When models begin to be "sufficient," what will be the next battleground for AI?

On September 21, Meta closed up more than 11%, marking its biggest single-day gain since April last year, adding more than $190 billion to its market capitalization in a single day.

The catalyst for this surge was Meta's recent launch of Muse, a personal AI assistant. Sensor Tower data shows that within six days of its release, the app accumulated over 902,000 downloads on iOS in the US, far surpassing the performance of Meta's previous generation product. By Monday, Muse simultaneously held the top spot on both the US iOS and Google Play free app charts, dethroning ChatGPT.

The market is buying into Muse's immense popularity. Technology analyst Ben Thompson commented in his article "Frontier Overhangs": "Muse Spark 1.3 isn't the most cutting-edge model, but that's a warning sign—a non-leading model is enough to support a highly sticky personal agent product."

Thompson believes that current frontier model capabilities are already "excessive," and model capabilities are no longer the most important moat for AI companies. The next turning point may be the "application moment."

After model capabilities crossed a critical point, the focus of AI competition is shifting. From Meta Muse for consumers to Tencent WorkBuddy for businesses, a common signal is emerging: what determines product value is increasingly not the model itself, but control over Harness, workflow, and user access.

On September 22, Tencent echoed the overnight rise in Meta, with its Hong Kong-listed shares jumping more than 6% in early trading, reaching a high of HK$463.40 during the session.

Meanwhile, research from consulting firms and venture capital firms is validating the same judgment from different dimensions. Gartner warns that by 2030, enterprise software spending of up to $234 billion faces the risk of "agent arbitrage"; a16z explicitly recommends that AI application companies stop pricing by token and shift to charging based on results ; data from more than 300 AI software companies surveyed by ICONIQ shows that the application layer has become the focus of industry development , and the gross margin of AI products is climbing from 45% in 2025 to an estimated 53% in 2026.

Is the "application moment" approaching?

Thompson proposed a key concept: when the capabilities of frontier models continue to improve, but the actual products and user scenarios cannot keep up with the speed at which they can absorb these capabilities, a situation of "frontier overhang" will occur.

He cited a specific example: Fable, an AI narrative game company, launched a new generation of models, but the market response was lukewarm. This was not because the product quality was poor, but because the existing model's capabilities were already "sufficient" for the scenario. The stronger model did not bring a perceptible improvement in the user experience, and naturally could not be converted into a higher willingness to pay or a larger market share.

Thompson believes that when performance is not enough, whoever binds the model and application layer (harness) tightly wins; once the "enough" line is crossed, customers start to care about time to market, convenience, customization, and how data is handled, and modularity will have an advantage.

This detail is significant. It illustrates that, in certain scenarios, the marginal value of model capabilities is diminishing. When a "stronger model" no longer equates to a "better product experience," the competitive advantage at the model layer begins to decouple from the business value at the application layer.

Thompson further points out that, in this context, what truly matters is the "harness"—that is, how to effectively organize, schedule, and encapsulate model capabilities so that they can stably and reliably complete tasks in specific scenarios. The model is the engine, and the harness is the transmission system. Without the latter, no matter how powerful the former is, it cannot output effectively. This framework provides a clear coordinate for understanding the product logic of Meta Muse and Tencent WorkBuddy.

Consumer-facing (C-end): Meta Muse's Product Bet

Meta's launch of Muse is a typical "application-layer bet." According to Meta's official introduction, Muse is positioned as a personal AI agent, with its underlying model being Muse Spark 1.3. However, the core selling point of the product is not the scale of model parameters or benchmark test scores, but the agent's ability to actually complete tasks—helping users make plans, execute steps, and track progress, truly "getting things done" in users' daily life scenarios.

Behind this positioning is Meta's assessment of the competitive landscape of AI products: with relatively abundant model capabilities, what users truly lack is not "smarter models," but "tools that can help me get things done." Muse aims to become the user's personal execution layer, rather than just a question-and-answer interface.

