Shen Chengang of Feishu Shennuo: Transcending 'Pseudo Multi-Agent', the New Generation Agent OS Reshapes the Capability Boundaries of AI Marketing

[Ebrun Original]On July 23rd, at Ebrun's "2026 Global New Brand AI Competitiveness Conference", Shen Chengang, Founder and CEO of Feishu Shennuo Meetsocial, delivered a speech titled "The New Generation Enterprise-Level Global Marketing Agent OS - The New Infrastructure for Global AI Marketing" and officially launched the new generation enterprise-level global marketing Agent OS – Marvy 2.0.

Shen Chengang believes that as model capabilities have significantly improved, the main challenge in applying AI to global marketing is no longer a "lack of tools," but how to translate model capabilities into tangible business performance.

During this process, companies expanding globally also face numerous pain points that urgently need resolution: general-purpose models lacking industry knowledge, different Agents being isolated from each other, decision-making processes being difficult to explain, and incomplete data chains. These issues make it hard for businesses to link efficiency improvements in individual touchpoints to sales conversions and to accumulate methodological assets.

To address this, Feishu Shennuo has developed a new generation Agent OS, which innovatively features a Task Scheduling Agent to provide top-level coordination for five sub-agents: Market Insights, Media Strategy, Intelligent Creativity, Intelligent Ad Placement, and Data Analysis, covering the entire global marketing chain.

This new system uses the Task Scheduling Agent to break down requirements and centrally coordinate, organizing multiple core functional Agents to work collaboratively. It relies on Feishu Shennuo's vast business database and continuous training and self-driven evolution through feedback from expert collective tuning.

According to Shen Chengang, the Marvy 2.0 AgentOS will empower frontline employees, liberate core managers, and enhance the overall productivity of the organizational ecosystem for companies going global. Furthermore, Feishu Shennuo plans to launch a more lightweight Marvy Lite in the fourth quarter of this year, aiming to empower more small and medium-sized businesses, enabling them to use AI Agents to participate in global competition.

This article has been compiled and edited based on the speaker's on-stage presentation, with modifications made without altering the original intent.

The following is the transcript of the speech:

01 Transcending 'Pseudo Multi-Agent', Global Businesses Need a 'Thinking Marketing Brain'

Hello everyone, I'm very pleased to be here today to introduce Feishu Shennuo. I believe sellers and e-commerce friends here involved in global business are no strangers to Feishu Shennuo.

However, our focus today is not on introducing the scale of our media business, but rather on sharing our technological breakthroughs in AI and global marketing, aiming to help sellers.

As AI applications in global marketing gradually take effect and model capabilities are significantly improving, how should the industry apply these capabilities to business and truly translate them into marketing results? People are still feeling anxious.

In fact, many different types of AI products have already emerged in the market. However, the information is very fragmented, and brands are often very confused about which one to choose.

At the same time, we also see that some products can indeed improve local efficiency, such as using a Creative Agent to produce materials that look good. But how effective is this creative work? How to link it to final conversions? This remains a headache.

During the AI application process, we have observed several pain points.

First, is 'unprofessionalism' – general-purpose models don't understand industry knowledge.

We can consult general-purpose large models on industry marketing questions, but their answers often lack sufficient professional backing.

Second, is 'unreality'.

The information they use mostly comes from public channels, lacking the support of authoritative third-party data, private enterprise professional data, and real business scenarios. Therefore, when companies see the data and conclusions provided by the model, they often take them with a grain of salt.

Third, is 'lack of synergy'.

Different AI Agents lack coordination and are not interconnected. Each tool only improves the efficiency of one point, unable to achieve a closed-loop across the entire marketing chain.

Fourth, is 'untrustworthiness'.

The entire deduction process of the AI application is still a black box: the model will tell you how to operate, but why to do it this way, what the advantages are, what potential problems may exist, and how to review and verify afterwards – companies are completely in the dark.

Finally, is 'lack of evolution'.

