Generalizing the Coding Agent to universal scenarios is a system-level competition.

Article author, source: Narrowcast AI

This week has been packed with major developments in the AI field: Jensen Huang redefining the AI PC, Microsoft announcing an “Agent-First” approach at Build 2026, OpenAI revealing the merger of ChatGPT and Codex, WeChat’s Agent progress being disclosed, Qwen beginning to integrate third-party skills, Doubao addressing rumors about paid features, and Meituan emphasizing the growing importance of service AI Agents during its earnings call.

Previously, some people asked why no one was talking about OpenClaw anymore. This week’s developments provide a clear answer: people have stopped mentioning OpenClaw because Coding Agent has emerged as a universal solution for task execution, merging with chatbots, while a corresponding ecosystem of skills and agents is being built, and new paid explorations are already underway.

Major companies are turning the insights from OpenClaw into tangible business progress. During this process, OpenClaw and the products we currently see may not represent the final form of AI products. As Yao Shunyu, Chief AI Scientist at Tencent, said at the 2026 Tencent Cloud AI Industry Application Conference, a long-term transformation has only just begun—true product forms, business opportunities, and usage patterns have yet to be fully invented.

What we can be certain of is that agents are becoming the core battleground for big tech’s AI efforts, and this competition is continuously evolving along four key fronts: who can expand user adoption across more productivity scenarios, who can deeply integrate within their internal products, who can build a sufficiently rich ecosystem of skills and agents, and who can accumulate sufficient context.

Colleagues have become the focal point of competition for Agents.

“Colleague” is the most frequently used term to describe agents today. Microsoft’s Scout is designed to work “like a colleague”; Koda 3.0 emphasizes collaboration between humans and AI teams; and OpenAI’s Agent plugins are described as “new colleagues who have already completed onboarding and understand all the processes.”

These statements mean that productivity scenarios have become a key battleground for big tech companies' agents.

Microsoft's Scout is an agent built on the OpenClaw framework, permanently integrated into Microsoft 365 and capable of running within Teams. It collaborates with productivity apps such as Outlook and OneDrive to browse emails, calendars, and work messages, automatically resolving meeting conflicts, drafting responses, and advancing tasks. Additionally, Microsoft has introduced Agent 365 to provide enterprises with centralized management of agent identities, permissions, policies, and risks.

OpenAI directly titled the event "Intelligence at Work." At this event, OpenAI introduced three core upgrades for Codex: customizable Agent plugins; expanding local annotation editing capabilities from code and web pages to documents, spreadsheets, and PowerPoint presentations; and the ability to generate websites for output reporting.

At the same time, Doubao mentioned in its response regarding paid features that, to meet the productivity needs of professionals, Doubao plans to launch Doubao Pro, which will include professional services such as software development, data analysis, professional design, process automation, financial analysis, and scientific research.

These product actions demonstrate that the substantial value of productivity scenarios—not just traditional enterprise scenarios—has been validated with real money.

Data released by OpenAI shows that since February this year, Codex's weekly active users have increased sixfold to 5 million, with knowledge workers growing at three times the rate of developers. Anthropic is expected to more than double its revenue in the second quarter to $10.9 billion, potentially achieving an operating profit of $559 million, with most of its revenue coming from enterprises and startups.

Integration and connectivity of internal products continue to deepen.

Product updates and iterations correspond to deeper structural reorganization of the product. On one hand, major tech companies have largely entered the Chatbot and one or more Agent product markets, and integration of these products has now begun. The most aggressive example is OpenAI’s integration of ChatGPT and Codex.

OpenAI aims to transform ChatGPT from a simple conversational interface into a central hub for collaborative agents, while Codex will evolve into a general-purpose agent platform capable of meeting diverse workplace needs—including office tasks, scientific research, enterprise workflows, data analysis, and business operations—by generalizing the use cases of coding agents. Through this integration, OpenAI hopes to introduce Codex to ChatGPT’s vast user base and expand its paid user base.

Additional reports indicate that OpenAI plans to involve its AI browser, Atlas, in this integration of super AI applications.

On the other side, major companies are rapidly integrating their existing internet product capabilities and services into AI products in the form of Skills or Agents. Alibaba’s early exploration involved adding functionalities like ordering food delivery, hailing rides, and shopping on Taobao to Qwen. Today, we can see ByteDance, Meituan, and Tencent undertaking similar efforts.

After connecting DouBao with the Douyin Mall, ByteDance is now adding recommendations for local life service stores and group deals in categories such as food and beverage, movie tickets, and homestays. Meituan stated on its earnings call that its AI assistant "Xiao Tuan" has been integrated into the Meituan app, serving over 100 million users during the May Day holiday across scenarios including dining, entertainment, travel, and online medical consultations. Tencent Docs has also transformed its accumulated document processing capabilities into Skills, which are now invoked by WorkBuddy.

