Source of article reprint: AI DreamWorks

Source: Titanium Media

Author: Zhang Shuai

Editor: Liu Xiangming

In the three years since Ye Jun came to DingTalk, DingTalk has been on the road to crushing itself, which is to turn a bloated DingTalk into a DingTalk with a better experience.

Image source: Generated by Unbounded AI

On the other side of the ocean, OpenAI's previously delayed GPT Store is about to be officially launched. At that time, Bill Gates wrote a blog post after the OpenAI Developer Conference, "AI is about to completely change how you use computers."

In his view, Agents will not only change the way everyone interacts with computers, they will also subvert the software industry and bring about the biggest computing revolution since typing commands to clicking icons. "In the computing industry, we talk about platforms - the technologies on which applications and services are built. Android, iOS and Windows are all platforms, and Agent will be the next one."

Just as the wind of big models blew to China, the craze of Agent in the other side also swept to this side. According to the "Terms of Computer Science and Technology (3rd Edition)", Agent is a software or hardware entity that embodies one or more intelligent characteristics such as autonomy, responsiveness, sociality, proactiveness, speculation (deliberation), and cognition in a certain environment.

AI Agent means that the big model has limbs and five senses, and can take action to execute commands, or the robot has a smart brain, an intelligent entity that can perceive, make decisions, and use actuators to respond. This is an imagination that previous agents did not have.

The industry is celebrating again and flocking to AI Agents, but new problems soon emerge. The challenge for manufacturers is how to define AI Agents. Only when there are enough users and the demands are sufficiently concentrated and layered, can AI Agents meet the needs of users. Otherwise, it is just a self-satisfaction from the perspective of manufacturers, rather than the form of AI Agents that users expect.

Among all AI agent vendors, DingTalk is one of the most distinctive representatives. When big models were just becoming popular, DingTalk announced its full intelligence earlier. It did not make basic big models or industry big models, but used big models to transform DingTalk itself.

In the early days, many people did not quite understand DingTalk's enthusiasm for intelligence. Although everyone knew that they wanted to do it, there were differences in how to do it, when to do it, and the degree of investment. Only DingTalk showed an extremely firm determination.

Over the past year, DingTalk's intelligent evolution has been very rapid:

In April 2023, DingTalk announced that it would be fully intelligent, and all products would be reshaped with a big model. In the following 100 days, DingTalk's 17 product lines completed intelligent reconstruction. In August 2023, DingTalk opened the intelligent base (AI PaaS) to ecological partners and customers, using big models to help the ecosystem remake products, and launched the innovative product "Digital Employee" based on AI PaaS, as well as a variety of intelligent scenario solutions and intelligent industry solutions. This marks that DingTalk's intelligence has fully entered the ecological layer. In November 2023, after more than 500,000 companies joined the DingTalk AI invitation test, DingTalk AI went online, and 17 product lines, 60+ scenarios, and nearly 100 AI skills were fully open to users for testing. This is the first national-level work application in China that fully opens AI. In January 2024, DingTalk's personal version of AI was fully launched. On January 9, 2024, DingTalk 7.5 launched products such as AI assistants, and everyone and every company can customize a personalized and exclusive super assistant. This is the latest judgment based on the AIGC wave entering productivity and application scenarios: AI Agent is the best entry point for AI applications, and DingTalk aims to become a low-threshold, high-frequency, open AI intelligent assistant platform.

As a result, DingTalk's strategy and prediction of the intelligent era have become clearer. In the three years since Ye Jun came to DingTalk, DingTalk has been on the road of breaking itself up, as shown in "DingTalk Breaks DingTalk" and "DingTalk Breaks the Ecosystem", which is to turn a bloated DingTalk into a DingTalk with a better experience.

So DingTalk made a series of changes, from the top-level product design, user experience, detail experience and other aspects, to make DingTalk, a B-end product, more user-friendly to C-end users. In the past, the means by which DingTalk smashed itself could be understood as sledgehammers, small hammers, chisels, and iron wedges. After becoming intelligent, DingTalk has epoch-making new technical means. Using a magic hammer to smash DingTalk, all logic becomes different.

