---
title: "AI's \"Chat Dividend\" Is Over, but the \"Work Dividend\" Is Just Beginning"
author: "Unique Research"
sourcePublication: "Unique Research Substack"
originalPublishedAt: "2026-07-18T13:00:44+00:00"
canonical: "https://ffcap.cn/en/research/src-20260718-01html"
source: "https://uniqueresearch.substack.com/p/src-20260718-01html"
language: "en"
---

# AI's "Chat Dividend" Is Over, but the "Work Dividend" Is Just Beginning

[![](https://substackcdn.com/image/fetch/$s_!kRsB!,w_1456,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F1cf012dd-00b6-46e6-9d49-768701bbf83d_2048x1152.png)](https://substackcdn.com/image/fetch/$s_!kRsB!,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F1cf012dd-00b6-46e6-9d49-768701bbf83d_2048x1152.png)

_Original · Unique Research · 2026-07-18_

_Editor's note: The first-person report and its judgments belong to the original Chinese author. This English rendition retains the traffic data analysis, product case studies, industry report findings, and the full rankings methodology appendix. Traffic figures, ARR numbers, and growth percentages are sourced from Unique Research's June 2026 AI Web ranking report and are self-reported by products or estimated by third-party measurement, not independently verified. Product names and company affiliations are preserved as source attributions._

AI Industry Observation

From Chat to Work: AI's Real Inflection Point Isn't in the Traffic Rankings

"Chat won't disappear, but it is shifting from being the product itself to becoming the communication interface of Work."

This week, OpenAI merged Codex and ChatGPT into a single desktop app and launched ChatGPT Work — its headline is "no chatting, just getting work done." GPT-5.6 and Grok 4.5 were released almost simultaneously. A few days earlier, Anthropic had moved Cowork from desktop to mobile and web. Meta wasn't idle either, rolling out Muse Spark 1.1.

The world's smartest companies were all telling the same story this week: AI's next step is not answering questions, but finishing the work for you.

Meanwhile, at WAIC in Shanghai, StepFun's Yin Qi (印奇) said something that resonated with the Silicon Valley narrative: "In 2026, model capabilities are crossing a critical threshold. Agents are advancing from executing for seconds continuously to working independently for tens of hours. In other words: one person + an Agent = a team." Tencent released its "2026 Top Ten Trends," stating that enterprise-grade Agents are evolving from point assistance to collaborative digital teams.

But if you look at June's global AI web traffic data, you'll see an uncomfortable fact: 64 of the top 100 apps are declining — nearly two-thirds.

Yet at WAIC in Shanghai, 57 application scenarios landed, with intended cooperation amounts totaling approximately 16.2 billion RMB. AI's money and scenarios are shifting from "technology" to "hands-on work."

For the past three years, we've measured AI products by visits, monthly active users, and conversation counts. But entering 2026, these metrics are distorting. AI's basic interaction unit is shifting from "a round of dialogue" to "a piece of work that can be delegated, executed, and verified." Users no longer just ask for an answer — they hand a goal to AI, let it plan, call tools, execute across systems, and deliver the result back. Chat won't disappear, but it is shifting from being the product itself to becoming the communication interface of Work.

Most People Haven't Understood This Decline

Open the data sheet and look at the Google ecosystem first: Gemini -1.5%, NotebookLM -13.7%, AI Studio -4.9%, even Stitch fell 16.5%. After months as growth stars, they've collectively turned downward for the first time.

What about OpenAI? ChatGPT MAU is 516 million, up a mere 1.1% — essentially flat. Claude, however, grew 5% against the trend, holding at 106 million MAU, becoming the only large model in the top 5 still in positive growth.

Category-level divergence is worth dissecting. Among emotional companion products, 7 of 10 are growing, averaging nearly 9% month-over-month — the only large category with a sufficiently large sample and positive average growth. But this is more of a client-to-web migration; our data only monitors web, so the growth you see may just be usage moving from app to browser, not an expanding user base.

More noteworthy is the broad decline across categories: coding assistants, 2 of 8 growing, averaging just 0.3%; image editing, all 7 declining; development tools, all 4 declining, averaging -8.9%; general chatbots, only 2 of 11 growing, averaging -3.6%.

