The real, capable Agent is here! Feishu has launched a large number of new AI products | The Latest Frontline

Author | Deng Yongyi

Editor | Su Jianxun

In 2025, is there still anyone who doesn't know that AI is the trend of the new era? Those overly fancy concepts are no longer of any concern.

Whether it's enterprises or many individual users, the focus of attention on AI has quickly shifted from large models to implementation. How to make AI help reduce costs and increase efficiency in real work, study, and life scenarios - that's the current hot topic in the AI circle.

On July 9th, Feishu released and upgraded several AI products, including Knowledge Q&A, AI Meetings, Feishu Aily, and Feishu Miaoda.

Xie Xin, CEO of Feishu. Source: Feishu

Compared with previous years, this can be regarded as another massive update from Feishu - not only were existing products such as Multi - dimensional Tables and Knowledge Q&A newly upgraded; the newly launched AI product suite like Aily and Miaoda are new products that address the current difficulties in implementing large models.

In the past few years, Feishu has gone through the early stage from product development to market coverage and has also accumulated a large number of customers in industries such as retail, high - tech, and advanced manufacturing.

According to the latest data disclosed by Feishu, in the new energy vehicle industry, 60% of the top 30 brands in terms of sales have started using Feishu; among the 6 listed brands in the tea beverage industry, 5 are Feishu users.

A competition for AI scenarios and new AI users has begun. In the past two years, how to capture the hearts of new players in the AI and embodied intelligence fields has become the focus of competition in the collaborative office market - during these two years, Feishu has also acquired new customers including DeepSeek and Zhiyuan Robotics.

△Source: Photographed by the author

An interesting little episode is that on July 8th, DingTalk suddenly launched a new AI table product, clearly showing its intention to compete. In response, Feishu also announced at today's press conference that it plans to launch Multi - dimensional Tables on the WeCom and DingTalk platforms - users on other platforms can also apply to use it on WeCom and DingTalk.

Massive Updates to Multi - dimensional Tables: From Tables to System - level Platforms

As a flagship product, Feishu Multi - dimensional Tables still took the "C - position" at this press conference.

According to the latest data disclosed by Feishu, the current monthly active users of Multi - dimensional Tables have exceeded 10 million - this is a significant number among domestic To B products.

From the underlying technology stack to the upper - layer applications, Feishu Multi - dimensional Tables have introduced a number of new features. In terms of performance, the capacity of Multi - dimensional Tables has increased, and the processing speed has also become faster.

Compared with 2024, the single - table capacity of Feishu Multi - dimensional Tables has now been increased to 10 million hot rows - ten times that of last year, but the processing speed has been significantly reduced. With a scale of 20,000 rows, the loading speed of Multi - dimensional Tables has decreased from 7.4 seconds to 0.94 seconds, and it only takes 1 second to load 50,000 rows, while similar domestic products may take up to 20 seconds.

What does this mean? It can already handle the business needs of a small e - commerce platform - it can directly manage 10 million SKUs/orders and cover information including order numbers, users, products, amounts, logistics, etc. on a segmented category platform.

△Source: Feishu

Why has the Agent - based transformation of intelligent table products become a battleground for all?

One reason is that Multi - dimensional Tables is a product built on aPaaS. In addition to connecting with documents, Multi - dimensional Tables has also become part of the core workflow of many enterprise customers.

In many small and medium - sized team businesses, Multi - dimensional Tables can replace the capabilities of many small business systems, and enterprises no longer need to purchase systems for sales, customer service, human resources, etc.

That's why table - based products have become a testing ground for large models + real - world business scenarios.

Specifically in terms of usage, in the past, users only needed to "drag and drop" in Feishu Multi - dimensional Tables, just like building blocks - this was also the core selling point of previous low - code and aPaaS products.

However, in reality, although this operation sounds simple, there are still thresholds. For users with no programming foundation at all, it's difficult to understand professional terms such as "fields" and "processes" and figure out the configuration process.