By 2024, DingTalk will have become a platform with 700 million users and tens of millions of corporate customer organizations. The huge number of users will create a vast number of scenarios, and the key to AI Agent is to maximize its capabilities in user scenarios. The user needs accumulated by the DingTalk magic wand will become the nourishment of AI Agent.

In the past, DingTalk was defined from its own perspective as a collaborative office platform and business development platform. Looking back at DingTalk today, it is a low-threshold, high-frequency, open AI smart assistant platform that is more user-oriented. The most exciting thing about AI Agent is that it will democratize expensive services, which coincides with DingTalk's strategy.

When DingTalk, a company that creates deep value, has an open ecosystem, and offers low barriers to entry, meets AI Agent, which naturally possesses such attributes, there is no reason for DingTalk not to jump on the express train to the future. How will the new productivity shape the new DingTalk? On the occasion of the release of DingTalk version 7.5, DingTalk President Ye Jun (nickname: Buqiong) and Titanium Media Group co-founder and co-CEO, and ITValue founding director Liu Xiangming, once again had an in-depth discussion on disassembling AI DingTalk in the new era.

AI Agent is the best form at this stage

Titanium Media: AI is developing very rapidly. What is your judgment on its future development trends?

Buqiong: Last year, the AI ​​industry was hot like a conductive heat, which spread from the United States to China, mainly large models. I think the situation will change next year. The large model has been in place for a year, and the change from buying and hoarding cards in the first half of the year to not knowing what to use the cards for in the second half of the year is very obvious. We saw it more clearly at the beginning, and DingTalk did not have a large model. I think returning to the basics is to create value.

We predict that next year, the big model will not play the leading role, but it will continue to improve. The impact of its improvement will not be as great as last year. Next year, products, that is, C-end AI native products and B-end AI products, will become the leading role.

I believe that the AI ​​Agent concept will be effectively implemented in many business scenarios in the second half of this year, which is also the direction that DingTalk is focusing on exploring in the release of version 7.5. We are taking the initiative to move forward, that is, the assistant AI Agent form. By moving forward through the product, we can let everyone see that DingTalk is a platform for the use and creation of AI Agents. This is the development trend and prediction of products next year.

Regarding SaaS, the big model was hot last year, while SaaS was cold. SaaS has been advocated in China for more than 10 to 20 years. In addition to the macro environment and technological changes, SaaS does have some unresolved technical issues. Therefore, SaaS is an intermediate form in the development of China's software industry. SaaS will appear in the future, but the form of SaaS will change and become another form, from an intermediate form to a more practical and inclusive form. This is the general direction.

We believe that the future will be towards new SaaS, which can be called new SaaS if it meets the following characteristics:

The first is fragmentation. SaaS is not as complex as previous integrated software and may be more fragmented.

The second is minimal interaction. In the future, SaaS will definitely use AI capabilities for interaction. Simple interaction does not mean that the previous user interface will no longer be used. It is still LUI+menu-based multi-modal interaction. In the future, interaction should be richer and more minimalist. Apple mobile phones have created a new touch-based interaction. New SaaS will also create a new interaction, which is LUI+conversation, voice, photo, menu collection multi-modal interaction, which will become the feature of new SaaS.

Third, AI Agent will become a standard feature of SaaS. In the future, each SaaS will come with several Agents. Agents can help SaaS better complete batch workflows and business flows, replacing the core problems that the original software solved. In the past, software was used to solve workflow problems, and using AI Agent assistants can make procedural events simpler.

There may be even bigger changes next year. I predict that if large-scale model applications become mainstream, another form of "Zhang Xiaolong" will appear in key scenarios next year, and there will definitely be a successful product manager for a certain product.

The popularization of many technologies depends on these key people. We don’t know where these people will come from. They may be in a startup, a large company, or an entrepreneur who has returned from overseas. The next Zhang Xiaolong will be a younger entrepreneur with more ideas. These are my three core predictions for the future: AI Agent, new SaaS products, and new product manager roles.

