Is there a bubble in the intelligent agent industry? What drives its real development?

/

Interview by | Zhang Hongyi

Written by | Yang Xiaoruo Edited by | Zhang Hongyi

Since the beginning of this year, with significant advancements in large model capabilities and a notable decrease in development barriers, numerous major companies and investors have been closely monitoring and increasing their investments in the intelligent agent sector. This has, however, sparked a "gold rush" of intelligent agent products, with some niches even seeing redundant offerings.

Particularly in consumer-facing segments like personal productivity and general-purpose assistants, some products lack genuine paid usage scenarios and rely solely on traffic and industry buzz to maintain visibility. Without sustained user engagement, achieving a complete commercial closed loop becomes challenging. For example, some basic AI applications, relying on simple workflows and fundamental tool integrations, are being marketed as advanced intelligent agents.

The year 2026 has been hailed as the inaugural year of "large-scale AI industry intelligent agent deployment." Amidst this rapid growth, debates have arisen over whether a bubble exists in the intelligent agent market. In the capital market, there is a trend where some projects prioritize securing funding over identifying clear application scenarios. Some investors predict that over 40% of Agent projects may exit the market this year due to ambiguous value propositions, uncontrolled costs, and implementation hurdles.

Is the booming intelligent agent sector genuinely driving industrial dividends, or is it just fueled by hype and conceptual speculation? Among the myriad of emerging intelligent agent projects, which segments represent genuine demand, and which are mere bubbles? With crowded and fiercely competitive niches, has the industry already entered a bubble phase?

Given the current industry landscape, will competition in the consumer market continue to intensify? How can the real implementation dividends in the business market be unlocked? How should companies navigate pseudo-demands and seize opportunities in certain segments? When will the true turning point for the intelligent agent bubble arrive? If we delve deeper, what is the real growth logic behind intelligent agents?

To address these questions, 「Business Show」 recently conducted an exclusive interview with Wu Yongjian, Vice President of Tencent Cloud and Head of Cloud Intelligence Product Research and Development, to dissect the bubble truth behind intelligent agent development, explore the boundaries between genuine and pseudo-demands, understand the differing development logics between consumer and business markets, and uncover future growth opportunities.

Wu Yongjian, Vice President of Tencent Cloud, stated in an exclusive interview with 「Business Show」 that the entire intelligent agent industry is still in its infancy and far from a bubble phase. With the widespread adoption of capabilities like code-writing and office productivity agents, barriers to development, PPT creation, and office tasks have significantly decreased, leading to a surge in demand from both individuals and businesses. Moreover, many current demands have yet to be fully met or refined by existing products.

He also believes that technological innovation in the industry is ongoing. From enhancements in reasoning capabilities and the implementation of the Lobster Claw model to the continuous deployment of new technologies like Harness intelligent orchestration and bidirectional workflow scheduling, these advancements are driving continuous industrial evolution. Coupled with the continuously lowering barriers to AI development, business implementation demands across various industries have exploded. Furthermore, the enterprise market features a clear and stable payment logic. As long as products can streamline business processes and ensure accuracy and stability, they can achieve high renewal and repurchase rates, representing tangible industrial growth.

During recent interviews by 「Business Show」 with several Tencent intelligent agent product leaders, there was a consensus that intelligent agent development is still in its early stages and far from reaching its final form, with diverse business models and participants emerging. Intelligent agents will differentiate by scenario, such as super agents for the general public (Yuanbao), office productivity tools (WorkBuddy), and enterprise-level platforms (ADP), each with unique capabilities. The core challenge for implementation lies in distinguishing tasks best suited for AI from those requiring human intervention, necessitating a distinction between the Claw (Lobster) model and workflow models. Standardized operating procedures (SOPs) and task-oriented processes should utilize workflows to ensure stable delivery. The value of enterprise-level intelligent agents evolves from information processing to scenario-based problem-solving and ultimately to result delivery, with the core being stable and secure operations in business environments to form a continuous value operation system.

On July 18, at the 2026 World Artificial Intelligence Conference, Tencent Cloud officially launched the overseas version of its intelligent agent development platform, ADP 4.0. The platform's core focus is on reducing construction barriers, enabling bidirectional scheduling between the Claw model and Workflow, balancing free orchestration and deterministic processes, and meeting enterprise requirements for real-time performance, cost, and permissions. Additionally, Tencent Cloud announced the "Top 10 Industries, 100 Key Scenarios Ecosystem Plan," inviting ecosystem partners to focus on high-frequency scenarios like customer service, business analysis, and transaction automation.

