NetEase's 'Bage Says' Offers ¥100K for 'Real Needs': A New User Co-creation Model for AI Voice Agents

NetEase's first AI-native voice agent, "NetEase Bage Says" (网易叭哥说), has launched an initiative called the "User Honorary Co-creation Program," putting up ¥100,000 to solicit product suggestions and declaring that there are no educational or professional requirements—just speak the truth. The news has sparked a modest discussion in AI product circles: why would a voice agent product spend real money to buy "real needs" from ordinary users?

An Unconventional Product Requirement Solicitation

Most AI products gather early feedback through beta groups, surveys, and community posts, with users offering suggestions largely on their own initiative. NetEase Bage Says has turned "giving suggestions" into a prize event with a ¥100,000 pool and a "no professional qualifications required" threshold—effectively putting professional product managers and ordinary users on the same starting line.

The subtext of this setup is worth pondering: for products like voice agents, the real pain points often aren't in technical specs but in the everyday usage scenarios—those details that are "hard to articulate but awkward to use." Users with professional backgrounds tend to fall into feature lists, while ordinary users are more likely to express genuine feelings.

What the "User Honorary Co-creation Program" Aims to Solve

Based on public information, the program's core is soliciting product suggestions, positioned as "honorary co-creation," meaning participants are given a co-creator identity in the product's evolution. For a newly launched AI-native voice agent, what's most lacking during the cold-start phase isn't compute power or models—it's an understanding of real usage scenarios. The ¥100,000 prize investment is essentially front-loading the cost of user research, using incentives to exchange for high-density real feedback.

This model isn't new in consumer hardware, but it's still rare in the AI voice agent category. The signal it sends is that the product team is willing to admit it hasn't figured everything out and to hand part of the definition power to users.

Product Logic of the AI Voice Agent Track

"NetEase Bage Says" is positioned as NetEase's first AI-native voice agent. The difference between a voice agent and a voice assistant lies in the word "native"—it's not about stitching together speech recognition, intent understanding, and speech synthesis, but building from the ground up with voice as the primary form of interaction.

The Real Barrier to Voice Interaction

The quality of a voice product's experience largely depends on the naturalness of conversation and scenario coverage. Text interaction is fault-tolerant and reviewable; with voice, once you say something wrong, you have to reorganize your language. Users have far less patience for voice agents than for apps—one off-target answer can lead to churn. This is why voice agents need real-world feedback even more than text agents—lab conversation samples can hardly cover what users will actually ask and how they'll ask it.

From this perspective, offering big money for suggestions isn't a marketing gimmick but a reasonable choice on the product iteration path.

"AI-native" emphasizes that a product is built with AI capabilities (especially large models) as core components from the start, rather than layering AI features onto traditional software. In the voice agent context, native means conversation management, context memory, multi-turn clarification, and other elements are driven by models rather than relying on preset rule trees and intent classifiers. The advantage of this architecture is handling open domains, ambiguous expressions, and topic shifts; the disadvantage is that behavior is harder to predict and debugging depends more on real data. Therefore, the maturity of a native voice agent often doesn't depend on single-point technical metrics but on how well the team understands the real distribution of conversations.

Cold start originally refers to the problem recommendation systems face when lacking user behavior data to make effective recommendations; in AI products, it broadly refers to the early launch phase when both user scale and real interaction data are scarce. For voice agents, the challenge of cold start is that without enough real conversations, you can't discover the model's failure modes under conditions like accents, interruptions, multiple intents, and background noise; and without these discoveries, the product experience struggles to retain users—creating a cycle. Front-loading user research and using incentives to exchange for high-density feedback is a common strategy to break this cycle.

Industry Implications of Spending ¥100K for "Truth"

User co-creation in AI products is shifting from "community operations" to "mechanism design." In the past, user feedback was free and spontaneous, passively collected by product teams; now some teams are willing to pay for structured, high-value feedback, reflecting a need to filter for feedback quality.

For ordinary users, "no professional qualifications required" lowers the barrier to participation and gives value to "speaking the truth" itself. As for how many suggestions that truly drive product evolution the ¥100,000 will ultimately bring back, it depends on whether the event mechanism can effectively filter out filler feedback. But at least, this approach of putting users in the product definition process provides an observable sample for AI product cold starts.

User co-creation refers to product teams involving users in the product definition and iteration stages, shifting from passively collecting feedback to proactively designing participation mechanisms. Early forms were mostly community posts, beta groups, and surveys, with scattered and uneven-quality feedback; advanced forms use bounties, crowd testing, and co-creator identities to structure, filter, and incentivize feedback. The core challenge is that incentives attract both high-quality suggestions and filler content, so mechanism design needs to include review criteria, deduplication, and validation steps—otherwise the prize investment may not translate into product improvements.

One-line Summary

NetEase's voice agent spends ¥100,000 to solicit user suggestions, exploring AI product co-creation mechanisms.

Paragraph Summary

NetEase's first AI-native voice agent, "NetEase Bage Says," has launched a "User Honorary Co-creation Program," investing ¥100,000 to solicit product suggestions from ordinary users with no professional qualifications required, emphasizing "speaking the truth." The article argues that the interactive nature of voice agents makes them more dependent on real-world feedback than text agents, as lab samples can hardly cover users' actual questions; offering big money is essentially front-loading user research costs to exchange incentives for high-density real feedback. This reflects AI product user co-creation shifting from community operations to mechanism design, but whether the ¥100,000 can bring back high-quality suggestions still depends on the event's feedback screening mechanism.