Looking further, Muse's product focus has shifted the breakthrough point for C-end agents from "whether the model is smart enough" to "whether users can confidently entrust tasks and whether the product can connect to real-world transactions." While the Muse Spark series models have helped Meta fill the gaps in its basic capabilities, what determines whether an agent can enter users' daily lives is not just reasoning and generation capabilities, but also authorization, trust, and execution chains. To this end, Muse incorporates an independent operating environment, credential protection, and key action approval into its product design to establish clear authorization boundaries; simultaneously, it combines long-term memory and background execution capabilities, enabling the agent to continuously advance tasks around user goals, rather than stopping at one-off question-and-answer or command responses.

From a business perspective, this design has profound implications. Once a personal agent is deeply embedded in a user's daily workflow, it accumulates a large amount of personal data, preferences, and usage habits, forming user stickiness that is difficult to migrate. This stickiness is the true moat—not the capabilities of a particular version of the model.

The market's positive response to Muse's release confirms this logic to some extent: investors' interest in "usable" agent products has surpassed their focus on "stronger" models.

B2B: Tencent WorkBuddy's Enterprise Path

On the enterprise side, Tencent WorkBuddy presents a similar product logic. According to Tencent Cloud's official introduction, the core capability of WorkBuddy Enterprise Edition lies not in providing a general chat interface, but in its ability to connect various internal systems and processes within an enterprise, execute work tasks across platforms, and truly embed AI capabilities into the daily operations of the enterprise.

This represents a crucial leap for enterprise AI, moving it from a "tool" to an "execution layer." In the past, when enterprises purchased AI products, they were essentially buying a smarter search or question-and-answer tool; while WorkBuddy represents a direction where AI becomes an agent capable of proactively initiating, executing, and completing work tasks, directly participating in business processes.

The business implications of this shift are equally profound. When AI can accomplish tasks across multiple software systems, the "functional value" of traditional enterprise software faces a direct challenge. Gartner's July 2026 report explicitly points out that the rise of Agentic AI puts up to $234 billion in enterprise application software spending at risk—because agents can bypass or replace workflows that originally required multiple separate software programs, potentially eroding the "functional moat" of traditional software.

This is a structural pressure that enterprise software companies need to take seriously.

The repricing of business value: from token to outcome

The "sufficiency" of model capabilities is driving a fundamental shift in AI business models.

Venture capital firm a16z, in its report "You are not a model. Don't price per token," directly points out this trend: the true value of AI applications comes from the data, tools, workflows, and quantifiable results they integrate, not from which model is invoked or how many tokens are consumed. In other words, the pricing logic of "selling model capabilities" is giving way to the pricing logic of "selling results."

a16z offers three pricing models: selling model access, priced per token; turning the model into useful work, priced in units of value recognized by the customer, typically credits; and delivering clear, attributable business results, priced per result. Of the 50 enterprise AI technology buyers surveyed, 27 preferred "credits linked to identifiable workload," while 14 preferred paying per token.

This has a direct impact on the value chain distribution of the AI industry. If models themselves gradually become a fundamental capability—similar to computing power in the early days of cloud computing—then those application-layer companies that can truly capture commercial value will be those that can build specific products, vertical workflows, and verifiable ROIs on top of models.

McKinsey's report, "Where AI agents pay off," provides corroborating evidence from a corporate perspective: as token costs continue to decline, the core issue companies focus on when evaluating AI investments has shifted from "how powerful this model is" to "how much quantifiable efficiency improvement and economic return this agent can bring." The verifiability of ROI is becoming a core criterion for corporate AI procurement decisions.

This means that the question AI application companies need to answer has changed from "What model did we use?" to "How much money can we help our clients save and how much efficiency can we improve?"

