A single market report from Moomoo says ByteDance’s Doubao and Alibaba’s Qwen are preparing to discontinue AI agent features, pointing to a possible retrenchment in one of the most heavily promoted areas of consumer AI products. If accurate, the move would matter less as an isolated product tweak and more as a signal that large Chinese AI platforms are reassessing how quickly autonomous agent workflows can be turned into durable mainstream features.

The source evidence available for this story is thin: the underlying article text is not available, and no direct statement from ByteDance, Alibaba, Doubao, or Qwen is included in the material reviewed here. That means the reported change should be treated as unconfirmed pending company disclosure or product updates. Even with that caveat, the report is notable because AI agents have been positioned across the industry as the next layer beyond chatbots, promising multi-step task execution rather than simple question answering.

What the report appears to say

According to the Moomoo item, "Doudou" and "Qianwen" are set to discontinue their AI agent features. Based on widely used English naming, that appears to refer to Doubao, ByteDance’s AI assistant, and Qwen, Alibaba’s family of models and related assistant products. The evidence provided does not specify which exact agent tools are affected, whether the reported discontinuation is temporary or permanent, or whether the change applies to all users, specific markets, or only certain in-app features.

That lack of detail matters. In the current AI product market, "AI agents" can describe very different things: browser-use tools, workflow automation modules, no-code task builders, in-app assistant actions, or developer-facing orchestration systems. A company can remove one visible consumer feature while still investing heavily in backend agent infrastructure. Without fuller sourcing, it is not yet possible to tell whether the reported move is a broad strategy shift or a narrower product cleanup.

Still, the pairing of Doubao and Qwen in the same report suggests a common commercial issue: consumer-facing agent experiences may be generating less sustained value than expected, especially when they are expensive to run, hard to evaluate, and difficult for users to trust on open-ended tasks.

Why AI agent features are under pressure

The reported pullback fits a wider pattern in AI product development. Companies have rushed to add AI agents to show progress beyond basic chat, but production agents remain difficult to deploy well. They often require more tool access, more memory, and more retries than standard assistant interactions. That raises cost and latency while also creating new failure modes.

For consumer products such as Doubao, the challenge is especially acute. A general-purpose assistant can impress users with a one-turn answer, but an agent is judged on completion: Did it actually book, search, summarize, coordinate, or follow through correctly? If not, the feature can feel unreliable rather than magical. That can hurt engagement faster than a conventional chatbot feature that makes fewer promises.

The same problem applies to Qwen-linked experiences if Alibaba has indeed chosen to remove some agent capabilities. For a feature to survive inside a large-scale assistant, it needs more than technical novelty. It needs predictable task boundaries, clear user controls, sensible cost economics, and support metrics that show real repeat usage. Many agent demos look compelling in launch materials but break down in daily use when tasks become messy, permissions are ambiguous, or the model loses context mid-workflow.

This is why some companies are narrowing scope rather than abandoning the category. Instead of broad autonomous behavior, they are emphasizing bounded AI agents inside customer support, software engineering, sales operations, and internal knowledge retrieval, where task structure is clearer and return on investment is easier to measure.

Evidence, attribution, and what remains unverified

The strongest fact available in this story is only that Moomoo published a report saying the AI agent features tied to Doubao and Qwen are set to be discontinued. The full article text was unavailable in the source material reviewed for this piece. No official product note, company blog, app changelog, executive quote, or regulatory filing was included.

That creates several unresolved questions:

First, it is unclear whether the report refers to Doubao and Qwen branded consumer apps, to embedded assistant functions, or to separate agent-building tools. Second, the reason for the reported discontinuation is not given in the available evidence. It could reflect low usage, product simplification, rising inference costs, safety concerns, weak reliability, or a shift toward different interfaces. Third, no timing details are confirmed beyond the phrasing that they are "set to" discontinue those features.