"The novelty dividend of tool-based AI is running out, but that doesn't mean tools have no value — it means tools in 'chat form' are receding while tools in 'delivery form' are taking over. And this handoff is happening largely in places our traffic rankings can't see."

From Coding to Work Is Gradually Becoming the Main Line for More People

If you look carefully at the products on the growth chart that "have scale and are genuinely growing," a clear thread emerges — many of them were previously coding tools and are now transforming into Work Agents.

Cursor is the most striking case on this thread. Traffic 22.31 million, up 7.2%; MAU 8.66 million, up 14.6%. ARR exceeds $2 billion, with over 50% of the Fortune 500 using it. But its key change isn't in these numbers but in product form: in March 2025, tab-completion users were 2.5x agent users; by 2026, this ratio has completely reversed — agent users are 2x tab-completion users. Agent Mode has become the default workflow.

Cursor's success can be summed up in one sentence: it doesn't sell the concept of "AI programming"; it sells the two hours engineers save every day.

Its evolution trajectory — from code completion to multi-file editing, to Agent Mode autonomous execution, to Background Agents running in the background — is precisely a microcosm of the entire AI industry's shift from Chat to Work.

The same trajectory is happening with Chinese tools, and with even greater momentum.

Qoder (Alibaba), June traffic 3.06 million, up 15.33%; MAU 870,000, up 13.40%. Growth is steady but not explosive. Its story lies in the progression of its product line: Qoder IDE launched in August 2025, QoderWork in January 2026, and QoderWake "digital employee" in April — three layers from pair-programming to autonomously shipping features. Pricing is Pro $30/month, Ultra $200/month. The core logic is the same as Cursor: from "helping you write code" to "finishing it for you."

Trae (ByteDance), this line is even clearer. Trae IDE launched in early 2025; in March 2026, SOLO became a standalone desktop/web app; in June, Trae Work officially launched — expanding from IDE to workspace, including Work mode, Code mode, and Design mode. At $10/month Pro pricing, it's roughly half of Cursor's. Its official positioning is direct: "workspace, not chatbot."

codebuddy, Tencent's AI coding assistant, had June traffic of 7.31 million, up 30%; MAU 2.26 million, up 40.75%. MAU growth even outpaced traffic growth. It also garnered significant attention after launching WorkBuddy — internal beta in February 2026, official launch in March, developed by the Tencent Cloud CodeBuddy team, essentially CodeBuddy's extension to the non-developer market. But its origin story is interesting: Tencent internally discovered that over 10,000 non-developers were "gritting their teeth" using CodeBuddy — they didn't understand directory trees and terminal windows, but endured the discomfort to organize data, write reports, and run workflows. Huang Guangmin called these "jailbreak users": they aren't CodeBuddy's target users, but their existence proves the demand is there, and the barrier is blocking a larger market.

WorkBuddy's design grew directly from this insight: you land on an input box, scenario-based entry points tell you what you can do, there's a preview after you finish, and you can share after previewing. It supports switching between Ask, Plan, and Craft modes — simple Q&A switches to Ask, letting it do work switches to Craft. It shares the same technical foundation and account system as CodeBuddy, but targets not developers but every ordinary person who needs to "use AI to get something done."

Four Chinese tools, four approaches, the same direction: from Coding IDE to Work Workspace.

Notably, Trae Work, Kimi Work, and WorkBuddy are primarily desktop clients, outside our web traffic monitoring. What you see on the charts is codebuddy-CN (+30%) and Qoder (+15%) — the tip of the iceberg visible to public data amid this transformation.

"Behind this is a judgment: if AI's ultimate form isn't 'a better chat box' but 'a digital employee who can finish your work,' then coding is the test bed where this path is validated first — because code is verifiable, the scenario best suited for AI to evolve from 'suggestion' to 'execution.' Coding tools have completed the shift from Chat to Work first; now this path is spreading to other scenarios."

Video Generation: Why Is Higgsfield Up 18% Against the Trend?