The overseas expansion of intelligent agents is also a key strategic direction for Tencent Cloud, though the domestic market remains the primary focus at this stage. Wu Yongjian told 「Business Show」 that significant differences in compliance, culture, and user awareness exist between domestic and international intelligent agent markets, making it impossible to simply replicate the domestic model overseas. "Leveraging the product strength honed through intense domestic competition is our biggest differentiator for overseas expansion. At this stage, Tencent's intelligent agent overseas expansion adopts a PLG (Product-Led Growth) strategy, prioritizing deep cultivation in Southeast Asia, focusing on exploration, pilots, and localized refinement to steadily accumulate overseas implementation experience." It is reported that the ADP 4.0 overseas version has undergone four major international adaptations in channels, interaction, ecosystem, and connectivity.

The following is an edited transcript of the interview between 「Business Show」 and Wu Yongjian, Vice President of Tencent Cloud:

「Business Show」: Intelligent agents are extremely popular right now, and many people are discussing whether a bubble exists. How do you view the development in both the business and consumer markets?

Wu Yongjian: I don't think we're anywhere near a bubble phase. Large models have gained popularity since 2023, so it's only been about three years. But the changes in these three years have been incredibly rapid, with new technological concepts and product forms emerging annually. Initially, the focus was mainly on model capabilities, and product forms were uncertain, mostly involving simple conversations paired with assistant modes like Copilot or Pilot.

As technology evolved, especially after Deepseek became popular early last year, models were no longer just answering questions but truly capable of deep thinking and reasoning, directly transforming the industry's product landscape. Intelligent agents evolved from simple, localized Q&A to deeper thinking modes resembling humans, spawning a wave of personal assistant products and completely disrupting traditional search models. Tencent Yuanbao emerged and grew in popularity amid this wave.

This year, the industry has seen new changes. The introduction of OpenClaw, or the Lobster model, early in the year opened up entirely new approaches. Tencent quickly seized this opportunity in February and March, launching products like QClaw and WorkBuddy, with WorkBuddy gaining particular momentum and positive feedback.

So, I insist there's no bubble yet. Technological iterations over the past few years have continuously spurred new demands. Last year, Deepseek accustomed users to deep questioning and thinking, disrupting traditional search models. This year, WorkBuddy elevated intelligent agents from "answering questions" to "getting things done," practically helping users complete entire work tasks.

The entire industry is still in its early stages and far from a bubble. With the widespread adoption of capabilities like code-writing and office productivity agents, barriers to development, PPT creation, and office tasks have significantly decreased, leading to a surge in demand from both individuals and businesses. Moreover, many current demands have yet to be fully met or refined by existing products.

Of course, some segments are indeed becoming crowded, with personal office productivity being the most typical example. It's not just Tencent; many companies across the industry are involved. However, this is just crowding in specific segments, not an industry-wide bubble. Products that can truly stand out from the competition are those that closely align with user needs and solve real problems.

「Business Show」: Overall demand is indeed growing, and the industry is on an upward trajectory. However, if we break it down, what are the genuine demands and pseudo-demands in different consumer and business scenarios? Which scenarios can truly be implemented and create value?

Wu Yongjian: From our practical experience, there's a clear trend of demand shifting from the consumer to the business market and penetrating deeper. Currently, the most crowded yet opportunity-rich consumer segment is personal productivity and intelligent office tools.

After the release of the Anthropic model early this year, many traditional SaaS companies saw their businesses impacted, and this trend has continued. It's evident that traditional enterprise office SaaS models are gradually being replaced by AI intelligent agents. WorkBuddy is slowly becoming the new gateway for enterprise office work, integrating various traditional office capabilities and delivering substantial scenario value. It aids in productivity, collaboration, automatic PPT creation, and report writing—all genuine rigid demands.

Although the office productivity segment is highly competitive, there's significant room for expansion. WorkBuddy is also building its ecosystem, integrating with various office systems and hardware devices. Many hardware manufacturers are developing their vertical intelligent agents based on WorkBuddy, covering more scenarios and meeting diverse demands.

Our goal with the Tencent ADP platform is to implement vertical scenarios across various industries. With AI construction barriers continuously decreasing, we hope to collaborate with partners to explore implementation opportunities across different sectors. Intelligent customer service is already a mature and validated scenario, replacing significant manual work with higher efficiency and accuracy. Additionally, enterprise knowledge bases, internal and external Q&A systems, and intelligent marketing are scenarios with consistent demand and excellent implementation results.

