Zhedie makes the case for specialist AI teams

Agent-based product work can now turn sufficient context and several rounds of conversation into prototypes and PRDs, shifting the scarce work away from first drafts. The harder question is who supplies the business knowledge, encodes sound methods, and takes responsibility for the result.

According to woshipm, AI is compressing copywriting drafts, visual exploration, storyboards, animation previews, basic editing, and post-production iteration. Yet the same source warns that an AI system can return incomplete work when it lacks a company's specific business context. Its author argues that product teams can transfer requirement-analysis methods into AI skills an Agent can invoke; after enough information is gathered, the system can form requirement-analysis conclusions and propose an MVP. That pattern explains why solo operators can remove handoffs in routine work, while demanding output may make specialist process design and accountable judgment more valuable.

Draft generation moves the bottleneck to context

AI can compress the first pass of routine production. According to woshipm, that includes copywriting drafts, visual exploration, storyboards, animation previews, multi-version materials, basic editing, and post-production iteration. Product work has a similar pattern: with sufficient context and several rounds of conversation, AI can draw prototypes and write PRDs effectively. Once the relevant information has been gathered, it can also generate requirement-analysis conclusions and an MVP proposal.

The scarce input is increasingly the context that makes those outputs usable.

A model does not automatically know a company's business, woshipm notes, so its result may be incomplete even when the draft looks polished. Earlier requirement background analysis, prototypes, and documents can let AI produce an operation manual with far less manual drafting. But those source materials embody decisions that somebody must still make, check, and update. The bottleneck shifts from typing an artefact to defining the conditions under which that artefact is correct.

For product managers, woshipm describes a progression from using AI to find information, to placing it inside a workflow, to making AI the product itself. The consequential step is not merely asking for a better prompt. The author argues that requirement-analysis experience and product methods should be transferred into an Agent as reusable AI skills. That turns tacit judgment into an invoked process, while leaving people responsible for the business assumptions behind it.

This also argues against rebuilding every existing system as AI-native immediately. woshipm recommends introducing AI first where it has been validated to improve efficiency clearly. Draft generation can move quickly; encoded methods and human review determine whether it can move safely.

A solo operator can remove costly handoffs

An OPC, in woshipm's definition, is a senior individual or very small team able to take work from problem understanding and strategy through creative expression, content production, and final delivery. That scope fits routine assignments where a client needs a single accountable operator rather than a chain of specialist handoffs. The woshipm author expects lighter arrangements for social-media communication, poster production, lightweight video, and platform operations.

The advantage is ownership of the whole routine, not a claim that AI can replace judgment.

A practitioner should package that ownership as a defined service. State the business objective. State the source materials, audience constraints, and approval owner. Then specify intermediate reviews, such as a strategy check before production and a draft check before release. Delivery should name the final format, the person responsible for corrections, and the condition for sign-off. This makes an AI-assisted workflow legible to a buyer without reducing it to a vague promise about prompting.

The evidence for capacity is encouraging but limited. A 2025 marketing experiment with 2,310 participants found human-AI teams raised per-person productivity by about 60%, according to woshipm, while overall advertising performance remained similar to traditional human teams. Faster output therefore needs accountable review. woshipm argues that fewer internal reports, coordination requirements, scheduling processes, and confirmation rounds can let an OPC respond more quickly than a large supplier.

Hiring signals also reward demonstrable workflow skill. According to sspai, data product manager roles almost all asked about relevant AI skills and agents used at work, with some requiring candidates to build AI workflows independently. A repeatable service definition turns that expectation into something a solo operator can show and deliver.

How AI changes work across advertising, content production, and product management

Premium production brings specialist teams back

Zhedie's operation shows why AI-enabled production does not always shrink to a single operator. Its main business spans mid- to long-form series, animated films, and advertising, made through a combination of established production methods and AI. That mix demands more than generated drafts. Directors, producers, artists, and technical staff must keep a project's visual language, production decisions, and audience expectations aligned across a sustained piece of work.

AI may widen the gap between adequate output and premium output.

According to 36kr, Zhedie has built agents for script breakdown, prompt generation, and character design, while a dedicated technical team supports AI game adaptation and internal film and television workflows. Sam said directors, producers, and artists are trying to distill their own methods into Skills, preserving more time for thinking. The company's founding team sees AI as making 60-point content easier, but says 100-point work needs people fluent in AI technology and traditional aesthetics.

Scale also changes what coordination means. Zhedie has production teams with capacity at the level of 100 people in multiple cities, and 36kr reports monthly capacity of 5,000 minutes of high-quality content, equivalent to about 40 films. It has partnerships with multiple platforms to adapt platform IP through AI hybrid production. As woshipm's author argues, large brand campaigns, crisis public relations, cross-regional coordination, celebrity resources, offline events, and high-budget film and television production still call for large organizations. The point is not that AI makes teams obsolete; it can make specialist teams more able to concentrate their judgment where the work carries the most risk or value.

Measure quality before calling the workflow better

A 2025 marketing experiment involving 2,310 participants found that human-AI teams raised per-person productivity by about 60%, according to woshipm. Yet their overall advertising performance resembled that of traditional human teams. More output, then, is not proof.

Better output remains unproven.

The measurement must match the work. Its real standard matters. People should retain ownership of decisions where quality determines the result. Fit also determines the result. Accountability does as well.

