve some JavaScript, which could be challenging for them.

Thanks to the continuous advancement of technology, we can now make requests to intelligent assistants such as this: “I made a form and took a picture of it, can you help me convert it into electronic data?” The convenience of this technology is self-evident. However, in my view, this cannot be considered a disruptive innovation. It merely signifies that the trend of the times has arrived, and we have made some adaptive adjustments accordingly.

The real revolutionary change, I believe, must be reflected in the appearance of the “Agent” form. Different from the enhancement of traditional tools, “Agent” possesses almost human-like characteristics. This is reflected in its ability to remember the preferences of users and have a deep understanding of these preferences. For example, if I made special requests to this Agent during my last use, such as using size-12 font, adopting DingTalk Progressive font, and hoping that each paragraph does not end with a summarizing language. When I use this Agent to write documents again, it will remember these specific requirements and apply these preferences to other scenarios such as writing logs or weekly reports.

Therefore, such changes are very critical. The personalized form of the Agent makes it not only able to understand your preferences but also capable of orchestrating actions across systems. This means that with the introduction of workflows, although the change may not be a drastic upheaval, it has indeed completely restructured the past software development and product usage patterns. I have a firm belief in this kind of change.

Taking the core functionality of DingTalk as an example: When you send a picture in an instant message (IM), you can immediately click an icon to understand the content of the image. This is just one of many instances. Another example is using mind-mapping software to plan a team building activity, where the Agent can provide inspiration based on the information you provide, representing an embedded application. On the edges of the attendance software, Copilot can assist you in setting up attendance rules. With Yi Da Copilot, even if you are unfamiliar with certain specific functions, you can still operate using the existing drag-and-drop mode, and ask it for help when needed with questions like, “How do I operate this?” or “How should I write this story?” It is capable of offering immediate assistance.

More and more Agent applications are emerging, such as masters of Zisha teapots, photography partners, and matchmaking go-betweens. They not only chat with you from time to time to share some interesting news, but also gradually get to know you better through the process. When you face difficulties, the Agent can comfort you and record the process. In the future, when you are looking for a life partner, the Agent will take note of your sensitivities towards certain things, making your experience unique due to this personalized perception.

As artificial intelligence technology continues to develop, existing work patterns are beginning to change quietly. For example, the award-winning Agent “partner” integrates various functions, including MBTI personality test, mutual matching, and information link sharing, making collaboration more efficient and convenient.

Another innovation worth noting is the operations assistant project, which has changed the way technicians analyze cloud machine link issues. With the help of the operations assistant, complex anomaly analyses can be automated, and users can continuously input new knowledge to make its analyses more accurate in conjunction with resource loads. This continuously evolving system fully demonstrates the potential of artificial intelligence in the technical field.

Discussing the originality and technical threshold of DingTalk in terms of artificial intelligence, Cheng Cao Hong (Babu) mentioned that AI technology is not just a competitive barrier at the technical level, but what is crucial is the cultivation of talent and the support of resource frameworks. He predicts that with the popularization of large models and the emergence of open-source projects, technical barriers will no longer be a problem.

Cheng Cao Hong believes that the core competitiveness of DingTalk lies in the memory capabilities of its Agents. An Agent’s memory is based on user data, such as user favorites, chat histories, and more. With user authorization, this data can be used for personalized assistants. The logs and documents on DingTalk reflect the users’ original memories and preferences, while information such as friend relationships and organizational processes can provide valuable resources for enterprise management and marketing. Therefore, DingTalk’s AI ecosystem is not just a display of technology but an intelligent system deeply integrated into users’ work and daily life.

In the current era of digitalization, perceptive abilities have become crucial for understanding and managing business events. Individuals separated from the vast collaborative network will not be able to accurately grasp various situations, such as not knowing which employee is late today, or which contract process has been delayed by three days without approval. However, platforms like DingTalk can easily control this information by integrating data. This brings out a point that the importance of using artificial intelligence (AI) thinking is often more significant than the technology itself.