So, Feishu hopes to lower the threshold further. After the capabilities of large models have been further improved, users can now "build systems with their voices".

On the basis of dragging and dropping, users can have a conversation with "Feishu Miaoda" and give specific instructions, such as "I want to build an e - commerce business dashboard that includes categories like orders, logistics, and turnover."

Then, Feishu Miaoda will call on the Agent capabilities to directly generate an AI application with one click.

Moreover, to enable the rapid implementation of AI applications, the newly released "application mode" by Feishu is like taking one more step to "decorate" the AI application with one click.

△By dragging and dropping, you can build "alternative" sales systems, human resources systems, inventory management systems, etc. Source: Feishu

After the AI application is generated, the initial result will be more like a demo. But the "application mode" has a built - in rich set of application components, such as various list components, Tab components, and carousel images, which can make the AI application look more like an interactive "AI system".

In simple terms, users have more choices and can quickly build a business application that is both beautiful and practical, and it's also easier to deploy it on different platforms.

A Real, Actionable Agent is a Good AI

Another focus of the press conference was the Agent.

Agents have been popular from 2024 to 2025, but they have always been stuck in the problems of vague concepts and difficulty in implementation. Even in the industry, it's hard to say exactly how intelligent an Agent should be to be considered a real Agent.

This has led to a big gap between the "buyer's show" and the "seller's show" in the Agent field - the demonstrations look very cool, but there are many errors once implemented.

How can we determine whether an AI application is a real Agent that can be implemented?

△Source: Feishu

In this release, Feishu also introduced an "AI application maturity model" for enterprise customers to refer to. Specifically, it divides AI applications into four levels:

• M1 - Proof of concept. Only for internal demonstrations, not for external release.
• M2 - Early trial. For users willing to try new things.
• M3 - Mature application. Can be used on a large scale.
• M4 - Fully mature. Can be used in any suitable scenario.

Taking Feishu's existing features as an example, the previously released "Knowledge Q&A" has reached the M3 level - "Knowledge Q&A" can answer relevant questions in the enterprise by calling on various daily - accumulated information such as chat records, meeting minutes, and documents.

Another well - received AI transcription product, "Feishu Miaoji", has reached the M4 level.

"Just as the comparison of large models is not about functional differences but about generation quality and understanding ability, AI applications should also be selected based on their effects," said Xie Xin, CEO of Feishu.

In addition to clarifying the concept of Agents, Feishu also launched a new "AI product suite" - the Feishu Development Kit, which helps enterprises implement Agents in specific scenarios more quickly.

Aily △Source: Feishu

For example, the newly released "Feishu Aily" is like an enterprise - version Manus. Users only need a few simple steps to configure the persona, connect to enterprise knowledge, and add the business system MCP to create their own exclusive or business - specific Agent.

The core difference between Aily and third - party Agents lies in whether they have enterprise data. In real To B scenarios, enterprises are very sensitive to data issues. Aily supports private - domain data and has a series of data permission and isolation functions configured.

Since last year, Aily has been implemented in several enterprises. A typical customer is Bull Group - now, Bull Group has launched the expert - level customer service Agent "Bull Smart Service", which can respond to customers and thousands of dealers' questions in real - time 24/7, increasing the customer service reception capacity by 30 times.

After quickly implementing ideas through AI coding, it's like bringing in an unlimited number of product managers and junior developers for the enterprise, who can directly output prototypes and develop systems according to business needs.

Feishu aPaaS has also completed several updates on this basis, enabling AI to assist the entire process of business system development. "AI is bringing new software paradigms and development paradigms. In terms of software paradigms, intelligent systems plus Agents not only meet digitalization needs but also provide more intelligent human - powered services; in terms of development paradigms, AI Coding plus PaaS can also achieve the optimal solution among efficiency, stability, and reusability," said Zhou Zhou, the person - in - charge of the Feishu Development Kit.