BossAgents · Intelligent Agent Performance Platform

Intelligent Agent Performance Platform · Detailed Design Plan v1 (2026-08-29) — To make users feel "Wow, this digital employee is amazing!" after watching a performance, immediately wanting to own one and place an order. The core is not to build another video platform, but to turn performances into hands-on sales tools—digital employees work online in real time, with users inputting data, seeing results instantly, calculating value on the spot, and placing orders immediately.

| Metric | Value | Description |

|------|------|------|

| 40 seconds | Full "wow experience" duration per performance | Hook viewers at 3 seconds, trigger order action at 40 seconds |

| 70+ | Real digital employees working online (ylxt.chat) | Roles: procurement/production/quality/safety/documentation/content/customer service, etc. |

| 8 major | Manufacturing pain point showrooms | Scenario-based engagement; customers enter via their most critical pain points |

| 48h | Enterprise-customized performance delivery cycle | Demo uses customer's own data |

Table of Contents: 01 Conclusion · 02 Why Now · 03 Our Confidence · 04 Core Loop · 05-12 Eight Breakthroughs · 13 Performance Format Matrix · 14 Script Design · 15 Technical Architecture · 16 Commercial Model · 17 Relationship with Official Version · 18 Cold Start Roadmap · 19 Risks and Countermeasures · 20 Data Caliber and Sources

01 Conclusion: Understand in One Sentence

Positioning: The "Douyin + Taobao" of the agent world—watch performances → get interested → directly purchase agents. BossAgents' 70+ real digital employees perform first; users watch, feel excited, want to own, and immediately order the official version.

The platform's sole goal: Turn "wow after watching" into "I want a set—how much?—I'll order." All designs focus on this conversion chain, prioritizing "watch → want to buy" over vanity metrics.

Eight Breakthroughs Overview

Breakthrough designs within the existing "Intelligent Agent Performance Platform" framework (details in corresponding sections):

| # | Breakthrough | One-Sentence Explanation | Core Problem Solved |

|---|------|-----------|---------------|

| B1 | Performance as Trial | Performance ≠ promo video; digital employees work live—users input data, see real-time results | Verifies actual capability, not superficial resemblance |

| B2 | Value Calculator | Shows "worth" before price—input factory parameters to instantly calculate person-day savings (¥10,000s) and payback period | Transforms "¥50k purchase" into "3-month ROI" |

| B3 | Manufacturing Showroom | Not abstract listings; 8 major pain-point scenarios let customers explore solutions for their toughest challenges | Ensures "fits my business needs" perception |

| B4 | Live Real-Data Performance | Each performance runs real interfaces, generates desensitized "work record cards," strictly separating demo data | Builds trust via "truly working" proof |

| B5 | One-Click Inquiry/Order Path | Five conversion exits: free trial → quote simulation → book demo → order → add WeChat; auto-generates quotes | Captures "I want one" intent without leakage |

| B6 | Performance Propagation Flywheel | Auto-generates shareable links/short videos; sharing = free lead gen; performance rankings drive virality | Turns every viewer into a salesperson |

| B7 | Enterprise-Custom Performances | Prospects submit scenarios → receive custom performance in 48h (using their data) → becomes sales tool | Filters genuine demand via high-quality leads |

| B8 | Open MCP Ecosystem | Self-run 70+ employees prototype → open MCP protocol for third-party agents → establish "Agent Performance MCP Standard" | Shifts from selling products to setting standards |

02 Why Now: Market Gap Confirmed

Giants have validated the space, but no one serves SMEs.

Industrial AI agents are now a real market: after Bosch's self-developed Shopfloor Agent ran successfully in its own factories, it was offered for sale externally at the end of 2025, resolving equipment downtime recovery 3-5x faster than traditional methods (Bosch official website); Black Lake Technology completed a Series D round of nearly RMB 1 billion, with annual revenue growth exceeding 60%, is profitable, and serves nearly 40,000 factories (Tencent News/Forbes); Gech Dongzhi's 'Octopus Brain' has deployed 200+ industrial AI projects (Gechuang Dongzhi official website).

But their common traits are heavyweight, expensive, and slow—large customers, large projects, large teams. China has over 60 million SMEs (China News Service/MIIT Minister Li Lecheng), and they still cannot afford industrial AI agents. That is BossAgents' opportunity: starting at RMB 50,000, one-time purchase, data never leaves the factory.

The “Agent Performance” category is empty.