Video-related categories are generally soft. On the traffic chart, video generation averages +2.2% (2 up, 2 down); the MAU chart is worse — video generation averages -3.26%, video editing averages -7.27%. HeyGen traffic -8.86%, VEED.IO MAU -13.02%, Media.io MAU -26.65%.

But Higgsfield is an exception: traffic 26.81 million, up 10.6%; MAU 7.17 million, up 18.8% — MAU growth even outpacing traffic growth, indicating improving user retention.

What did Higgsfield do right? It isn't a single video generation model but a multi-model integrated production platform — integrating 15+ video models including Sora 2, Veo 3.1, Kling 3.0, Seedance 2.0, and WAN 2.6 into one workflow, split into two product lines: Cinema Studio 3.5 for narrative/cinematic content, Marketing Studio for commercial content. 20 million users, $200 million in revenue over 9 months.

"This confirms the shared logic of Cursor and WorkBuddy: what users want isn't 'a stronger model' but 'a smoother workflow.' When underlying models iterate quickly and differences narrow, whoever can orchestrate multiple models into a complete production pipeline has the moat. What Higgsfield does in video is essentially the same thing Cursor does in code and WorkBuddy does in office productivity."

Another point worth noting is Pollo.ai — traffic 9.16 million, up 18.96%. It follows the same path as Higgsfield: multi-model + workflow integration. The video generation sector is shifting from "whose model is better" to "whose workflow is smoother."

What's Happening Where Traffic Can't Reach?

Two reports from OpenAI and Anthropic fill in what the traffic rankings can't see.

A key data point in OpenAI's Codex report: 80.6% of sampled users submitted tasks equivalent to 30+ minutes of human work, 70.2% submitted 1-hour+ tasks, and 25.6% submitted 8-hour+ tasks. Codex weekly active users exceed 5 million, with over 6x growth since the desktop launch. Knowledge workers account for about 20% of users but are growing 3x faster than the developer cohort — Work is spreading from programming to general knowledge work.

Anthropic's "Cadences" report provides evidence from another angle. The same task required an average of 13 rounds of human interaction in a Chat/Cowork environment, versus a median of just 1 human instruction in the Claude Code environment. Among 31 output categories, 26 showed higher autonomy in the Code environment. The same task in an Agent environment is closer to "goal delegation" than round-by-round co-creation.

The report also has an easily overlooked data point: among 1.2 million sampled Cowork sessions, over 90% were not writing code at all. Number one was business process processing; number two was content creation. Agents are spreading from coding to legal, finance, recruiting, content creation, and other non-technical roles.

Microsoft 365 data corroborates this direction: active Agents grew 15x year-over-year, and large enterprises grew 18x. But the Stanford AI Index 2026 simultaneously notes that while 88% of surveyed organizations already use AI, Agent deployment rates remain in the single digits across most business functions.

The direction is clear; implementation is early. Gartner even gives a more sober prediction: over 40% of Agent projects may be cancelled by 2027 due to unclear cost, unclear value, or insufficient risk control.

Product Directions Are Converging

Several leading companies are expressing the same direction under different names:

ChatGPT Work (OpenAI): for long tasks, cross-app actions, and finished deliverables

Cowork (Microsoft/Anthropic): from conversation to action, supporting background execution

Gemini Spark (Google): from question-answering assistant to all-day work-completing Agent

Kimi Work (Moonshot AI): desktop AI Agent, 300 sub-Agents, local files + browser + code execution

QoderWork (Alibaba): from pair-programming to autonomously shipping features

Trae Work (ByteDance): workspace not chatbot, across Code/Work/Design three modes

WorkBuddy (Tencent): desktop Agent, starting from "jailbreak users" insight, targeting non-developers

These names differ, but the underlying logic is the same: users no longer just ask for an answer — they hand a goal to AI, let it plan, call tools, execute across systems, and deliver the result back.

What This Means for Your Work

Twisting three threads together — June web traffic data, OpenAI and Anthropic reports, and leading companies' product launches — I see a market that is layering.

Layer one: the dividend period for general chat entrances is over. Products like ChatGPT, Gemini, DeepSeek, Doubao, and Kimi — the first wave to scale — have generally entered negative web traffic growth. This doesn't mean users are leaving; it's more likely the natural slowdown after "everyone has tried it and penetration is already high." More importantly, much usage has migrated from "opening a webpage and sending messages" to desktop clients, IDEs, CLIs, and APIs — behaviors not counted in web traffic.