Titanium Media: AI Agent is the focus of this release. We all come from programming backgrounds. From Program to Application, and then to Assistant Agent, do you think this is another big change?

Infinite: The transformation from traditional PC software to Internet software is a change from CS to BS structure, which truly realizes the distributed system. From PC to mobile Internet, the devices have changed, so the entire technical architecture has changed greatly, even the language has changed, design habits, user operation interaction, and the keyboard that was used before is no longer used now, and it has become another form of product.

The AI ​​Agent intelligent era will also see a change. The devices will still be the same devices, but perhaps the only change is that they will become more diversified and multimodal. There will be many single small devices and special devices, such as more and more unique devices in certain scenarios of IoT networking.

No matter how the devices change, I think this change is different from the last one. The main device has not changed and is still based on the mobile terminal. The system has been constructed and you can still use Hongmeng, Android, and iOS. No matter what kind of device it is, you can use these systems. Compared with the BS architecture of the previous PC era, from Windows to iOS and Android, these have not changed.

What has changed is the intelligent body AI Agent, including the product form of AI assistant, which will exist in a symbiotic state with traditional apps for a long time. Now, innovation and research and development, from the creative system to the product interface design to the product distribution system, will be connected to the old system to a certain extent. In the past, mobile Internet was almost abandoned and a new system was developed. This time, there will be a lot of interactions between the old and the new.

In addition, the No App trend will become the general direction. This process will be a bit long. The premise of No App in the future is the continuous development of AI Agent and smart assistant products. When intermediate products and fragmented products become extremely simple to enter, No App will truly appear.

Everyone is looking forward to the super apps in the intelligent era. In the future, super apps may have great opportunities on mobile phones. The first interface of the mobile phone may replace the original entrance to the traditional super apps and take up a lot of users' time. I think there is this possibility.

DingTalk also has opportunities. With the cooperation between DingTalk and mobile operating systems, including Huawei’s Hongmeng, many of DingTalk’s functions will be directly mapped to the Huawei screen and displayed directly like a small AI Agent. I think AI Agent is also a fragmented intermediate state. In the future, AI Agent can interact with itself. At this time, the product status will be a little different from the original Taobao and Douyin.

Therefore, the fragmented processing capabilities of AI assistants, such as their perception of the environment, memory of past historical data, and ability to execute batch processes, will become an important external connection port for popular e-commerce, gaming, short video, and other platforms. This connection port will become very, very high-frequency. Interaction will be very important in the future. The closer one is to the interactive interface, the more ability one has to influence user decisions.

Titanium Media: What impact will the emergence of AI Agent have on the future software development ecosystem?

Not poor: In the past, software development was more about SaaS or mini-programs. You either developed mini-programs to make some money or get some commissions and advertising fees, or you developed SaaS to sell to customers. These are the two mainstream forms of software development, and the other is customized projects.

In the future, when agents are available, first of all, all SaaS software vendors should develop agents simultaneously, which is the new SaaS I just mentioned. If AI agents are not developed, I think SaaS customer acquisition and user experience will become very poor. There is no harm without comparison.

In addition, new entrepreneurs, including college graduates, do not need to do SaaS from start to finish. They can develop their own business logic based on the AI ​​Agent framework platform and focus on business logic. They should not develop a SaaS from start to finish, and they should not even do the registration and login by themselves. Because the platform PaaS capabilities are getting stronger and stronger, try to develop based on the platform.

Now, the top ten AI products in the United States are all developed based on the platform. This is a high-probability event. Creators should focus on business-related matters and gain customers through business value innovation. In the past, traditional competition relied on traffic, and the essence of traffic is information asymmetry. In the new AI Agent era, quality value and scenario function value may become the focus.

The third is mini programs. In the past, mini programs were displayed by pulling down the negative one screen on WeChat. To put it bluntly, they were products for acquiring traffic and customers. Mini programs are a form of the past. In the future, the symbiotic relationship between AI Agent and the main application will be very close, and it will coexist with not just one main application, but many main applications.