Due to the absence of a complete customer data chain, the model cannot learn from previous applications, making it difficult to continuously improve in the next application.

From IDC's reports, we can see that, real AI Agent applications need to gradually transition from human-led to fully autonomous. In this evolutionary process, how companies move from basic automation, semi-automation to full automation is not a simple path.

Feishu Shennuo is also rapidly advancing within its full-chain marketing service system, hoping to provide more support in areas like insights and creativity, continuously improving the maturity of AI applications.

Today, I want to introduce Marvy 2.0 to you. We hope it can become a 'thinking marketing brain.'

The name Marvy represents its core capability architecture:

M stands for Multi-agent, meaning multi-agent collaboration;

A stands for Autonomous, meaning self-learning and evolution;

R stands for Reliable, meaning providing trustworthy actions, reasoning, and decision-making basis;

V stands for Verifiable, meaning all operational data is traceable and can be cross-validated through actual results;

Y stands for Yield maximization, meaning maximizing marketing value.

We hope that Marvy 2.0 transcends the simple parallel connection of 'one portal + N large models + search engine,' breaks through the capability barriers of 'pseudo multi-agent,' and becomes a truly operational decision-making and execution system, bringing 'marketing democratization' to global businesses, and moving advertising growth further from possibility to certainty.

02 Supported by Massive Data and Empowered by Expert Cases, Five Core Agents Unlock the Full Global Marketing Chain

The Marvy 2.0 system contains five core Agents: Market Insights, Media Strategy, Intelligent Creativity, Intelligent Ad Placement, and Data Analysis. They correspond to the complete marketing chain from understanding the market and formulating strategies to generating creativity, executing placements, and analyzing feedback.

First is the Market Insights Agent.

A major challenge in global marketing is localization, and the biggest difficulty in localization is whether a company can truly understand local consumers, market competition, and competitors' placement strategies. Ultimately, companies need to enhance their understanding of the local market.

In the past, helping a client complete a localized market analysis typically required professionals to spend two to three days to produce a report aiding decision-making. Now, we have accumulated over a decade of industry data and experience in a data warehouse, including thousands of exclusive historical industry cases and dozens of continuously updated external third-party data sources.

Based on AI collaborative cognition, companies only need to input unified target requirements, and the system can increase market analysis efficiency by 10 times, producing a professional market analysis report within 5 minutes, providing decision-making references for business owners.

After understanding the market, companies also need to determine which media to choose, how to allocate budgets, and what market effects might be achieved. This corresponds to the Media Strategy Agent.

In the past, completing a full set of media strategy planning for a client took at least a day, requiring professional marketers to conduct extensive research.

Now, we have established a database containing over 60 media outlets and accumulated execution references from tens of millions of campaigns. Simultaneously, Feishu Shennuo's expert team deconstructs the execution logic of campaigns and media strategies, digitizing the entire operational process, then entrusting it to AI for prediction and optimization. In this way, the operational efficiency of media strategy can be improved by over 5 times.

The Intelligent Creativity Agent is also a very important part of AI marketing applications.

Now, almost any large model can generate creativity according to requirements, but whether the creativity it generates can lead to effective marketing results is another question. A company might test 1000 creatives, invest a lot of money, but the final results are not ideal.

Therefore, the most important element of creativity is not production cost, but whether 'effective' creative production can be achieved. By effective, we mean the output creativity can help the company achieve the expected marketing results.

In the past, our creative personnel would take one to two days to complete creative strategy planning for a client. Because conceiving a good creative not only requires understanding consumers but also integrating product selling points into the content and verifying through more diverse creative variations and different testing angles.

During this process, we have accumulated vast amounts of data. Currently, Feishu Shennuo adds over 800,000 new pieces of material daily. Data behind each piece of material, such as clicks, impressions, conversions, and purchase behaviors, are tagged and undergo structured deconstruction. Focusing on information like selling points within the material, we have extracted over 27,000 tags.