There is no “Agent Performance Platform” category in the market; whoever defines it first sets the standard (market research observation):

| Product | Positioning | Difference from Us |

|------|------|-------------|

| Baidu Yijing | Digital human live-streaming/video platform | Shows “digital humans that look human,” not “agents that can do work” |

| AiTucao AI Agent Theater | Users input a topic, and AI generates comic drama/talk shows | Entertainment-oriented, passive viewing, no agent purchase involved—but it is the engine foundation we have already built |

| ReUp | AI project showcase and sales platform | Creators upload project showcases, not live agent work performances |

| AI Puppet Theater | Multi-agent puppet show (Hugging Face experiment) | Open-source experimental project, no closed-loop business model |

Conclusion: “Agent Performance” = agents working live, capability as the product, and watching the performance is the seeding. This category is currently blank. We have real digital employees, a performance engine, and manufacturing assets—we are the only ones who can establish the category immediately.

03 Our confidence: assets others cannot copy

The breakthrough is not designing from scratch; it is turning existing real assets into sales tools. The following are assets unique to BossAgents that competitors cannot catch up with in the short term:

• Real digital employees on the job: 70+ digital employees have been deployed on the public network at ylxt.chat and are doing real work, covering procurement, production, quality inspection, safety, documentation, content, customer service, and other positions. Real interfaces are online and runnable.
• 15 years of industrial knowledge foundation: 15 years of PLM accumulation, 100,000+ industrial object models, and million-level relationship topology. This is not empty talk—it is real capability backed by industry know-how.
• The performance engine is already implemented: AiTuCao has successfully run the "Comic Drama Workshop" and "Talk Show Workshop," with 60 performance scripts and 68 video scripts accumulated, ready to be directly upgraded to the engine layer of the performance platform.
• Scenario show-format specification: "Intelligent Agent Performance · Scenario Show-format Transformation Implementation Specification v2" + procurement collaboration show-format transformation have broken real business scenarios into performable, reusable show templates.
• Cost control capability: the rule engine cuts model call volume to 10%, allowing performances to run at high frequency, free, and at scale without fear of burning money—this is the confidence to afford traffic burn during cold start.
• Real competition endorsement: all six major competitions reached the finals, gaining validation through real competition. External data red line: only real work records + real data + iteration process are shown.

04 Core closed loop: Watch the show → Feel thrilled → Place an order

The entire platform is a five-step conversion funnel, and each step has an exit designed to move users from attraction to action:

① 发现（公域引流：WeChat/短视频/Share）
   → ② 看表演（数字员工现场实时干活）
   → ③ 被种草（"这个数字员工太牛了"）
   → ④ 询价下单（试算/报价/预约/下单）
   → ⑤ 拥有使用（私有化部署·数字员工上岗）

五步转化漏斗：每一步都有明确出口，表演平台负责 ①-④，正式版承接 ⑤。

Key design of each funnel stage

05 Breakthrough 1: Performance as Trial · Engineering Pleasure Points

Breakthrough B1 Core Concept: The performance isn't a promo video—it's the digital employee working live. Audiences don't "watch a clip" but "personally direct a digital employee to work." This fundamental difference from all "digital human/promo videos" is the strongest proof of actual utility.

"Pleasure" Origins: Seven Sources

• Instant feedback: Input-to-output flow, streaming typewriter effect, zero wait/circle-spinning—users feel "it's genuinely working."
• Control: Users personally input business data (material names, product names, customer issues) and see the digital employee execute—"I commanded it."
• Contrast: Side-by-side display of "Manual: 2 hours" vs. "Digital employee: 30 seconds"—time savings visualized.
• Tangible results: Beautiful visual outputs (price comparisons, reports, solutions, schedules) generated "bang!" before your eyes.
• Value demonstration: On-the-spot calculator showing "This system saves you X million/year."
• Scarcity: Exclusive customization, limited demo slots, 48-hour custom performances—creating "I need this now" urgency.
• Relatability: Demo scenarios mirror the user's factory routine—"Your company can achieve this too."