No single platform can cover all the nuanced experiences and details of every industry, so we insist on ecosystem collaboration. Tencent focuses on building the platform and foundational capabilities, while partners delve deep into their respective industry scenarios to ensure thorough implementation. With increasing demand and lowering implementation barriers, only by working together can we truly solve various practical problems for enterprises.

「Business Show」: As consumer and business markets continue to penetrate each other and boundaries blur, many enterprise clients are still unfamiliar with intelligent agents and in the exploratory stage. How do we help enterprises build awareness and understand the mutual empowerment value of consumer and business markets?

Wu Yongjian: Many enterprise clients' awareness of intelligent agents primarily comes from industry reports and personal experience using products. Tencent's greatest advantage lies in its massive consumer user base and mature consumer scenarios. Many business owners and employees personally use products like WorkBuddy, QClaw, and Marvis.

A few years ago, people's understanding of intelligent agents was superficial, viewing them as mere Q&A assistants like Copilot. However, intelligent agents can now independently create PPTs, write marketing copy, build workflows, and perform deep work, delivering tangible value. Once enterprises experience this value firsthand, they naturally want to implement these capabilities internally.

However, the usage logic and error tolerance standards for consumer and business markets differ significantly. Individual users are highly tolerant of consumer products, accepting occasional errors or slower responses. If a product doesn't meet their needs, they can simply stop using it. Enterprise services, however, brook no mistakes. Internally, corporate financial, salary, and business data are highly sensitive, requiring intelligent agents to deliver precise content. Externally, in customer service scenarios, any incorrect output can directly impact the enterprise's brand and reputation. Therefore, enterprises have much higher requirements for product stability, accuracy, and security than consumer users.

Having enterprise users experience consumer products first and then implementing them in the business market is the best way to cultivate enterprise awareness. It allows them to appreciate the value of intelligent agents while understanding the rigor and high standards of enterprise applications. Many large enterprises now aim to build their AI platforms and entire intelligent agent systems, a clear demand among industry leaders.

However, the market is predominantly occupied by mid-sized and small-to-medium enterprises. These enterprises don't require comprehensive AI platforms; instead, they prioritize addressing specific business challenges and achieving measurable outcomes. Consequently, our strategy has pivoted from a general-purpose platform to scenario-specific implementations, offering more customized scenario solutions for small and medium-sized clients through the ADP.

「Business Show」: The ADP platform has also undergone refinement based on actual needs identified during Tencent's client service engagements. With products like the Buddy series and QClaw already in the market, many are curious about the positioning of ADP. Furthermore, some enterprises are concerned that the implementation results may not meet their expectations. What are your thoughts on this?

Wu Yongjian: Indeed, from last year to this year, some enterprise clients have voiced concerns about implementation results falling short of expectations. This is primarily because the industry is still in its nascent stages. Typically, the pattern is to focus on the consumer (C-end) experience first, followed by a gradual rollout to the business (B-end) sector. Only when individual users recognize the value do they become willing to adopt and implement the technology at the enterprise level.

The rapid technological advancements in large models and intelligent agents have significantly expedited B-end implementation. In contrast, during the deep learning era over a decade ago, although there were breakthroughs in computer vision (CV) and voice technologies, their pace of iteration was much slower. Now, with faster technological progress, B-end commercialization has also picked up speed.

Precisely because the industry is in its early stages, many enterprises find it challenging to swiftly leverage platforms after purchase. While building simple Retrieval-Augmented Generation (RAG) applications is relatively straightforward, creating comprehensive, business-ready workflows demands professional implementation—a common hurdle for clients last year.

To tackle this, we enhanced our Lobster capabilities this year, optimizing cloud-based Harness scheduling and integrating traditional workflows with Lobster's intelligent orchestration. Complex, flexible scenarios leverage Lobster Harness for intelligent orchestration, while rigorous, fixed-process businesses rely on established workflows, significantly reducing barriers to enterprise implementation.

We also placed a strong emphasis on Frontline Delivery Execution (FDE) services. In the AI era, merely selling tools and platforms is insufficient; we must deeply engage with clients' business scenarios to ensure successful implementation. Our FDE teams collaborate on-site with enterprises, providing product training, scenario exploration, workflow construction, and utilizing proven templates and workflows from other clients to facilitate rapid implementation and avoid pitfalls.