The reason Artificial Intelligence (AI) is so powerful is because it is backed by a vast quantity of high-quality data. Enhancing data density and quality – providing AI with a continuous supply of “nourishment” – is particularly crucial. Data is seen as a key element of production, but the issue is that the quality and density of past data have often been unsatisfactory. Insufficient digital coverage of scenarios is one of the main reasons. Many real-world situations are not digitally transformed, and thus cannot be fully recorded. For instance, without proper processing, the perspectives and information encompassed in a deep exchange and discussion can only be superficially browsed, rather than utilized further.

In collaborations between enterprises and external partners, enhancing data density and quality is of great significance. An enterprise’s digital upgrade involves not just the optimization of organizational governance, the development of business models, and the improvement of organizational efficiency; it also means that utilizing the right methods to achieve all these is extremely important. Traditional methods, such as purchasing expensive professional software or conducting complex custom developments, often do not fully meet an enterprise’s own needs. With swift changes in market environments and business models, outdated marketing systems may require substantial upgrades in a short period to maintain uniqueness and competitiveness.

With this in mind, a few years ago DingTalk introduced low-code technology to lower the thresholds of custom development and accelerate the speed of application development, striving to solve various challenges faced by enterprises. Today, low-code technology is widely used and has helped enterprises rapidly enhance their digital capabilities.

In the field of information technology, low-code development, though convenient for solving problems, still has a limited range of abilities. Traditional professional software might only cover a few scenarios, while low-code can handle an increased range, likely about half of the overall scenarios, yet the remaining ones have not been effectively addressed. As time progresses, the development of Artificial Intelligence (AI) technology may bring more profound impacts. Currently, it seems we have only solved a small portion of the problem, with the vast majority of scenarios still untouched. The key is that there is still a wealth of multidimensional information yet to be integrated and utilized.

A major obstacle in early digital transformation has been the need to structure data. Because early information systems could mostly only handle structured data, with unstructured data, limited by technology, we often only stayed at the “storage” level. However, even though some progress has been made in areas such as image recognition, a great deal of information, such as human expressions and subtle communication signals, cannot be accurately captured and understood by systems. This has led to a significant loss of detailed data.

The advent of AI has brought about new possibilities. Through its multi-modality, the memory and action capabilities of agents, AI can “operate” systems more like humans and reuse existing systems. It has introduced improved models for data connection and collaboration and has participated in building interactive networks. While it’s unpredictable how many gaps AI can fill, one thing is certain: it will further expand our ability to cover data processing.

This expansion also means that data density and quality will receive a significant boost. Only by first addressing the issue of scene coverage can we further enhance the value of data. Afterwards, by solving data quality and density issues with AI, and combining Agents with large models to process the data—whether for use in pre-training materials, memory, or as indicators for perception—will help to elevate the value perception of scenes. Recognizing the value of data, various business processes such as expense reimbursement, financial management, marketing strategies, and even training, policy formulation, and resource allocation will become more efficient, leading to substantial transformations.

In summary, AI acts like a powerful driving wheel added on top of low-code, promoting the continuous enhancement of data quality and density, and creating a positive cyclical momentum. As for when this pivotal turning point will arrive, that requires our continuous effort and attention, as its timing is not easy to predict.

With established product planning and market forecasts, this year might witness significant transformations. Current AI technology still holds certain uncertainties, and to optimize its effectiveness, we are trying to streamline the responsibilities of each Agent, by improving workflows and human involvement, and employing Multi-Agent systems. However, this involves the societal acceptance of AI technology, requiring people to understand the differences between AI and traditional software systems. If we expect AI to have the same 100 percent accuracy as old software, we might encounter difficulties. Thus, deeper contemplation is required in organizational governance and the development of new mechanisms with AI.

Take programming as an example; even if the participants are experts, crucial systems still require multiple checks after completion. For instance, in Tmall transactions, at critical moments like Singles Day, it’s standard to perform double or even triple verification to avoid serious consequences in case of a problem.

Therefore, when using AI, we cannot simply replicate old thinking patterns, management methods, or tolerance standards. It is important to consider how to reasonably allocate the division of labor among Agents, as well as how people collaborate with Agents. These issues may require in-depth thinking, and all of this hinges upon two aspects: one is the product base layer, where significant developments are expected this year; the other is the market environment and the public’s understanding of AI application models. Just as when cars were first introduced, accompanying issues of oil prices and noise were challenges that needed to be resolved.

