Claude Pro recorded 11 opens while dual subscriptions cost 290 yuan

Productivity applications such as ChatGPT, Claude, Gemini, and Grok drove US productivity app revenue up by approximately 100% and user time spent up by approximately 55%. Yet this surge masked an engagement void. Reporting from ifanr highlights that monthly newly released iOS apps increased from approximately 40,000 to 120,000 between early 2025 and April 2026. Across major app stores following early 2025, the proportion of newly launched apps with fewer than 10 ratings and fewer than 100 downloads climbed significantly, swelling the low-engagement software Sternstein coined 'App-Slop'.

Individual buyers are not sustaining the deluge. As ifanr reported, Consumer Edge data cited by a16z indicates that approximately 3% of US consumers personally paid for AI services as of the first quarter of 2026.

The standalone wrapper provides little durable utility. Real advantage arrives only when software moves from speculative downloads into repeated operating workflows governed by measurable accountability and human judgment.

The gym membership trap of personal AI subscriptions

A marketing professional tracked by woshipm spent 290 yuan per month maintaining concurrent Claude Pro and ChatGPT Plus subscriptions, yet opened Claude Pro only 11 times in a month for basic drafting and translation tasks. That expense stands apart from ordinary software costs. In China, standard video streaming memberships for platforms such as iQIYI and Tencent Video cost approximately 20-some yuan per month. Claude Pro and ChatGPT Plus each demand approximately 145 yuan per month. The gap between billing and utilization is vast.

The woshipm author compares underutilized AI subscriptions to a gym membership business model. Providers collect predictable monthly fees from subscribers who rarely open the service, generating profit from friction rather than delivered output.

Most consumers have not fallen into that pattern. According to ifanr, Consumer Edge data cited by a16z indicates that approximately 3% of US consumers personally paid for AI services as of the first quarter of 2026. The payer base skews young. In that period, 5.9% of consumers aged 18 to 24 and approximately 5% of those aged 25 to 34 paid for AI services, compared to 1.4% of those aged 65 and older. The proportion of cardholders paying for at least one AI service roughly doubled between 2025 and 2026, but overall adoption remains thin.

Individual spending accelerates quickly when tools accumulate. The woshipm author reported spending nearly 1,000 yuan on AI tools in a single month, allocating 559 yuan for Kimi, 145 yuan for ChatGPT Plus, 200 yuan for Qwen, and 300 yuan for DeepSeek API credits. That expansion mirrors broader market totals. Driven by ChatGPT, Claude, Gemini, and Grok, US productivity app revenue grew by approximately 100% while user time spent increased by approximately 55%. Paying 145 yuan each month for sporadic text generation leaves users funding capabilities they never convert into continuous work.

The economic owner test and employer capture

A payment framework from woshipm benchmarks AI pricing models against four tangible counterparts. Free tiers resemble WeChat and Meituan, while pay-as-you-go packages match video streaming memberships. Monthly subscriptions reflect hiring hourly domestic workers, and professional tiers act like buying production equipment or paying employee salaries. Yet framing workplace AI as a personal helper collapses under corporate dynamics. Columnist Zhang Mu argues that charging consumer users directly for workplace tools will fail because workers will not pay out of their own pockets for software that merely expands their corporate workload.

Organizations capture those efficiency gains immediately. Extra output quickly becomes the daily baseline. For that reason, Zhang Mu argues that organizations rather than individual workers should be the paying party for productivity software.

Corporate procurement habits already confirm this migration toward organizational budgets. Most organizations in provincial capital cities and higher-tier cities contacted by Zhang Mu purchased Claude or Codex membership accounts. In second- and third-tier Chinese cities, enterprises contacted by Zhang Mu purchased AI tokens from Alibaba and Zhipu AI, or secured token plans from Xiaomi and MiniMax.

This budgetary discipline arrives as broader technology economics cool down. Silicon Valley Bank data cited by a16z shows median tech revenue growth rates dropped from roughly 40%-70% in early 2022 to roughly 15%-30% by 2026. The bank estimated that approximately 2,345 venture-backed US companies would fail or shut down in 2026, with over one-third founded between 2019 and 2021. Meanwhile, ifanr argues that the primary value of vibe coding lies in reducing the cost of validating product ideas rather than reducing the cost of building a viable business. True business infrastructure requires an organizational sponsor.

Free platform utilities versus specialized agent infrastructure

A woshipm author classifies AI users into six consumer tiers. Tier 1 and Tier 2 users rely entirely on free tools like Doubao, Yuanbao, Kimi, and unpaid versions of DeepSeek and Qwen. Tier 3 users cap monthly spending below 100 yuan. Tier 4 custom agent builders spend under 200 yuan, while Tier 5 vibe coding enthusiasts spend 200 to 300 yuan combining subscriptions with the DeepSeek API. At the top, Tier 6 professionals spend hundreds to thousands of yuan on Claude Code and Codex, treating software as capital equipment evaluated on direct returns.