Tang Dao-sheng, Senior Executive Vice President of Tencent, stated during the 2026 Tencent Cloud AI Industry Application Conference that many functions of traditional applications need to be transformed into capabilities that can be invoked by agents in order to further unlock the value accumulated over the years. This year, WeCom has opened up its original data capabilities through APIs and Skills, enabling other agents to access them. This trend toward openness is becoming increasingly evident.

Third-party ecosystem development is now being prioritized.

A key difference between this Agent and previous products is its ability to invoke tools, which requires a sufficiently rich ecosystem of tools behind it. Even large companies find it difficult to build such an ecosystem on their own, necessitating third-party development of Skills or Agent ecosystems.

The development of this ecosystem has now been prioritized.

After integrating Alibaba’s first-party products and services, Qwen has announced full openness to third-party Agents and Skills, enabling all businesses to operate their own branded Agents on Qwen. This week, Luckin Coffee, KFC, Mixue Ice Cream & Tea, and China Eastern Airlines have launched their Skills on Qwen. Businesses will soon be able to customize their Agent personas and specific services on Qwen.

Tencent is meanwhile integrating Meituan’s Xiao Mei into Yuanbao to provide users with services such as food delivery and ordering, while also accelerating the development of the WeChat Agent ecosystem.

Media reports indicate that WeChat's Agent has completed prototype testing and could begin its compliance approval process before public launch as early as this month. This Agent can orchestrate WeChat Mini Programs to deliver integrated services such as food ordering, ride-hailing, ticket booking, shopping, and local lifestyle services.

In addition, WeChat is also seeking to establish Agent-to-Agent connections with smartphone manufacturers such as Honor and Xiaomi, enabling their Agents to access WeChat’s core capabilities. This means smartphone manufacturers will become new entry points into WeChat’s Agent ecosystem, creating an architecture where multiple entry points share a single Agent ecosystem.

OpenAI's Agent plugins can bundle all the tools, knowledge, and skills required for a specific role in one package. For example, a creative production plugin can generate campaign boards, ad variants, product lifestyle images, and e-commerce image sets based on a brief, while also invoking tools like Figma, Canva, Shutterstock, Picsart, and Fal. In simple terms, this is a professional skill transmission system designed for Agents.

Currently, Codex's Agent plugins cover 62 popular applications and 110 skills. Moving forward, OpenAI aims to open the plugin ecosystem to partners, enabling third parties to create and deploy their own plugins directly within Codex and ChatGPT.

Context becomes even more important

Yao Shunyu stated that models are becoming increasingly adept at transforming complex inputs into outputs, but only if they are provided with sufficiently high-quality inputs. This requires users to supply models and agents with detailed, useful information that enables them to understand key questions such as “Who are you?”, “What are you doing?”, and “What answers would be valuable to you?”—thereby anchoring the correct path.

On the development side, adequate contextual communication is also essential for AI product development. During the above discussion, Yao Shunyu and Tang Daosheng noted that AI product development requires using product feedback to determine what the model should reward or penalize—what constitutes a good response and what qualifies as poor behavior. This means that the model team and the product team must collaborate through a shared-context process to co-design a better user experience.

Therefore, AI products must connect to and accumulate multi-source contextual information on the user side, then align with agent task intentions by determining which information to provide and which to withhold; on the development side, they must establish seamless feedback mechanisms to align the development goals of the model and product teams, accelerating experience optimization.

Both the accumulation of user-side context and the sharing of developer-side context are not merely development issues, but organizational ones that must be achieved through collaboration.

This is why, in January of this year, OpenAI reorganized its teams to bring the product team and researchers working on the underlying models closer together, and later merged the ChatGPT, Codex, and API teams into a single department led by Thibault Sottiaux.

At the same time, emphasizing context may also drive the agentization of hardware, turning hardware into an effective means for agents to collect user context. Microsoft’s Project Solara is exploring this very direction. While seamless communication anytime and anywhere is one goal of developing agent-enabled desktop terminals and wearable devices, the primary objective is to provide agents with richer contextual information in desktop and mobile scenarios.

Over the past few years, the AI industry has followed a relatively clear technological path: pre-training → post-training → Agent → Coding Agent. This path may not be the only future主线, but it is currently the most effective one that major companies can leverage.

The four trends we have identified serve as interconnected foundational coordinates along a fixed path, all aimed at enabling the Coding Agent to generalize across universal scenarios. This is another system-level competition.