Because all available evidence comes from a single media item, builders and buyers should not treat this as a confirmed shutdown until ByteDance, Alibaba, Doubao, or Qwen update their products or issue statements. In fast-moving AI markets, product labels are also fluid. Features can disappear, be folded into another workflow, or reappear under a different name.

This is also a reminder about vendor and market reporting standards in AI. Performance claims for AI agents are often benchmark-driven or demo-driven, while discontinuation signals tend to surface through app changes or user reports before companies frame them publicly. In the absence of primary evidence, caution is warranted.

What this could mean for builders and enterprise teams

If major consumer AI products are stepping back from visible AI agents, the lesson for builders is not that agent architectures are over. It is that broad autonomy is a harder product to ship than many launch narratives suggested. Product teams building on AI agents should assume that users will tolerate less error than they do with a conversational assistant, because the system is taking action rather than only generating text.

For founders, one implication is focus. Tools that expose bounded, testable workflows may fare better than open-ended agent promises. An internal procurement assistant, coding helper, or support triage system can be evaluated against completion rates and business outcomes. A consumer agent asked to handle arbitrary life admin has to solve too many edge cases at once.

For enterprise AI buyers, any reported retreat by Doubao or Qwen would reinforce the case for careful rollout discipline. Teams considering agent deployments should ask practical questions: What tools can the model access? How are actions logged? Can workflows be sandboxed? What is the fallback when a task fails? How much human approval is required? Those questions matter more than whether a product is marketed as an agent.

The story also has competitive implications. If ByteDance and Alibaba are indeed trimming public-facing agent features, that may leave more room for specialized AI agents in vertical software, workplace automation, and developer tools. It may also push large platforms to reposition agents as APIs and orchestration layers rather than headline consumer features.

A broader reset in enterprise AI and workplace automation

The most plausible interpretation of this report is not a rejection of agent technology itself, but a reset in where it fits. Across enterprise AI, buyer interest remains strongest where automation can be constrained, audited, and priced sensibly. That is why many successful deployments look closer to workflow software with model assistance than to fully independent digital workers.

In workplace automation, companies increasingly prefer systems that draft, route, retrieve, and recommend while keeping humans in the loop. That lowers the cost of mistakes and makes procurement easier for regulated or risk-sensitive teams. If consumer platforms such as Doubao and Qwen are adjusting course, it could reflect the same economic reality: users like helpful automation, but they do not necessarily want opaque autonomy.

For model providers, this matters because AI agents consume more than model quality alone. They need tool integrations, execution frameworks, memory management, safety filters, and monitoring. If any one layer underperforms, the user sees the whole feature as broken. Pulling back a consumer-facing agent can therefore be a rational choice even while investment continues behind the scenes.

What to watch next

The first signal to watch is whether ByteDance, Alibaba, Doubao, or Qwen publish official product updates confirming feature removals or rebranding. App release notes, web product pages, and help center documentation will be more reliable than secondary headlines.

Second, watch whether the reported discontinuation affects only front-end agent experiences while model APIs or developer tooling continue to expand. That would suggest a packaging change, not a technical retreat.

Third, pay attention to where these companies redirect emphasis. If future messaging shifts from AI agents to search, reasoning, coding assistant workflows, or enterprise AI tools, that would support the view that broad consumer agents are being deprioritized in favor of narrower, more measurable products.

Finally, monitor whether peers make similar moves. If other major assistants also simplify or remove agent-style features, the market may be entering a phase where reliability and cost discipline matter more than autonomous-product branding.

Creati.ai perspective

Even with limited evidence, this report fits an increasingly visible market truth: AI agents are easy to announce and hard to operationalize. The real dividing line is no longer who can demo an agent, but who can make one dependable enough that users come back after the novelty wears off.

For AI builders, the likely lesson from any Doubao or Qwen pullback is to narrow scope, strengthen observability, and define success at the workflow level. For enterprise AI teams, the message is similar. Buy outcomes, not labels. If a product marketed as AI agents cannot show reliability, cost control, and auditability, a simpler form of workplace automation may deliver more value.