These middle consumer tiers face rapid elimination. When tech giants bundle advanced models into free workplace utilities, personal software subscriptions lose their economic justification.

Chinese platforms are aggressively converting standalone agent features into free workplace infrastructure. Alibaba unified three agent tools-QoderWork, Wukong (悟空), and MuleRun-into Qianwen Office (千问办公), offering free access to its Qwen Max model for work tasks. ByteDance took a similar path. It integrated TraeWork directly into Feishu (Lark) without point deductions or paid memberships, while Tencent merged Qclaw into Workbuddy to distribute daily login points. As columnist Zhang Mu observes, office AI tools cannot survive on direct user subscription fees. They compete on delivery capability and platform scale instead.

Standalone tools also struggle because raw generation does not equal delivered value. An ifanr report on an MIT and University of Pennsylvania study, titled 'Writing Code vs. Shipping Code', marked February 2025 as the beginning of the 'Agentic coding era'. The metric gap was stark. Analyzing GitHub activity across more than 100,000 developers, researchers found that AI tools increased code commits by approximately 180% but expanded published releases by only about 30%. Creating code became cheap, but shipping software still required structured organizational pipelines.

Three Paradigms of AI Adoption: Personal Subscriptions, Vibe Coding, and Enterprise Workflows

The evaluation gap and App-Slop accumulation

According to ifanr, researchers from MIT and the University of Pennsylvania detailed this productivity imbalance in a study titled 'Writing Code vs. Shipping Code'. Examining GitHub activity across more than 100,000 developers, the study found that cumulative adoption of AI coding tools increased code commits by approximately 180% but increased published software releases by only about 30%. Sternstein coined the term 'App-Slop' to describe the resulting deluge of low-engagement software flooding app stores. Abundant syntax does not guarantee completed work.

Generating code is cheap, but finishing software remains rare. When evaluation lags behind generation, teams accumulate unverified volume instead of usable tools.

This imbalance mirrors structural economics. In development economics, the middle-income trap refers to an economy that stagnates after reaching middle-income levels because its original low-cost labor and resource advantages disappear. The sspai author argues that users of AI agents face an identical bottleneck as models become cheap and commoditized. Tool-bound skills-such as prompting techniques tailored to specific model generations or tuning particular harness frameworks-rapidly lose their value. The initial productivity boost evaporates.

To escape that stagnation, the sspai author separates tool-bound tricks from cognitive judgment capabilities. Durable competence requires identifying fundamental problems, determining required context for agents, setting quality criteria, and deciding when to accept or reject agent outputs. Cognitive science indicates that the brain relies on discrepancies between predictions and actual outcomes as a critical learning signal. The sspai author models this judgment as an iterative loop: prediction, actual result, prediction error, explaining the error, and updating the mental model. The filter matters more than the engine.

Workflow-Centric AI Resource Allocation Framework

• Validating new software or product concepts amidst an influx of low-engagement apps Use vibe coding setups (spending 200 to 300 yuan per month across a pro subscription and APIs like DeepSeek) strictly to lower idea validation costs. Do not treat automated code generation alone as a viable business strategy, especially as monthly iOS app launches increased from approximately 40,000 to 120,000 between early 2025 and April 2026 while sub-10-rating and sub-100-download releases surged.
• Managing engineering productivity where code generation outpaces actual deployments Focus developer effort on cognitive judgment-problem identification, agent context setting, and acceptance criteria-rather than tool-bound prompt tweaks. Cumulative AI adoption increases code commits by approximately 180% but published releases by only about 30%; reserve professional tooling expenditures (such as Tier 6 investments in Claude Code or Codex) for workflows with proven return on investment.
• Handling routine information retrieval, drafting, or occasional office tasks Rely on free Tier 1 and Tier 2 tools (such as Doubao, Yuanbao, Kimi, or free versions of DeepSeek and Qwen) or platforms offering subsidized access (such as Qianwen Office providing free Qwen Max access, TraeWork on Feishu without points or paid memberships, or Workbuddy daily login points) rather than purchasing individual paid subscriptions.
• Assessing individual employee software subscriptions for corporate productivity tasks Halt personal out-of-pocket subscriptions for corporate work, avoiding the gym-membership pattern where users pay 290 yuan per month for tools like Claude Pro and ChatGPT Plus but only access them 11 times. Shift procurement to organizational accounts (such as Claude or Codex seats in major cities, or token packages from Alibaba, Zhipu AI, Xiaomi, or MiniMax in other regions), as organizations should fund tools that increase enterprise output.
• Defending against productivity commoditization and the AI middle-income trap Prioritize intrinsic capital (human judgment and cross-domain taste) and compounding external capital (engaged user communities, high-follower accounts, and proprietary iterative workflows) over transient, tool-bound skills like generational prompt engineering or framework tuning, which diminish in value as model execution cheapens.