With the development of agent intelligence, mini programs will gradually switch to a new state of intelligent entities. The design of mini programs is function-centric and application-centric, while AI Agents are truly people-centric. They are closer to people, and their core logic will affect many of people's decision-making judgments. They are centered on people's personality and use, and personalized processing. This will gradually replace or substitute mini programs.

In addition, mini programs do not have the ability to execute automatically and are largely passive, but intelligent agents are different. They perform many operations in batches and focus on execution and decision-making, which is fundamentally different from information systems. AI agents are driven by model neural networks, while mini programs are still driven by human clicks.

In the future, DingTalk's cool applications and low-code will serve as action systems, the AI ​​Agent big model will serve as a command system, and various functional data on DingTalk will serve as a memory system. The entire system will be integrated into a platform like DingTalk, which will bring many opportunities in the future in the intelligent era.

AI Reshapes DingTalk

Titanium Media: DingTalk has undergone many major changes in recent years and its technology has developed rapidly. Can you review the changes in your understanding of DingTalk and your mental journey? What kind of product do you think DingTalk is now?

Buqiong: Looking back at what I said at the first DingTalk launch conference three years ago, I feel that some of my ideas at that time were relatively simple. There have been quite a lot of changes in the past few years, but some things behind them have not changed.

What has not changed is that we are essentially deepening our customer value. We discovered very early on that DingTalk's value was too shallow, and now all our competitors have this problem. They chat and hold meetings, and there seems to be no difference between having you or not. You can only say that you think you are very important, but you are not that important to customers.

Therefore, three years ago, DingTalk began to add deep value and enter business scenarios, which is our cognitive lead.

The second point is to build an open ecosystem. We have not changed our work on PaaS, because if we want to deepen the value, we will find that we cannot do so many things. We still need to work with the ecosystem and PaaS, especially since the problems in various industries are different. From three years ago to now, we have insisted on supporting the ecosystem and investing in PaaS capabilities. At that time, many people were not optimistic about our investment in low-code, PaaS, and cool applications. In fact, I think the three-year investment is still very worthwhile and has yielded some good results.

The third point is that there is a certain degree of persistence, but there are also certain adjustments. We have always found that DingTalk products are very bloated. It should be said that all toB software have this problem. Every time a new function is added, there are always some people who like it and say it is good, while others say that this function is not needed. There will definitely be people who criticize it. Maybe if 1% of people criticize it, I will feel very uncomfortable, so DingTalk has become very uncomfortable now.

Over the past three years, DingTalk has been improving the experience, from top-level product design, user experience, to detailed experience. It has also invested a large number of product managers internally to make B-end products friendly to the C-end. This has been the focus, but there have been adjustments in the approach. The biggest change this time is the use of intelligent methods to provide the experience, and the idea is completely different.

In the past three years, DingTalk itself has both changed and remained unchanged. If I define DingTalk in my mind now, starting from the end and looking to the future, DingTalk used to be a collaborative office platform and a business development platform. Looking back at DingTalk today, the two platform concepts are more like a platform perspective rather than a user perspective. Today, DingTalk is an intelligent assistant platform, which is more inclined to the user role rather than the platform role.

A characteristic of the smart assistant platform is its low threshold and universal access. This has not changed. We have always talked about universal access before. We also talked about this before we opened up, and we did not do so many things ourselves. Third, it has become very high-frequency, which also determines that subsequent product investment, including our encouragement of the ecosystem, will tend to develop in the direction of high frequency. We hope to promote the network density of DingTalk's collaboration through high frequency, so that DingTalk will not only become a low-frequency communication tool, but we hope it will become a high-frequency smart assistant platform that users cannot do without. This is the definition of the future.

Titanium Media: The bloated problem is indeed a big challenge. People have very different expectations and habits for using DingTalk. Will there be any changes in the main interactive interface of the next version of DingTalk?

Buqiong: Version 7.5 has been gradually released in grayscale. We don’t want to release too many futures, but spot products. Especially the PC interface, everyone will feel it very clearly. The full launch of GUI+LUI will definitely make it easier to use DingTalk after the magic wand is launched.