At the same time, creative strategists continuously optimize strategies. Marvy 2.0 will be trained based on this historical data and daily new materials, continuously improving material output and operational efficiency. The efficiency improvement in this stage is calculated in multiples of 10.

With insights, strategy, and creativity, companies also need to quickly complete batch placements across different media. Especially for clients with large placement volumes, advertising placement involves a lot of account setup, parameter setting, material uploading, and subsequent adjustments, making the entire process very lengthy.

In the Intelligent Ad Placement Agent, we have accumulated advertising creation and adjustment skills honed over the past decade. Feishu Shennuo's operations team manages global advertising budgets of up to $8 billion annually, with over 300 optimization experts. Marvy 2.0 digitizes and tags the experience of optimizers during the placement process, then combines it with the SOP of the entire optimization workflow. From internal application results, the operational efficiency of advertising placement has increased by 20 times.

Finally, there is the Data Analysis Agent. After completing the previous work, the marketing chain generates a large amount of data. In the past, optimizers needed to invest a lot of time daily, combining data from different media and different creatives, then interpreting and attributing it.

During the development of Marvy 2.0, we continuously accumulated historical data and the know-how of optimizers in data analysis. The daily influx of new business data also continuously provides high-quality training for Marvy 2.0. In this way, insights, strategy, creativity, placement, and analysis are no longer isolated links but form a complete chain capable of continuous feedback and optimization.

03 Liberating Organizational 'Productivity' for Global Expansion, Empowering SMBs to Tap the Global Blue Ocean

In terms of maturity, Marvy has evolved from version 1.0 last year, which was roughly at L2 (Partial Automation) stage, to the current version 2.0, which is between L2 and L3 (Conditional Automation), and is still continuously evolving. We expect that by next year, Marvy will hopefully reach the L3 to L4 (Highly Automated) stage, capable of efficiently and accurately automatically handling complex marketing events.

Here is a real case study.

Our client deployed a dedicated private Marvy 2.0 model, covering modules like Market Insights, Media Strategy, Intelligent Creativity, Intelligent Ad Placement, and Data Analysis. The system overall helped the client improve efficiency by 74%, reducing time spent by 89%.

This is a real commercial case; the client paid for the Marvy 2.0 product, and it represents a successful commercial landing of Marvy 2.0 as an AI intelligent marketing OS in the market.

To summarize, we hope to use Marvy 2.0 to change the past state of scattered and chaotic AI tools, reconstructing it into a concise and clear Agent OS like an operating system, forming a full-chain, verifiable intelligent marketing platform.

In this system, we are applying cutting-edge technologies such as large-scale intelligent agent reinforcement learning, self-evolving meta-agents, and self-organizing multi-agents. At the same time, regarding underlying technology interfaces, Feishu Shennuo has many partners, including several leading overseas large models and leading domestic multimodal models.

The strategy of Marvy 2.0 is not to replace humans with AI. We hope that in marketing work scenarios, AI can take on low-value, repetitive tasks, thereby unleashing employees' higher creativity in decision-making, analysis, and judgment.

We hope Marvy 2.0 can help companies accumulate digital assets, empower frontline employees, liberate core managers, and thereby drive the entire industry ecosystem to operate more efficiently.

Currently, Marvy 2.0 primarily serves large clients, but we hope to make these capabilities available to more small and medium-sized businesses in the future, using AI to promote the popularization of global digital marketing capabilities.

Therefore, we plan to launch the Marvy Lite full-chain closed-loop growth engine in Q4.

Compared to the current version, Marvy Lite will be more lightweight, more automated, have higher operational efficiency, and consume fewer tokens. We hope to truly achieve 'every merchant can afford an AI Agent,' enabling more companies to participate in global competition.

Additionally, at this summit, Feishu Shennuo Meetsocial officially announced the significant news of Marvy 2.0 integrating with Seedance. On-site, a representative from BytePlus and Shen Chengang, Founder and CEO of Feishu Shennuo Meetsocial, jointly held an 'Official Partner' accreditation ceremony.