About the transformation of AI thinking, the public may witness changes in the interaction between computers and mobile applications. Regarding some media reports about DingTalk becoming the next-generation internet portal, in reality, we have not planned to make DingTalk an internet portal. Instead, DingTalk is envisioned as a super AI assistant platform, we hope it will be more like a foundational base, carrying capabilities such as AI, data consumption, and product information presentation, delivering this base to various industries to establish a network of connections.

In the future, Internet gateways will be ubiquitous, involving diverse hardware fields, and enterprises themselves will have their own operational gateways. They may simply use DingTalk as a foundation to strengthen their business capabilities, and some customers have already modified DingTalk, such as Zhejiang Province, which refers to it as “Zhezheng Ding.” Therefore, we focus more on making DingTalk a strong base that supports customer business and organizational governance networks.

In an era where technology is rapidly advancing, it is crucial to recognize that artificial intelligence (AI) technology is no longer an insurmountable barrier. By establishing this core competency and building a healthy ecosystem—an environment that both attracts participation and ensures benefits for all—we have accomplished a significant mission. Furthermore, this ecosystem enables individuals and small organizations to efficiently solve tasks that they could not complete on their own.

One of the objectives of DingTalk is to create a foundational setting for industrial collaboration and network effects, prompting all parties to make progress together. In some ways, DingTalk has been gradually stepping back, leaving more of the spotlight to the customers, while providing a range of basic functions and service support.

In fact, the participation of our AI assistant is also to attract more attention and to showcase the guidance we hope to provide in this field. Sometimes customers understand this even more than we do, creating solutions on top of the infrastructure that surprise us. We aim to influence the bigger picture through small details, not to control everything. As the network effect continues to expand like a snowball, the quality and density of data keep increasing, and a highly collaborative network will produce many interesting outcomes, greatly enhancing the utility and value of AI.

The emphasis on high collaboration and ecosystems means we need to work with more partners, including empowering companies to ensure all their internal systems and applications can connect with each other. Therefore, we are committed to lowering the barriers to these integrations.

Lowering the barriers to data usage and the transition of products towards AI are tasks that we must face and address. These are being achieved through the concepts of low-code, AI capabilities, and “Agent.” We hope that Agents can learn to operate existing applications like humans, thereby simplifying the entire process. In the business environment, this capability allows people to consider the practical issue of whether they can receive immediate returns on investments.

When it comes to user interfaces, with the emergence of “large models,” people have begun to explore the revolution of LUI (Language User Interface). For the future interface interactions of DingTalk, a transformation is expected. This change is currently underway and will bring a new interactive experience to users.

Our latest tool “Cool Applications” has started to enter the development phase two years ago. It was initially designed to solve current software usability issues. Most collaborative software, such as the excellent tools used by Belle, are group-centric and handle all collaborative activities in a centralized manner. However, the problem with many complex software currently lies in their high cost of use and their detachment from actual user scenarios.

Imagine, when you are thirsty, the ideal situation is that you could immediately get a bottle of water, instead of having to trek in search of one or fetch it from the refrigerator. Similarly, when you are tired and want a cushion, one should be readily available nearby. In enhancing user experience, we believe that applications should be more flexible and fit various real-world scenarios.

For example, if you need to adjust inventory, the system should provide a simple “Transfer Stock” button; when you need to increase inventory, just a light tap on the “Add Inventory” option will do. Currently, many software functions are buried in deep menus, requiring users to spend time and effort searching. This is why we focus on the issue of “data density”; the cost of using software is simply too high.

We emphasize on reducing the difficulty of the search process, letting functions appear automatically in front of the user, rather than forcing the user to seek them out. The introduction of “cool applications” is to proactively provide the users with the necessary functions, just as delivering the functions directly to their hands. Using artificial intelligence (AI), the experience is as smooth as water automatically dripping into your mouth when you’re thirsty.