Layer two: emotional companion is the only consumer Chat category that has crossed the cycle. But this doesn't contradict the Work trend — companion value is in the interaction itself; Work value is in task results. They serve different needs.

Layer three: the truly "delegated" Work dividend is currently happening mainly in enterprise and developer scenarios. What we see in the traffic rankings — codebuddy-CN (+30%), Qoder (+15%), and Cursor (MAU +14.6%) — is the tip of the iceberg visible to public data. But Trae Work, Kimi Work, and WorkBuddy — these desktop Agents — are simply outside web traffic monitoring, quietly growing in their respective clients.

"Two things are simultaneously true: C-end traffic is cooling, and Work Agents are exploding. They're just happening in two worlds that haven't fully overlapped yet."

And this means three concrete changes for how you and I work.

First, the metrics for measuring AI products need to change. In the past we looked at visits, MAU, and conversation counts. But now, a user might open WorkBuddy only once a month yet delegate an 8-hour task in that one session — this "low-frequency, high-value" usage model completely overturns the traditional "high frequency = good product" logic. New core metrics should be: task completion rate, deliverable adoption rate, human takeover rate, and repeat delegation rate.

Second, AI is shifting from "consultant" to "colleague." Before, you asked AI a question, it gave you an answer, and you did the rest yourself. Now you tell it a goal, and it decomposes steps, calls tools, executes, and delivers the finished product. The human role shifts from "continuous input and round-by-round revision" to "setting goals, providing constraints, supervising, and verifying." CodeBuddy has already cut requirement delivery cycles from two weeks to two days at Tencent, with code production efficiency up 4x to 5x.

Third, barriers are lowering, but not as fast as you think. WorkBuddy targets exactly those 10,000 "jailbreak users" — non-developers who grit their teeth through IDE interfaces they can't understand to do office work with coding tools. This group proves the demand is huge, but supply hasn't caught up. Gartner predicts 40% of Agent projects may be cancelled by 2027, often because ROI doesn't work out. The direction is right, but enterprise-grade reliability, organizational transformation, and commercial returns are still in early stages.

A fairer judgment: from Chat to Work has become a clear product and technology direction, and market migration is happening, but penetration is "fast but uneven" — fast for early adopters like developers and knowledge workers, slow for the mass market of ordinary office scenarios.

And at WAIC, 208 embodied intelligence terminals and over 300 real devices were on display, with Zhiyuan Yuanzheng A3 Ultra selected as one of the "Ten Treasures of the Venue." Agents are walking out of screens and onto factory floors. When Agents have "bodies," Work's boundaries are completely opened — at that point, it's not just your colleague but your worker, your driver, your assistant.

"Chat won't disappear. It will become the communication interface within Work — you send a message not to get an answer, but to trigger the start of a piece of work."

Unique Research — Authoritative Source of Global AI Market Insights

Unique Research's June 2026 AI Web ranking details are as follows:

AI Website Traffic Ranking: TOP 100 ranking of AI web products by visits and monthly active users, distinguishing global and China markets.

All rankings also provide growth charts, helping investment institutions discover promising projects and helping entrepreneurs and developers identify potential opportunities.

Our Influence and Recognition

Unique Research's data and insights have earned widespread trust and citation from top domestic and international investment institutions, academic organizations, and media, becoming an important benchmark for market analysis, investment decisions, and trend assessment. Our partners and citers include (but are not limited to):

Securities firms and investment banks: Sinolink Securities, CSC Financial, Huayuan Securities, Haitong Securities, and others

Authoritative media: China Securities Journal, TMTPost, Jiemian News, and others

We firmly believe that only data that stands up to verification by industry core participants has long-term value.

We believe in the power of open source and openness. Unique Research hereby makes the following three commitments:

This AI product ranking is permanently open-source and free, with complete methodology and raw data supporting full reproducibility.

This AI product ranking's data sources are strictly limited to independent third-party monitoring platforms; we do not accept self-reported enterprise data as evaluation input.