Preserving taste and compounding judgment in repeated loops

Since ChatGPT was launched at the end of 2022, generative utilities have dramatically sped up early-stage building. ifanr observes that the median age of newly minted unicorn startups has dropped to 4 years. Yet as ifanr points out, the real merit of vibe coding is cutting the expense of testing product ideas rather than decreasing the cost of creating an enduring enterprise. Human judgment decides what survives.

The sspai author argues that human judgment and human-oriented taste function as intrinsic capital. Because automated agent execution cheapens by the month, these personal faculties resist depreciation.

Cultivating that judgment relies on focused immersion rather than mechanistic prompting. Product designer Iglika contends that design taste forms through sustained personal exposure to high-quality products until intuition takes over, rather than through rigid tutorials. Thiago Costa, a designer and co-founder of Fey, similarly argues that building taste requires gathering design inspiration across distinct disciplines. He points to art, architecture, film, and games instead of staying bounded within a single field.

This refined intuition translates into durable assets outside the model. According to the sspai author, practitioners protect their work by cultivating proprietary iterative workflows alongside core engaged user communities. High-follower social media accounts provide another compounding channel. Under the woshipm author's classification, Tier 6 professional users spend hundreds to thousands of yuan on tools like Claude Code and Codex. They treat AI software strictly as capital production equipment evaluated on direct return on investment. The woshipm author argues that a user's willingness to pay for these systems is determined primarily by perceived service value and usability, rather than by personal income or educational background.

Practitioners managing sporadic duties can offload execution costs onto free tiers. Qianwen Office (千问办公) provides free access to the Qwen Max model, while TraeWork operates inside Feishu (Lark) without point deductions or memberships. Daily login points from Workbuddy cover occasional tasks without expense. According to woshipm columnist Zhang Mu, charging individuals for workplace AI tools will fail because staff will not pay out of pocket to increase company workloads.

Organizations must be the paying party instead. Zhang Mu argues that office AI tools will ultimately compete on enterprise delivery capability or foundation model capability, rather than direct software subscription fees.

Purchasing patterns already follow this split. Most organizations in provincial capitals and higher-tier cities contacted by Zhang Mu purchased Claude or Codex memberships. Meanwhile, enterprises in second- and third-tier Chinese cities bought tokens from Alibaba and Zhipu AI, or token plans from Xiaomi and MiniMax.

For readers outside China

• Availability: The domestic Chinese productivity suites and agent platforms discussed-including ByteDance's TraeWork and Feishu, Tencent's Workbuddy, Alibaba's Qianwen Office, Kimi, Doubao, and Yuanbao-are primarily documented in the context of mainland Chinese enterprise and consumer ecosystems. International availability, regional account restrictions, and overseas service terms for these specific platforms are not disclosed in sources.
• Pricing: Individual Western subscriptions cited include Claude Pro and ChatGPT Plus at approximately 145 yuan per month (or 290 yuan per month when held concurrently). By comparison, standard video streaming memberships in China (iQIYI and Tencent Video) cost approximately 20-some yuan per month. Documented spending tiers classify free users in Tiers 1 and 2; Tier 3 users maintaining simple templates with willingness to pay capped below 100 yuan per month; Tier 4 agent builders spending under 200 yuan per month; Tier 5 vibe coders spending 200 to 300 yuan per month; and Tier 6 professional users spending hundreds to thousands of yuan per month on tools like Claude Code and Codex. One professional user reported a monthly expenditure of nearly 1,000 yuan (including 559 yuan for Kimi, 145 yuan for ChatGPT Plus, 200 yuan for Qwen, and 300 yuan for DeepSeek API credits). Enterprise token access is available from providers such as Alibaba, Zhipu AI, Xiaomi, and MiniMax, though exact commercial token rates are not disclosed in sources.
• Closest Western equivalents: ChatGPT Plus; Claude Pro; Claude Code; Codex; Gemini; Grok
• Data residency: The source material does not cover data residency, server infrastructure locations, cross-border data transfer protocols, or regulatory compliance frameworks for the Chinese platforms (such as Workbuddy, Feishu, Qianwen Office, or DeepSeek) or their Western counterparts.

Sources

• ifanr AI 正在制造 App 过剩时代 https://ifanr.com/1681265
• woshipm 免费用户不是抠，他们只是把AI当成了微信和美团外卖 https://woshipm.com/ai/6446238.html
• sspai 当 AI 让执行力变得廉价，我们该拿什么脱颖而出？ https://sspai.com/post/114519
• woshipm 无论是Workboddy、千问办公还是TraeWork，收费=没戏！！！ https://woshipm.com/ai/6452074.html