Over the past two years, we have been encouraging our ecosystem and customers to crack and simplify complex software functions and integrate them into DingTalk’s documents, instant messaging (IM), and even audio and video conferences. In this way, when you need information, a single click is all it takes, very convenient. We inspire everyone to do so through the openness of the underlying architecture. As long as you’re willing, this platform is always open to you.

Despite functions being within reach, users sometimes forget to use them. Our first-step strategy is to integrate functions into groups, the so-called “cool applications”; the second step, with the help of AI’s power, directly interpret natural language, voice, and even video. Imagine, when a respected person sends you a video, and you need to understand and respond, AI can intervene at this moment to provide suggestions, which is clearly different from the traditional software interaction experience, because AI has deeply integrated into your actual usage scenarios.

When discussing user interfaces, the conversation often involves graphical user interfaces (GUIs) and technical issues. But fundamentally, the transformation is about seeing AI as your personalized companion, always ready to serve you.

In the past, only the leadership of companies might enjoy good data insight services, which provided them with in-depth information analysis, yet it didn’t offer them much in the way of substantive advice. But the current situation is changing, every employee can now receive a personalized, customized service experience.

Therefore, I set two goals for my team: The first is that each member creates at least one assistant in the first half of this year and ensures that these assistants are actually used. The second goal is that, in addition to the default assistants, everyone should learn and use at least six more. The key point here is that if you do not master the use of these tools, then your way of thinking will not change at all. As you mentioned before, we are experiencing a major shift in interaction methods, and the current trend is like communicating with a person.

If we take a look back at the software market, we’ll notice an interesting similarity. Typically, senior management of companies do not directly use the system; they prefer to assign specific tasks to subordinates, like instructing the HR department to obtain data on salary distributions. This reflects human instinct; everyone tends to seek convenience and save effort. For instance, when you are comfortably lying on the beach in Sanya, you would certainly prefer to simply instruct your assistant: “I want to go home tomorrow, please book my flight.” Nowadays, we crave systems that can offer such convenient services.

In the aspect of technological cooperation, can the encouragement of partners and ecosystem players to build their applications on top of the underlying systems be viewed as the so-called “No App” trend? In the future, as our cognition and dependence on Agents increase, the traditional app-centric thinking might gradually dissolve. For instance, when you’re hungry and want to eat something, you may prefer simply telling your assistant to order takeout for you rather than opening an app to choose for yourself.

As Agents get to know us better, they can provide services that are more precise to our needs. It’s like it might know you have caught a cold recently and would suggest, “Although I understand you like spicy food, considering your current health condition, or the important presentation you have coming up, I recommend ordering something light. How about seafood porridge?” In such scenarios, the plethora of apps on traditional smartphones might lose their significance.

Therefore, our core focus might be shifting towards building the best Agents, continually improving the experiences and cognitions of individuals and organizations, and adapting to this transformation. This could be the real issue we need to focus on.

So, will the App stores of the future evolve into Agent stores? Indeed, this also means that the current market competition might see a new shift, no longer just a battle between app developers or a rivalry between internet platforms, but a new battlefield for services and experiences.

In the future trends, smartphone manufacturers might venture into broader portal fields, such as hardware portals, automobile portals, etc., becoming comprehensive Agent stores. Then, what exactly are the core capabilities and barriers for Agents? I believe it still comes down to the accumulation and processing capability of data and the way of collaboration. The compactness of the collaboration chain and work efficiency are critical factors.

The ultimate product barriers lie in the product form, whether they can continuously innovate and upgrade, whether they can quickly establish a mechanism for data circulation and a positive cycle within the ecosystem. Moreover, whether they can fully realize the support for collaboration. In other words, between people, existing hardware systems, and Agents, can they handle and cooperate seamlessly? Development habits, ways of collaboration, and the support of diversified carriers, along with information processing efficiency, these are all essential elements for comprehensive barriers.

Regarding DingTalk’s underlying system, I am convinced that it could potentially evolve into an Agent Store in the future. DingTalk is essentially a base platform and with the creation of more and more Agents on it, naturally, an Agent Store would be required so that users can acquire Agents for promotion of sharing and exchange across various industries. For example, personally crafted Agents, like those by master Zisha teapot artists, photographers, or AI matchmakers, have already transcended conventional online boundaries, as the network of DingTalk is not confined to a single internal organization.