In the future, when using DingTalk, users no longer need to look for a certain menu function. They just need to use the magic wand to say: I want to project the screen; I want to hold a meeting; I want to share this document with XX; I want to turn off all DingTalk messages; I want to modify my avatar; I want to complete all approvals, etc.

We are gradually strengthening the LUI batch command capabilities. The previous GUI graphical user interface experience will not be affected. In the future, the DingTalk image user interface buttons will become fewer and fewer. The number of interface buttons today may be the peak, and various buttons will be gradually offline in the future. For example, we will officially offline all collections. In the future, there will be no need to collect. You only need to tell the smart assistant to please help me remember this article, please help me remember this message. It has become the long-term memory or short-term business memory of the AI ​​assistant. The user experience will change greatly, which comes from the enhancement of LUI's intelligent capabilities.

With the release of version 7.5, not only the magic wand in the upper right corner of the first screen, but every corner of DingTalk will support natural language interaction. Documents, schedules, YiDa, low-code, meetings, to-do lists, electronic invoices and other interfaces can be used, so it will definitely be easier to use in the future.

The same goes for search. DingTalk is currently designing a completely new search product, which will not be released at this press conference. We want to make DingTalk fully intelligent when it is upgraded to 8.0. Therefore, 7.0 is an important intermediate version. After this version, we will enter a new stage.

Titanium Media: It sounds like DingTalk’s functions have undergone some major adjustments. Will there be major adjustments to the underlying architecture and design concepts?

Buqiong: Yes, there should be several adjustments to DingTalk's product design concept. Before the emergence of intelligence, more adjustments were made to the bottom design to make the product experience in line with usage habits, from the original boss's habits to the boss and employees' habits. This has been a big work done in the past two or three years, taking care of individual usage experience through experience optimization.

The biggest change this time is due to the emergence of intelligence. All functions of product design are first or synchronously called up using LUI. In the past, after a function was completed, it was necessary to find a menu to operate it. A bubble popped up here to tell the user what to do when clicking here. Now LUI will be used by default. What you think is what you get, not what you see is what you get. The premise of what you see is that you see the menu first. If you don’t see the menu, you can’t get this function. So this is the first change after the emergence of intelligence.

The second change is the full launch of the personal version. DingTalk personal version has been polished for a year from last year to now. This time, the personal version is officially released. In the era of big models, personal capabilities will be greatly enhanced. Personal assistants are a very big change in product design architecture. Many functions that cannot be done on standard DingTalk can be implemented in the personal version, and may be put into standard DingTalk in the future.

The third change is that the product architecture design will highlight connectivity. We talked about our general direction at the DingTalk 7.0 launch conference, and this direction has not changed. This is also the main theme for the next few years, which is the era of great collaboration.

Big collaboration means that in addition to collaboration between employees, we also hope that individuals and organizations can collaborate. For example, recruitment and job hunting can be done on DingTalk. It also includes collaboration between peers. We will also emphasize the collaboration between helpers and AI assistants. These are all big collaborations. Therefore, collaboration between individuals and organizations, individuals and intelligent entities, and individuals and individuals will become the big direction, and the stickiness will be enhanced.

On the other hand, the collaboration between organizations will also be enhanced. Since we released the upstream and downstream organizations of external groups in December last year, we also released Dingxuan. In the future, we will further promote the enhancement of various connection capabilities.

I think a product like DingTalk is a bit like Slack Connect. It is simpler than ours. It is more about connecting people with people, external individuals and organizations. The large-scale collaboration that DingTalk is doing now is truly full-connection, which can be better achieved through AI capabilities.

For example, I asked the AI ​​assistant to help me book a hotel and air ticket. We don’t know whether the hotel and air ticket come from Ctrip or Fliggy. This connection is a connection between an individual and an external supply partner. I think these connection capabilities can truly unleash the value of DingTalk.

To sum up, the biggest adjustments in the architectural design ideas are: first, intelligent improvement of experience, changing the user's experience of interaction and design creation, second, enhancing individual product capabilities, and third, fully connecting network density. We hope to make DingTalk break through the boundaries of organizational tools, that is, local area network tools, and gradually move towards full connectivity. This full connectivity is a process that DingTalk will inevitably go through in the next 2-3 years. The emergence of intelligence should make this thing more feasible, and it is much more predictable and certain than when we held the press conference in December last year.