40-Second Pleasure Points Timeline (Standard performance rhythm)

Performance delivers 4 pleasure peaks within 40 seconds: Hook at 3s, force order action at 40s:

| Time | Action | Pleasure Point | UI Element |

|------|------|------|---------|

| 0-3s | Hook opening | One-line capability preview: "Compares 3 suppliers in 15s" | Digital employee persona + large-font capability tagline |

| 3-10s | User hands-on | User personally inputs business parameters | Input field + "Let me try" button |

| 10-25s | Real-time work | Streaming output visible—watch it work like live broadcast | Typewriter effect / task cards lighting up sequentially |

| 25-35 sec | Results burst out | Full results (tables/reports/plans) pop up with visual impact | Comparison table + highlighted "Save X hours" |

| 35-40 sec | Peak payoff + CTA | "Your company can do this too" + calculator entry | "Calculate how much you can save" button (enters Breakthrough 2) |

06 Breakthrough 2: Value Calculator

Breakthrough B2: Before users ask "how much does it cost," first let them see "how much it's worth." The calculator translates the demo's "wow" into money: enter your own factory's parameters, and it instantly calculates how many person-days this set of digital employees saves you per year, how many 10,000 yuan that equals, and how many months to break even. "50,000 one-time buyout" becomes "3 months to break even" in the user's mind.

Interaction design: 3 inputs → 3 outputs

Inputs (3 parameters) — by role:

• Procurement: monthly purchase order volume / supplier count / average inquiry time
• QC: daily inspection batches / time per batch
• Documentation: monthly document volume / time per document
• Customer service: daily inquiry volume / average response time

Outputs (3 numbers):

• Save X person-days per year
• Equivalent to Y × 10,000 yuan (converted based on person-month cost)
• Buyout price Z × 10,000 yuan → payback period N months

CTA: "Let the procurement digital employee calculate for you" → generate a Digital Benefit Calculation Sheet with the company name → one-click book a dedicated demo / direct inquiry.

Example (procurement assistant calculation): A mid-sized factory has 400 purchase orders per month, each requiring about 2 hours of manual inquiry and price comparison → annual investment of about 9,600 person-hours ≈ 4.8 person-years. The digital employee cuts a single inquiry to 15 minutes → saves about 4 person-years per year. At an estimated average annual cost of 120,000 per person, that's about 480,000/year. A 50,000 one-time buyout → payback in about 1.3 months.

07 Breakthrough 3: Manufacturing Showcase

Breakthrough B3 is not an abstract listing of 70+ digital employees, but 8 major manufacturing pain-point showrooms. Each showroom = one digital employee dedicated to solving one specific pain point. Customers don't care how many employees you have; they only care about "the most painful issue in my factory—can you solve it?"

8 Showrooms: Customers enter based on their most critical pain points, with corresponding digital employees providing on-site solutions

| Showroom | Pain Point | Digital Employee | Live Demo Content |

|--------|------|---------------|-------------|

| Procurement Comparison | Time-consuming quoting, comparison, and delivery follow-up | Procurement Assistant | Input material → Generate 3-supplier comparison sheet (price/delivery/payment terms) in 15 seconds |

| Downtime Recovery | Slow root cause identification after equipment stops | Equipment Management Assistant | Input failure symptom → Provide troubleshooting steps + historical solutions for similar faults in seconds |

| Production Scheduling | Reliance on veteran experience for scheduling | Production Planning Assistant | Input orders + capacity → Generate suggested Gantt chart schedule |

| Quality Inspection | Manual inspection records, difficult traceability | Quality Assistant | Input inspection items → Generate inspection records + automatic severity-based alerting for anomalies |

| Safety Patrol | Unreviewed patrol checklists | Safety Assistant | Input patrol results → Generate hazard list + rectification tasks |

| Document & Report | Manual creation of quotes/reports | Document Assistant | Input company + product → Generate complete quote in 15 seconds |

| Content Marketing | Struggling to create professional product descriptions | Content Marketing Assistant | Input product name → Generate copy + matching images in 30 seconds |

| Customer Service | Slow response, no after-hours support | Customer Service Assistant | Input customer query → Generate professional response script in seconds (24/7) |

Name showroom homepages by 'Pain Point' (e.g., 'Exhausting Procurement', 'Excessive Downtime')—not by 'Employee Name'—so customers find their pain point within 3 seconds.

08 Breakthrough 4: Real-Data Live Performance

Breakthrough B4: Performances use real interfaces running live, not pre-recorded demos. Each demo is driven by real ylxt.chat / CMS interfaces—what users see is genuinely executing. Every performance generates a desensitized 'Real Work Record Card' proving actual execution.

Real Work Record Card (Auto-generated per performance)

• Performance timestamp, called Digital Employee ID & role
• Actual executed step chain (Rule engine + Model call path)
• Task type processed & deliverables generated
• Key snippets from real interface response (sensitive fields masked)

Why it breaks trust barriers: Upholds the data principle 'Only show real work records externally'; Strictly distinguishes from 'demo data'—pages must be labeled 'Real Interface Data' or 'Demo Data'; Answers the client question 'Is this pre-recorded?'—the record card provides proof; Optionally enables 'Performance Replay': Show clients the full step-by-step execution chain.