The organizational collaboration within DingTalk is divided into two types: one is internal organizational collaboration, where people can set boundaries for cooperation; and the other is cross-organizational industry chain collaboration, which can even extend to the entire society. For instance, we might not be in the same organization, but we can still become friends, and your assistant can also join our group chat, thus greatly promoting the efficiency of collaboration.

In the initial planning for the Agent Store, we launched the Assistant Market on April 18th, which includes many agents from our ecosystem, developed by a multitude of developers, clients, and DingTalk itself. There is great enthusiasm for this market, and we are organizing related competitions. The goal of this launch is more about letting everyone experience the diversity in forms and unique features of agents. Additionally, we have conducted continuous upgrades to the product for two months, improving workflows and memory capabilities. In the future, agents will be able to perform an even more diverse range of complex tasks, which is very exciting.

When it comes to mobilizing the ecosystem, developers, and internal staff, the three groups have very different situations. For internal staff, especially those who are keen on AI technology, they are basically acting on their own initiative and actively participating in agent development. In the early stage, when others were still hesitant, I had already started creating my own agent.

Many people initially wanted to just focus on their professional areas and adopted a wait-and-see attitude, but we took a firm stance: non-participation would result in assessment. This signifies that participation is mandatory. Now, as everyone gradually embraces new challenges, their passion has been reignited. Over the past two weeks, we added 150 new agents. Among these new agents, some have performed exceptionally well, fully demonstrating the infinite potential of human creativity.

In terms of staff participation, we have adopted a strategy both mild and forceful. For customers, I believe it is critical to seek shared value. This is because the spirit of co-creation among customers is particularly prominent on the DingTalk platform. As soon as there is a new release, some particularly active customers will immediately take the initiative to make contact. These customers usually innovate according to their own actual circumstances and are active contributors within a large community. For example, some have unique ideas, like the Hangzhou Public Security Bureau, which has created three eye-catching works that serve the public and involve many other aspects. Including in fields like agriculture, numerous customers are actively co-creating.

Moreover, customers themselves have noticed this change and are particularly eager to find ideas that can bring breakthroughs in their own fields or work environments. In terms of the ecosystem, we believe that the design of its business model must be carefully considered. After all, every participant is essentially an entrepreneur, and many members of the ecosystem are starting businesses with their own resources. Therefore, fundamentally, we cannot simply rely on personal intuition to demand changes from them, we must also consider issues from the perspective of the healthy development of the business ecosystem.

We need to guide them gradually, whether it was creating cool applications in the past or developing agents now. This is to showcase their solutions to DingTalk users and various groups, to make their ideas clearer, and to let them feel that synchronizing with product upgrades is a great thing. Next, if they do release some high-frequency usage agents for specific scenarios, this will significantly increase customers’ interest in them, whether for renewing services or new purchases.

I believe it is crucial to grasp the key point of commercial interests. Our open platform mechanism will complement this, with the Agent Store itself being jointly operated by the open platform team and the operations team.

The attitudes and actions of a business often have a close relationship with the organizational culture it is part of. Some customers pursue flawless processes, which usually relate to the characteristics and emphasized culture of their organization, especially in areas where there is extremely low tolerance for errors. In the case of Zhejiang Zhengding, the benefits of digital construction are obvious, with over 1.6 million active daily users on the digital platform. Seeing these benefits, they encourage an innovative atmosphere and continuously push for the development of applications.

Some department employees even work until the early hours, motivated by their desire for results. Meanwhile, some companies may realize the limitations of their current models and begin to seek new methods, exploring new business models. Facing various challenges, like in the greenhouse planting industry, if a base has achieved success in planting technology, they start to consider how to effectively replicate these experiences elsewhere. They also take into account market demand, such as how to effectively convey the market’s preference for the sweetness of blueberries to the production base. They seek to solve these problems through digitalization and intelligent methods, continuously optimizing the connection between production and market.