Titanium Media: Everyone is curious about the cooperation between DingTalk and Microsoft. Can you disclose the form and scope of the cooperation?

Buqiong: We prepared for a full year and a half, and finally Satia approved it. Satia was also very curious about the development of DingTalk in China, and thought it was amazing that we had built a platform like this.

The cooperation between DingTalk and Microsoft will first focus on Microsoft M365, the integration of Microsoft documents and DingTalk documents. DingTalk documents will become the most compatible documents in China. We are not only compatible with WPS, PDF and 100% compatible with the full set of Microsoft M365 Office document products, including DingTalk's self-developed Internet documents, similar to lotion. DingTalk is suitable for different work habits and work characteristics in thousands of industries. This is the first cooperation option.

In addition, DingTalk will cooperate with M365 to release a commercial version. M365 will provide a nationwide preferential price version for the DingTalk base, and we will jointly promote the sales of the product.

The third is the next step of cooperation with the M365 App desktop. This time it is mainly on the browser side, and the next step will be to move towards the coordination of various desktop products.

AI assistant for everyone and every business

Titanium Media: We just talked about the new smart assistants. Last time we talked about digital employees. Are there any changes and inheritances from digital employees to smart assistants?

Buqiong: In the enterprise scenario, we upgraded digital employees to intelligent assistants. It is essentially the same product and technical architecture, but for the enterprise side, we think the name of enterprise assistant is more acceptable. Since we have enhanced the capabilities of the C-end this time, the product architecture has been enhanced on the C-end, so it is not appropriate to call it a digital employee. Entrepreneurs and self-employed individuals on the C-end can also have an assistant.

Especially for some small businesses, I predict that in the future, small business entrepreneurs will emerge in this field like mushrooms after rain. It is difficult to start other businesses and the competition is fierce. New businesses should be generated from new tracks and new technologies. It is more appropriate to call them smart assistants than digital employees.

From a technical perspective, the original product architecture, technical architecture, data, models, and application frameworks are all inherited, and the product experience will not change.

Titanium Media: In the past, apps were still programmed, and the relationship between them was relatively simple, that between users and function providers. What do you think about the relationship between assistants and users in the future? How will they evolve in the future?

Buqiong: I understand the user as the manager, and AI as the assistant. The symbiotic relationship is a companion relationship. It will always accompany you, and the longer the time, the greater the value of the assistant will be. So the relationship between the two, AI, is that in the future everyone should have one or more assistants, and everyone should have at least one assistant.

AI assistants may belong to individuals or to companies, which will definitely bring new problems. I haven’t thought about this issue in depth, such as how to divide intellectual property rights. If a person leaves a company, I can isolate the data, but who the AI ​​assistant belongs to is still debatable.

Titanium Media: Indeed, you just talked about symbiosis. Will AI assistants in the future pose data security risks, as they require a lot of data to train them. Will this bring about legal issues? Do you have any plans in mind?

Buqiong: We have indeed considered this matter, but there is no good answer. We are now working hard to inform users which parts belong to you and which parts belong to the enterprise, but this is all we can do.

Titanium Media: If there is a writing assistant in the future, and we need to feed him all the original writing, how can we ensure the data security? Where will the data be used? Many people are worried that the data will be directly used to feed the big model?

Buqiong: All AI assistants on DingTalk will not input data into the big model, whether it is a personal assistant or an enterprise assistant, they are limited to the RAG search enhancement generation layer on the assistant. We are firmly opposed to inputting data into the big model. Since April 18, we have advocated the application layer. I think the big model needs many industry big models and small models, because not everyone is willing to share things with the big model.

Titanium Media: Now AI Agents, including industry large model data, should all be stored locally. It used to be a private deployment process.