Our public cloud ADP data clearly demonstrates that as long as we assist in achieving tangible business results in the first month, renewal rates surge, leading to sustained collaboration as businesses expand. This is the crux of the difference between B-end and C-end: once enterprise products take root and deliver value, they cultivate long-term, stable partnerships.

The newly launched ADP 4.0 overseas version also incorporates numerous optimizations to address client implementation challenges. Given the rapid technological updates in the industry, from RAG and workflows to MCP and Skill last year, and Harness this year, we must proactively embrace new technologies to continually lower client implementation barriers.

C-end products prioritize user experience, such as enhancing PPT presentations. However, the enterprise-level ADP platform focuses on assisting enterprises in building, managing, and deploying their intelligent agents, all centered around genuine business needs. While reducing implementation barriers, it prioritizes addressing enterprises' core concerns, including controllability, response speed, cost, and permission security.

Although Lobster's intelligent orchestration offers flexibility, it can sometimes be slow to respond and lack controllability, making it unsuitable for scenarios like customer service and warehousing inquiries that demand rapid and accurate output. Therefore, ADP 4.0 emphasizes upgrading the Claw construction mode to enable bidirectional scheduling between Claw's intelligent orchestration and traditional workflows: vague and complex innovative tasks are assigned to Claw for intelligent processing; for tasks with fixed processes, clear standards, and high speed and accuracy requirements, mature workflows are automatically scheduled for execution. Simultaneously, workflows can also coordinate multiple intelligent agent teams to complete collaborative tasks such as R&D, testing, operations and maintenance, and release, striking a balance between flexibility, precision, and stability.

Our product iteration primarily consists of two steps: first, we concentrate on serving large clients, creating benchmark cases, and thoroughly validating the platform's value; second, we delve into vertical scenarios, developing standardized, ready-to-use AI solutions for mature scenarios like intelligent customer service and industrial quality inspection, reducing implementation costs for small and medium-sized clients.

Additionally, we have fully opened up ADP's OpenAPI capabilities, making core functions like Lobster orchestration and workflow scheduling accessible to external users. Enterprises from various industries possess a deeper understanding of their own sectors than we do, enabling them to conduct secondary development based on our capabilities and jointly foster an ecosystem. Our initiative, the 'Top 100 Scenarios in Ten Industries' plan, aims to collaborate with partners to implement more enterprise scenarios with genuine value.

「Business Show」: Nowadays, overseas enterprises are also paying close attention to intelligent agents, and going overseas entails addressing differentiated issues such as data compliance, security, and regional cultural nuances. What do overseas clients care about the most? Where do our differentiated advantages lie in expanding overseas?

Wu Yongjian: The overseas and domestic markets differ significantly, whether it's data compliance, model adaptation, or local cultural and localization needs. Our current overseas expansion is still in its early exploratory phase. The launch of ADP 4.0 Overseas Edition is primarily to gather feedback from overseas users and refine product capabilities tailored to overseas markets.

Similar intelligent agent platforms overseas are also in their infancy. The domestic market is highly competitive and undergoes rapid iteration. The product capabilities we have developed through intense competition in China are highly competitive overseas, granting us a significant product advantage.

Promotion strategies also vary between domestic and overseas markets. In China, we can adopt both Sales-Led Growth (SLG) and Product-Led Growth (PLG) approaches, with direct sales teams being highly effective in expanding large clients in the early stages. However, overseas local teams have limited manpower, making it challenging to scale the SLG model. Therefore, our core strategy overseas is PLG, relying on the product's strength to penetrate the market and build a reputation.

There are already some competitors overseas offering workflow and pipeline construction tools, and domestic peers are also expanding overseas. However, our opportunity is clear: to fill gaps in the overseas market with robust product capabilities and differentiated strengths. Currently, we are primarily focusing on the Southeast Asian market, exploring and advancing steadily.

Overseas enterprises' understanding of intelligent agents lags behind that in China by a few years, so our current focus is on exploration and pilot projects, gradually accumulating experience in overseas localization and implementation.

「Business Show」: The intelligent agent sector is thriving this year, with everyone developing products, building platforms, and establishing ecosystems. However, the industry must ultimately deliver real value in practical scenarios. Only with tangible results will enterprises be willing to invest. Now, when enterprises procure AI products, they are becoming increasingly rational, meticulously calculating TOKEN costs and return on investment. The industry will eventually move beyond the hype of conceptual speculation and begin to refine products meticulously, focusing on delivering practical and quantifiable real value to enterprises. 'The End.'