Within the company, for employee training and customer Q&A, using functions like “smart customer service” can allow users to quickly experience significant advantages. Since many enterprises possess a large amount of professional knowledge, especially in the electronics industry involving numerous materials, this knowledge is difficult for the average person to fully grasp. Therefore, when enterprises make internal knowledge available to employees or customers through tools like smart customer service, the value can be immediately realized, leading the enterprises to be willing to invest in such tools.

Despite various challenges the economy may have faced over the past few years, businesses are paying more attention to the cost-benefit ratio when investing in innovative projects. DingTalk stated that last year, the average cost per call on the large model platform was only about 0.05 yuan, and DingTalk has nearly 700 million users and has launched a version for individual users. Under these circumstances, how to balance application cost and business value during mass usage becomes a key consideration.

Mr. Babu (Cheng Caohong) mentioned that cost is a factor that must be considered. But at this stage, they encourage customers to co-create high-value scenarios and form exemplary models. This can greatly promote industrial transformation, and DingTalk and its parent company Alibaba have been actively investing in this area. They hope to have more ecological partners join the field of artificial intelligence, and provide subsidies for these partners. For ecological partners who have completed AI restructuring, DingTalk will also offer special commission discount strategies. In the usage of artificial intelligence, DingTalk currently does not charge much. Although this is a short-term strategy, from a medium- to long-term perspective, maintaining such a model is clearly unsustainable. Ensuring model performance, cost and charging standards will be important considerations, especially as the scale of the model continues to expand and costs sometimes even exceed the initial estimate of 0.05 yuan.

We always insist on optimizing the application strategy of our models without increasing customer costs. For example, to avoid using a large model for every repeated request for similar images, a mix of models of various sizes becomes an important cost optimization approach, particularly in scenarios where small models suffice, reducing unnecessary resource consumption.

What’s more noteworthy is that we are pursuing not only cost control, but also value creation. When artificial intelligence is genuinely recognized by customers and significantly saves labor and material resources, the direct experience and willingness to pay are undoubtedly enhanced. This prospect is clearly superior to the traditional Software as a Service (SaaS) model.

Customer/User Feedback

InfoQ: During the process of applying artificial intelligence in content generation (AIGC), what do you consider the most valuable customer/user feedback?

Cheng Caohong (Babu): Based on customer experience and objective data analysis, the message summary function is undoubtedly frequently used, especially for heavy users with a massive amount of information. This feature can effectively summarize and review a large number of unread messages or key tasks, and performs exceptionally well in organizing calendar messages. Additionally, real-time information transmission is also crucial in instant messaging (IM), where the image tail feature is particularly popular. Whether sending documents, log links, or lengthy articles, it helps users quickly skim and read.

The third aspect is that users are starting to create applications similar to AI assistants, such as their popularity on DingTalk, which highlights everyone’s needs and innovation. AI assistant applications are constantly emerging, with different users setting up various small assistants according to their needs. These may help children complete homework or serve as data analysis tools for companies, providing automated detection of essential information. These small assistants are becoming increasingly personalized and popular.

In addition, many built-in intelligent functions are quietly serving users, such as intelligent features in document processing. Modern text editors already incorporate tools to help refine language, create mind maps, and even generate presentations (PPT) at the click of a button. These smart tools are being widely utilized by an increasing number of knowledge workers, significantly improving work efficiency.

In today’s fast-paced work environment, audio and video meetings are becoming more common, and an emerging feature accompanying them is “flash notes,” which allows users to quickly generate summaries after meetings, including textual abstracts and some PDFs with images, greatly promoting the efficiency of understanding meeting content. The use of this feature is gradually increasing, with varying needs guiding preferences for its use.

For teams, adapting to changes brought about by the development of large models is an ongoing process. Last year, we profoundly felt the shift in the software development paradigm. Large models now have advanced planning abilities, can process voice, and recognize speech intonations, meaning that development teams need to understand and master the best capabilities of large models thoroughly.

For programmers of the past, understanding the many features of operating systems was essential, and now, understanding large models is equally crucial. In the future, we may no longer rely on specialized programming languages, but instead turn to natural language. Based on this prediction, we might need to relearn natural language and build on it with prompts and fine-tuning. We have established a dedicated team to handle this fine-tuning work, which requires not only an algorithmic background but also more people to understand how large models work.