Not Endless: Now they are all exclusive deployments within the enterprise. If any enterprise knowledge base document is trained, this model has nothing to do with the big model, and other enterprises cannot access it. It can only be accessed within the enterprise. It is an exclusive deployment and an exclusive model.

Titanium Media: DingTalk has done a lot of work on PaaS. Will the PaaS capabilities based on AI Agent be different in the future?

Buqiong: We launched the first version of AI PaaS at the Ecosystem Conference on August 22 (last year). Based on the AI ​​PaaS capabilities, in addition to increasing the model docking capabilities, it can help users dock and save a lot of workload; in addition, basic capabilities have been added, such as environmental perception, like the previous RPA fast batch processing logic chain, these capabilities are all ready; thirdly, the exclusive training capability of customized training of RAG or vector database models has been added, and users can also perform exclusive training and SFT tuning on the model.

Titanium Media: We have been talking about the transition from software to services for many years, including SOA and SaaS services at that time. I feel that we have never been able to break out of the traditional software development demand development interaction model. You also mentioned that AI Agent can see some substantial breakthroughs. Has AI Agent truly completed the turning point from software to services? Do you think the turning point has come?

Buqiong: I personally think that the turning point has definitely arrived. First of all, the phrase "software as a service" is correct, including the era of intelligent entities, which is also software as a service. In the past, the biggest problem with software as a service was that the service was not good. It was very troublesome to change the requirements, the operation and maintenance costs were very high, and the customer service interface was poor. In essence, the service was not good. In the future, AI Agent AI will solve these problems, the user experience will be very good, the operation and maintenance costs will be very low, and personalization will be easy.

Titanium Media: Can the emergence of AI Agent fundamentally solve the problem of DingTalk's commercialization? In the future, if the Agent service is good, it will be easy to charge.

Buqiong: If our logic holds true, the new SaaS can first solve the problem of SaaS monetization, that is, the problem of software value monetization, because the value will be more concise and intuitive. The greatest value of AI Agent is that it solves the problem of commercialization of large models, not the problem of DingTalk commercialization. There has been no solution to the commercialization of large models. I think we should first help large models solve sustainable income other than selling cards, because selling cards is a monopoly business, but selling services is a long-term business, so AI Agent is a way out for the service of large models.

DingTalk will also benefit to a certain extent from commercialization, but DingTalk does not rely solely on this to achieve commercialization. The basic products of the DingTalk platform itself are also valuable. As Chinese companies become more accepting of digitalization, the value of our audio, video, document and other services is also becoming apparent, so it is the value of the combination of the two.

Over the past year or so, DingTalk's business model has almost come into being. In essence, DingTalk's first growth curve has been established. Now the question is how to find the second growth curve for DingTalk's commercialization. AI Agent has helped DingTalk find its second growth curve.

Titanium Media: Whether it is DingTalk's Agent or other Agents, they all face a big challenge, which is the need for other apps or even other Agents to connect and interact. Will some standards be formed in the future? Or has DingTalk's Agent started to consider connection standards?

Buqiong: We are now developing PaaS in the hope of providing scaffolding to a certain extent. If enough people use this scaffolding, we will encourage it to become a standard. For example, permissions now support automatic login and registration to external systems, but you need to enter the password and account string for configuration. From login and registration to interaction to exception information processing, some standards will gradually be formed. For example, AI assistants for purchasing air tickets and booking hotels will definitely come up with standards as long as they are used frequently.

Titanium Media: Have you considered linking DingTalk Agent with friendly companies Feishu and Qiwei in the future? Do you have such design considerations?

Not poor: I think there will be in the future, but definitely not today because we haven’t reached that stage yet.

What we most want to integrate is WeChat, but there are various challenges. This is still based on user value, not the platform perspective. As long as customers have demands, we will definitely do it, but it is not the main contradiction for customers now.

Customers have very low demand for data integration of Feishu product functions, and there will be relatively little integrated data for Qihoo, because their functions are all available on DingTalk, so there is no need for data interoperability on both sides. Instead, there are integration demands for third-party systems and non-homogeneous products. Everyone has integrated WeChat, and everyone wants to integrate Douyin and Taobao, so the demand is relatively large.