Red Line: External data must be real—only real-time interface outputs can be labeled 'Real Interface Data', all else must be 'Demo Data'. This red line is non-negotiable.

09 Breakthrough 5: One-Click Inquiry-to-Order Conversion Path

Breakthrough B5: Seamlessly captures 'I want this system' intent, leaving no lead behind. Every post-performance action leads to inquiry/order, with 5 paths catering to different customer decision types.

Five conversion paths, routing by customer decision type

Auto-generated Quote (Core Materials)

On the quote page, users select a version (Single Scenario Package / Complete Platform / Source Code Privatization) → The system auto-generates a company-named quote based on enterprise size, number of positions, and deployment method. The quote includes: Product list, price, delivery cycle, deployment method (data on-premise explanation), ROI calculation, and after-sales service terms.

10 Breakthrough Six: Performance Propagation Flywheel

Breakthrough B6: Every performance is customer acquisition material; every viewer is a salesperson. Each link shared by users after watching a performance brings us free customer acquisition.

Flywheel Design:

• One-click sharing material generation: Each performance auto-generates a unique link + 30-second short video (automatically edited from "User input → Digital employee working → Results output").
• Enterprises share with peers after watching: Manufacturing bosses watch procurement price-comparison performances and share in industry groups—this is word-of-mouth fission.
• Performance leaderboard: Ranked by "views + shares + trial calculations" to trigger herd mentality: "This performance is trending—I should watch it too."
• Share backflow tracking: Each shared link carries a source tag to identify lead sources and optimize ad placement.

Example: The "Procurement Collaboration" performance is naturally shareable among executives: "Watch this procurement digital employee negotiate with three suppliers." Each share = one precise manufacturing customer acquisition.

11 Breakthrough Seven: Enterprise Custom Performances

Breakthrough B7: Opportunity engine—Potential customers submit their business scenarios; we deliver custom performances in 48 hours. Using their own data for demos shown to executives/board members, this yields the highest-quality leads and filters genuine demand.

Process: Customer submits scenario → Platform understands requirements → 48h custom performance delivery → Customer uses for internal demo → Orders official version

Customer submits:

• A one-sentence business scenario (e.g., "Our factory procures 300 materials monthly—price inquiries are exhausting").
• Optional: Upload a real document/table (quote, quality inspection sheet, schedule).
• Contact information.

We deliver (within 48h):

• A custom performance using customer data (anonymized).
• A scenario solution page (shareable with decision-makers).
• Quote and ROI calculation.

Commercialization (decided 08-29: Free initially): No upfront charge—custom performances are free to showcase Left & Right Arm AI capabilities, build trust, and impress clients. Pricing discussed only for genuine demand, with quotes tied to opportunity progression. Free ≠ no screening: "More specific scenarios get priority scheduling" replaces paywalls; templated processes control delivery costs.

12 Breakthrough Eight: Open MCP Ecosystem

Breakthrough B8: From "selling our own employees" to "setting industry standards." First, popularize performances with 70+ in-house digital employees; then open MCP protocol for third-party agents to join performances, seizing authority over "Intelligent Agent Performance MCP Standards."

Three stages from self-operation to platform:

Key MCP access specifications:

• Registration: Register agent basics, capabilities, and pricing via MCP protocol.
• Performance format definition: Agents define their own input type, output type, and duration.
• Performance interface: Standardized "Input → Process → Output" performance call API.
• Performance records: Each performance is recordable, replayable, shareable, and integrated into real-work documentation.

13 Performance Format Matrix

Expanding on Ai Tucao's existing comic drama and talk show capabilities, 8 performance formats cover diverse decision-maker types:

| Performance Form | Description | Applicable Staff | Engine Foundation | Conversion Focus |

|---------|------|---------|---------|---------|

| Live Work | User inputs requirements, staff executes and displays results in real-time | Purchasing/Customer Service/Documentation/Quality Inspection, etc. | Real Interface | Directly proves "can perform tasks" |

| Comparison Demo | Side-by-side "Manual Work" vs. "Digital Staff Work" | Any staff | Extensible | Directly proves time savings |

| Data Visualization | Real-time visualization of staff processing data | Data analysis roles | Extensible | Demonstrates "intelligence" |

| Comic Drama | 4-panel comic showing end-to-end solution for a business scenario | Any staff | Ai Tucao implemented | Shows process closure |

| Short Video | 30-60 second video performing tasks in real scenarios | Operators/Equipment roles | Extensible | Highest realism |

| Talk Show | 2-3 minute audio explaining capabilities humorously | Any staff | Ai Tucao implemented | Lowers cognitive barrier |

| Roast | Input topic, staff delivers sharp critique | General roles | Ai Tucao implemented | Creates viral buzz |

| Podcast | Two digital staff discuss industry issues | Industry expert roles | Extensible | Builds professional trust |