This AI product ranking's system remains absolutely neutral and does not accept any commercial cooperation or paid behavior that would interfere with ranking results.

How to Use This Ranking?

If you are an investor/analyst: This ranking provides you with a pure data benchmark that can withstand reverse-engineering verification — a reliable foundation for building investment models, discovering undiscovered projects, and conducting cross-validation.

If you are an AI entrepreneur/product manager: You can not only pinpoint competitors but deeply reproduce our methodology to analyze your own product's market performance and optimize iteration.

If you are a researcher/student: The fully open-source nature makes this report an excellent free resource for academic research and market analysis. We encourage innovative research based on this data.

For the complete data behind the charts and rankings shown here, please visit 100aiapps.cn. All visible data can be copied and is freely available for personal study, research, and other non-commercial use. Commercial use requires prior authorization from Unique Research. We encourage listed teams to apply rankings to business plans, recruiting materials, and similar contexts under compliant use, providing authoritative support for their capabilities.

Join the AI Think Tank Open-Source Community

Unique Research is not only focused on AI industry research but aspires to build an open-source community promoting data transparency in the AI market. We invite developers, data scientists, and industry analysts to:

Reproduce and verify: Visit our Feishu knowledge base for complete data and methodology, verify our results, and propose improvements.

Co-build: Contribute code to help us expand data dimensions (such as API call volume estimation, Agent monitoring models, etc.).

Exchange insights: Join our Feishu group to share insights derived from Unique Research data with peers.

One person's insight is limited; a group's wisdom is infinite. Let us define the standard for measuring the AI market together.

01\. AI Website Traffic Ranking

1.1 Global AI WEB TOP 100 by Monthly Visits

Based on its independent research on the AI industry and continuous tracking of thousands of AI companies and nearly 10,000 AI products globally, Unique Research ranks the top 100 AI web products worldwide by cumulative monthly visits. The specific ranking is as follows:

1.2 Global AI WEB TOP 100 by Monthly Active Users

Based on its independent research on the AI industry and continuous tracking of thousands of AI companies and nearly 10,000 AI products globally, Unique Research ranks the top 100 AI web products worldwide by monthly active users. The specific ranking is as follows:

1.3 China AI WEB TOP 100 by Monthly Visits

Based on its independent research on the AI industry and continuous tracking of thousands of AI companies and nearly 10,000 AI products globally, Unique Research ranks the top 100 AI web products by Chinese entrepreneurs targeting the domestic market, sorted by cumulative monthly visits. The specific ranking is as follows:

1.4 China AI WEB TOP 100 by Monthly Active Users

Based on its independent research on the AI industry and continuous tracking of thousands of AI companies and nearly 10,000 AI products globally, Unique Research ranks the top 100 AI web products by Chinese entrepreneurs targeting the domestic market, sorted by monthly active users. The specific ranking is as follows:

02\. Research Statement

2.1 Data Description

The AI product revenue data referenced in this report, article, ranking, or chart is based on real-time tracking of redirect traffic from AI product official websites to payment gateways (such as Stripe), combined with time-series analysis, deeply analyzing historical visits, monthly active user fluctuations, and pricing strategy changes to build a dynamic revenue estimation model. Core model parameters are determined through a triple validation system: paid conversion funnel monitoring, reverse calibration against enterprise disclosures, and industry benchmark conversion rate comparison, with timely correction against corporate financial reports, financing announcements, and authoritative media disclosures. Through sampling verification, the average absolute error rate between predicted and actual values is stable within ±10%, meeting industry research-grade accuracy standards.

The AI product traffic data referenced in this report, article, ranking, or chart is based on multi-source data integration and Unique Research's intelligent algorithm processing. Monitoring scope is strictly limited to users' direct access through official apps and websites. Specifically, this data only covers users directly accessing AI products through websites and native apps, and does not include traffic from: browser plugins/extensions, desktop client software, WeChat/Alipay mini-program ecosystems, embedded services on third-party platforms such as Discord, local deployments of open-source models, API calls, and other non-direct access scenarios. This data focuses on core end-side access behavior monitoring of AI products, aiming to objectively reflect direct usage on mainstream user terminals.