Furthermore, we also face the challenge of how to integrate software engineering or application engineering with large models. We need to look at existing assets in a new way of thinking. Taking DingTalk as an example, we have a low-code application platform like Yida, open APIs, and a wealth of data assets such as OA approval processes. How to effectively combine these assets with the latest Agent form, how to transform data assets into Agent’s memory, and how to convert low-code applications into actions that Agent can execute are issues that we need to think deeply about.

To adapt to these changes and achieve this goal, our technical committee operates continuously. Last year, we held AI-themed events for two consecutive months, with weekly all-staff sharing sessions, focusing on learning Prompts, understanding model features, familiarizing with programming systems and frameworks based on large models, and how to combine DingTalk’s core data application interfaces and other practical experience.

In our technical field, we always encourage innovative thinking, transforming routine operations like quizzes and medal awarding into new modes with artificial intelligence. We should rethink our understanding of artificial intelligence concepts, such as when facing challenges in operations and capacity planning, we should break away from traditional patterns and use AI methods for capacity planning.

Rather than being confined to building applications, establishing processes, or process approvals, we should introduce computational resources and cloud resources at the application layer, and collaborate with an AI assistant called “resource management expert” to provide new intelligent planning. This AI expert does not need to develop complex systems, nor relies on manual labor to compensate for deficiencies, and can efficiently implement capacity planning and answer related questions.

This approach changes the original work pattern where team members had to deal with various inquiries every day, such as questions about “Why am I throttled by 500,” and now the AI assistant can participate in the entire process from early consultation to actual planning to providing recommendations. It may even proactively suggest that the business can expand capacity, creating a completely different work experience.

This is an upgrade in thinking and philosophy, and by continuing to implement these changes, some people will stand out. The recruitment section also reflects the new way of thinking, examining candidates’ understanding of artificial intelligence, grasp of technological progress, understanding of large models, knowledge of the prompt concept, and corresponding insights.

In the process of driving technological change, we advocate a combined approach of hardware and software, improving from various aspects such as culture, organizational structure, process improvement, and internal members are also actively creating Agents, practicing the philosophy of “using one’s own products.”

Looking ahead, predictions on the trend of artificial intelligence include the trend of “No App,” suggesting that more personalized assistants will emerge in the future, and operations can be completed through simple voice interactions. Additionally, it foresees the transformation of individuals and organizations, with an increase in smaller organizations, while large organizations may seek new dimensions of development at the capital level, holding level, or market level.

With the advancement of artificial intelligence (AI) technology, organizational patterns and service methods are undergoing tremendous changes. We have attempted to establish small, flexible organizations similar to amoeba models, but they were difficult to implement in the past due to technological complexity. However, empowered by AI, these organizations will operate more flexibly and have more comprehensive empowerment in the future.

These small organizations will be better able to utilize the conveniences brought by large platforms and conglomerates, and their independent operational space will be greatly expanded. With the support of AI assistants, as well as more efficient data flows and better collaboration methods, small organizations will become mainstream.

AI assistants serve organizations as well as individuals, and their service models will become increasingly rich and varied. Just like individuals can become internet celebrities through videos or live broadcasts today, AI assistants will promote more diverse individual development. Although they may not be large in scale, everyone has the opportunity to showcase their own specialties, such as ant farming, blueberry planting, or pickling vegetables. They can transform these skills into assistants and spread them through collaborative networks.

We anticipate that more such small organizations will appear in the future, and even a single person could form an organization, which is a very interesting development trend. In this trend, the increased degree of digitalization and the deepening participation of AI assistants in collaborative networks will lead to reduced reliance on human labor in fields such as resource allocation and project management.

As many work natures are shifting towards pragmatism, outstanding individuals can be identified by AI systems through the analysis of their work performance and be given priority consideration in future work assignments. AI is able to allocate resources and organize personnel more effectively, which is not only a farsighted idea but also a very interesting direction for development.

The conference is currently in the early bird ticket sale period, with a 20% discount. You can call 17310043226 to contact the ticketing manager for the latest discount information. At the same time, you are welcome to scan the QR code to add the conference welfare officer and receive your customized welfare gift package.