For product work, woshipm's author argues for gradual adoption in scenarios already validated to improve efficiency, rather than rebuilding every existing system as AI-native. A ten-year-old e-commerce system could start with automatic completion of listing information. It could add AI image polishing. It could then use AI to filter and summarize negative reviews. Each addition creates a testable comparison against the existing process.

That approach also sets sensible expectations for practitioners. The woshipm author argues that AI can sharply reduce the time product managers spend filling business-knowledge gaps, and can serve as an above-average industry baseline across fields. It is less suited to the standards expected of top experts. The bigger step is to integrate proven AI capabilities into products so a broad user base benefits, rather than treating AI only as an individual product manager's efficiency tool. Quality gains require evidence in the specific scenario, with people still answerable for the standard.

Where AI-assisted workflows fit-and where they do not

• You need routine marketing output such as copywriting drafts, visual exploration, storyboards, animation previews, multi-version materials, basic editing, or post-production iteration. Use AI to compress production work, while keeping an experienced operator responsible for strategy, creative direction, and final delivery. The evidence suggests this is the territory where a senior individual or very small team can operate across the full process.
• You are building a product requirement, prototype, PRD, or operating manual but lack complete business context. Use AI as a structured research and drafting partner, then supply company-specific information through multiple rounds of conversation and review the result. AI may provide incomplete results when it does not know the company's specific business, even though it can produce prototypes and PRDs well with sufficient context.
• You want to automate a repeatable internal process rather than merely use a chatbot for one-off questions. Convert proven product methods and requirement-analysis experience into reusable AI skills or agents. Zhedie's workflow uses agents for script breakdown, prompt generation, and character design; its directors, producers, and artists are also researching how to distill their own expertise into Skills.
• You are considering an immediate rebuild of an existing system as an AI-native product. Start with validated, bounded use cases instead. Suggested examples include automatic product-information completion for listings, AI image polishing, and AI filtering and summarization of negative reviews; the recommended approach is gradual introduction where efficiency gains are clear.
• The work involves a major brand campaign, complex media buying, crisis public relations, cross-regional coordination, celebrity resources, offline events, or high-budget film and television production. Do not assume a lean AI-enabled operator can replace a large organization. These cases still require large organizations, and premium content may depend on people who understand both AI technology and traditional aesthetics.

A portfolio should show the workflow, not confidential work

A portfolio can make AI workflow-building visible without publishing the work itself. Show a public-safe method: how you frame a requirement, what context the system needs, where human review intervenes, and what a finished deliverable must satisfy. That matters because AI can return incomplete work when it lacks a company's specific business knowledge, as woshipm notes.

The useful signal is not a polished model draft presented as a personal accomplishment. It is a reusable process that distinguishes analysis from writing, records the checks applied to each stage, and explains which decisions remain accountable to the operator. sspai recommends that separation when using AI for resume preparation, along with reviewing the reasoning and results throughout the process.

Do not reveal former employers' proprietary context to prove that you have it. Abstract the method instead.

For product candidates, woshipm argues that requirement-analysis experience should be converted into Agent-invocable skills. A portfolio can demonstrate that conversion with a sanitized requirement template, an example of the questions an Agent must answer, and a review rubric. Keep any example fictional or safely generalized; the point is to show judgment encoded into a method, not to reproduce confidential material.

This evidence addresses an increasingly direct hiring question. According to sspai, data product manager roles almost all asked candidates about relevant skills and agents used at work, while some stressed the ability to independently build AI workflows. Automated applications may reach 8-10 roles per day, yet sspai's author reports no interview progress from system-recommended opportunities and believes such submissions receive low priority. A documented workflow gives a recruiter something concrete to assess in a conversation of about 10 minutes.

Treat AI experience as a work sample, not a line on a résumé. For roles that ask about agents or independently built AI workflows, prepare a short account of the input, the method you encoded, the intermediate checks, and the decision you retained. According to sspai, reliable headhunters use conversations lasting about 10 minutes to test fit and explain why they recommend someone; make your workflow explainable in that window.

Be wary of volume as proof of progress. sspai's author reports that platforms can invite candidates to apply for 8-10 jobs per day, while automatically submitted résumés may receive low priority and did not produce an interview in the author's case. Remove handoffs where AI genuinely helps, but show where specialist review remains necessary.

For readers outside China

• Availability: The source material names Figma, WeCom, DingTalk, and Feishu, but does not say where these tools are available outside China or whether the described AI capabilities are offered internationally. It also discusses internal workflows at Zhedie rather than a publicly available product.
• Pricing: Not disclosed in sources.
• Closest Western equivalents: Figma; No Western equivalent is identified in the source material for Zhedie's internal production workflow.; No Western equivalent is identified in the source material for the AI features described for WeCom, DingTalk, and Feishu.
• Data residency: The source material does not cover data residency, hosting location, cross-border data transfer, or enterprise data-handling terms.

Sources

• sspai 与 AI 搏斗失败后重新开始找工作：经验分享与半可靠避雷指南 https://sspai.com/post/114461
• woshipm 公关、广告行业的OPC时代来了 https://woshipm.com/ai/6452759.html
• woshipm 产品经理在AI时代的三层进阶 https://woshipm.com/share/6461983.html
• 36kr 爱诗科技投资、主做精品内容，AI影视公司折叠完成数百万美元融资｜36氪融资首发 https://36kr.com/p/3954709821258889