Industry Landscape: Application Layer Competition in the Multi-Model Era

In its 2026 State of AI Report, ICONIQ surveyed more than 300 AI software companies, revealing a clear industry trend: nearly two-thirds of the surveyed companies are developing horizontal or vertical AI applications, and the use of multiple models has become the norm.

The fact that "multi-model usage has become the norm" is itself a strong indication of the trend towards model commoditization. When enterprises can flexibly switch between different models according to task requirements, the bargaining power of model providers will be suppressed, while application layer platforms that can effectively integrate multiple models and build stable workflows will gain a stronger bargaining position.

ICONIQ's report tracks changes in both AI revenue and gross margin. The importance of this data dimension lies in its ability to help the market determine whether application-level business value capture is truly occurring or merely a narrative. For investors, this is a key indicator distinguishing between an "AI application story" and an "AI application business."

From an industry structure perspective, the current emerging landscape is as follows: the model layer (OpenAI, Anthropic, Google, Meta, etc.) provides foundational capabilities; the application layer (various vertical agents and workflow platforms) is responsible for translating these capabilities into deliverable results; and traditional enterprise software companies face pressure to be bypassed or replaced. The value distribution among these three layers will be the most noteworthy dynamic to observe in the AI industry over the next few years.

The traditional agent logic of SaaS is being penetrated.

If an agent can complete tasks across systems, employees won't need to frequently access each traditional software interface. The software will still run, but the interface can be moved to the background.

Gartner predicts that by 2030, up to $234 billion in enterprise application software spending will be at risk from Agentic AI, representing about 20% of enterprise application SaaS spending.

"Agentic AI is changing the economics of software," said George Brocklehurst, managing partner at Gartner. "Agent systems deliver results directly, bypassing applications that heavily rely on user interfaces, making software invisible." The traditional link between agent numbers and software revenue growth may therefore be weakened.

The pressure first falls on vendors who rely on dashboards, feature modules, and agent licensing fees. Enterprise buyers won't increase their budget just because there's an AI button; they'll ask whether the agent reduces manual operations, shortens processes, improves conversion rates, and lowers unit costs.

But this is not the end of traditional SaaS. Vendors with industry data, customer relationships, system connectivity, and business process control can still restructure themselves into agent execution layers. The risk lies in charging based on the interface, while the opportunity lies in securing the position of task execution and result delivery.

If the model is leading-edge, can it be directly monetized?

It's important to clarify that the above analysis does not imply that "models are unimportant." Cutting-edge model capabilities remain the foundation of the entire AI industry; without sufficiently strong models, building application layers is impossible. The real question worth discussing is: can model superiority directly translate into product advantages and commercial value?

Current evidence suggests that this transformation path is becoming longer and more complex. The poor demand for Fable's new models illustrates that, in some scenarios, stronger models no longer deliver tangible improvements in user value. The product logic of Meta Muse and Tencent WorkBuddy demonstrates that leading tech companies are already using a "Harness application layer + workflow + user entry point" to build differentiation, rather than simply relying on model capabilities. Research by a16z and McKinsey shows that enterprise clients' payment logic is shifting from "model capabilities" to "quantifiable results."

This means that competition in the AI industry is entering a new phase: model capability is a necessary condition, but no longer a sufficient one. Whoever can translate general-purpose models into concrete product experiences, stable workflows, and verifiable business returns will truly gain a competitive advantage in the next phase of competition.

This shift suggests a valuation framework worth re-examining: while model parameter size and benchmark rankings are important in the AI field, the depth of application-layer workflows, user engagement, and the verifiability of ROI may be becoming more predictive indicators of business value. As models become "sufficient," the battlefield for the next battle has quietly shifted.

Risk Warning and DisclaimerInvesting involves risk; please exercise caution. This article does not constitute personal investment advice and does not take into account the specific investment objectives, financial situation, or needs of individual users. Users should consider whether any opinions, views, or conclusions in this article are suitable for their specific circumstances. Any investment decisions made based on this information are at your own risk.