Given the complexity and dynamism of data collection and processing, and continuously changing market conditions, the data shown may contain some degree of error and omission. Therefore, this data should be viewed as a reference for research and analysis, aiming to provide a window into understanding the AI product market, rather than a definitive basis for formulating specific investment strategies or providing advisory recommendations.

Data coverage expansion note:

We are actively developing and testing monitoring models for the following emerging areas, expected to be gradually incorporated in future quarterly reports:

AI Agent ecosystem

AI hardware device sales and activity

Large model API call volume estimation

Stay tuned.

2.2 Concept Definitions

• Geographic dimension

Overseas AI products: Products founded by non-Chinese entrepreneurs or teams and primarily launched for the global market (including but not limited to their home market).

Domestic AI products: Products founded by Chinese entrepreneurs or teams and primarily launched for the China domestic market.

Going-global AI products: Products founded by Chinese entrepreneurs or teams but primarily targeting overseas markets (non-China domestic market).

• Functional dimension

AI-native applications: Applications that deeply integrate AI technology and algorithms from the earliest product design stages, where core value, business processes, or user experience are entirely built on AI. These applications don't just use AI to optimize or enhance existing features but treat AI as their core component — without AI, these applications would cease to exist or lose their core value.

AI-feature applications: Applications that enhance existing features or provide new ones by integrating or embedding AI technology on top of existing business logic and application frameworks. These applications may already have mature business models and market positioning, but introducing AI technology significantly improves efficiency, user experience, or creates new value points.

AI ecosystem applications: Platform-type applications that support, connect, or facilitate exchange for the AI ecosystem comprising AI technology developers, application developers, service providers, and end users. These applications go beyond the technical level, involving market, community, resource sharing, and other dimensions, aiming to promote AI technology adoption, innovation, and collaboration.

2.3 Metric Definitions

• WEB data metrics

Visits: The sum of all website visits within a specific time period (e.g., one month), measuring website traffic scale. Page views within a continuous active session count toward the same session; a user's reactivation after 30+ minutes of inactivity or a new calendar day is counted as a new visit, comprehensively and accurately measuring website traffic and visitor activity.

Unique Visitors: The number of unique IP addresses visiting the target website within a specific time period (e.g., one month). If one person visits a website on multiple days within a month, they are counted as a single unique visitor.

ARR (Annual Recurring Revenue): The total revenue a website generates from subscription services over a period (e.g., one year), excluding advertising revenue, transaction commissions, and one-time revenue such as professional services.

• APP data metrics

Downloads: The number of times an app is downloaded by users within a specific time period (e.g., one month), typically measured as downloads per app on app stores (e.g., App Store, Google Play). Downloads are a key indicator of app popularity and user appeal, reflecting the effectiveness of app promotion and marketing campaigns.

Active Users: The number of unique users who performed at least one activity within a specific time period (e.g., one month). Active users measure user engagement and activity levels, typically used to assess app user scale and user activity.

IAP (In-App Purchase): Revenue generated from users purchasing virtual goods, additional services, or premium features within an app, excluding advertising revenue, direct user payments (such as tips), and third-party Android app store revenue.

2.4 Disclaimer

This report is published by Unique Research, and its copyright belongs to Unique Research. Any Chinese reprint or citation must credit the report source; overseas organizations seeking to reprint or cite should contact us in advance for authorization. When citing data, please mark: \[Data source: Unique Research\]. When directly citing charts, please mark: \[Data source: Unique Research, Chart production: Unique Research\].

This report is independent original analysis by Unique Research as a third-party organization. Its content does not represent the position of any enterprise and does not constitute investment advice to anyone. Investors should note that any investment decisions made based on this report are unrelated to Unique Research, its employees, or affiliated entities.

To the extent permitted by law, Unique Research and its affiliated entities may hold equity in companies mentioned in this report, or provide or seek to provide fundraising, financial advisory, or related services for such companies, and its employees may serve as directors of companies mentioned in this report.

---

Original publication: https://uniqueresearch.substack.com/p/src-20260718-01html
On-site reading page: https://ffcap.cn/en/research/src-20260718-01html