Live Work and Comparison Demo are the "primary order drivers", while others serve as "traffic and trust entry points". Initial launch will prioritize Live Work + Comic Drama + Talk Show (engines ready).

14 Script Design: Four-Step Structure and Initial Staff

Each performance follows a fixed four-step structure (grasp "what it is, what it does, and its value to me" in 30 seconds):

① 开场（5 秒）自我介绍 → ② 演示（15-20 秒）现场干活 → ③ 结果展示（5-10 秒）成果爆出 → ④ 行动召唤（5 秒）引导下单

七个首发数字员工表演剧本（可直接上线）

| 员工 | 开场 | 演示 | 转化话术 |

|------|------|------|---------|

| Content Marketing Assistant | "30 seconds to help write a product introduction" | Input product name → Generates article word-by-word + images | "Want similar copy for your product? Click to try" |

| Procurement Assistant | "15 seconds to compare 3 suppliers" | Input material name → Generates comparison table (supplier/price/delivery) | "Want procurement this efficient? Click to learn" |

| Customer Service Assistant | "Instantly reply to customer inquiries, 7×24 online" | Input customer question → Generates response scripts in seconds | "Want instant replies for your customers? Click to try" |

| Documentation Assistant | "15 seconds to generate a quotation" | Input company + product → Generates quotation preview | "Want automated document generation? Click to learn" |

| Product Assistant | "20 seconds to generate full product description" | Input product name → Generates description + parameters + selling points | "Need similar product listings? Click to try" |

| Mall Assistant | "5 seconds to list products with one click" | Input product name → Shows successful listing preview | "Want automated mall listings? Click to learn" |

| Health Assistant | "5 seconds to assess your physique" | Photo/Input → Generates physique report + adjustment plan | "Need employee health management? Click to try" |

The initial 7 staff cover seven high-frequency demand categories: "Content/Procurement/Customer Service/Documentation/Product/Mall/Health". All are lightweight adaptations of existing capabilities, deployable in 1-2 weeks.

15 Technical Architecture: Based on Real Assets

Entry strategy (finalized 08-29): Primary entry: Mini Program (fast reach, short sharing chain, natural WeChat user acquisition); Secondary entry: Web version—stronger features, including backend management, operational configuration, data dashboards, quotation sheets, and lead management.

Four-layer architecture, fully leveraging existing BossAgents assets without new heavy platforms:

| Layer | Content |

|----|------|

| Frontend Display Layer | Mini Program "Left & Right Arm Agents" + Web ylxt.chat · Performance Square / Player / Detail Page / Inquiry Page / Showroom |

| Performance Engine Layer | Comic Drama Engine / Talk Show Engine (Ai Tucao implemented) + Live Work Engine / Comparison Demo Engine / Calculator / Record Card Generator |

| Agent Access Layer | MCP protocol compatible · Registration/Authentication/Review · Demo template marketplace · Rule engine (model calls reduced to 10%) |

| Agent and Data Layer | BossAgents 70+ digital employees + 15-year PLM / 100k+ objects / million-level topology + ylxt.chat & CMS real interfaces + third-party MCP agents |

The four-layer architecture fully reuses existing assets: AiTucao engine, 70+ employees, real interfaces, and the PLM data foundation.

Key Implementation Points

• Demo means calling real interfaces: the live work engine directly invokes ylxt.chat / CMS digital employee runtime interfaces with streaming output.
• Automatic record card generation: each demo logs a running record (desensitized), used for real work record cards and replay.
• Low-cost operation: rule engine as fallback + model calls reduced to 10%, enabling high-frequency free demos.
• Data labeling: outputs from real interfaces are labeled 'real interface data'; the rest are labeled 'demo data'.

16 Commercialization Model and Pricing

Phased strategy (finalized 08-29): everything is free in the early stage—free demo viewing, free custom demos—with the goal of promoting the ZuoBangYouBi agents, building trust, and getting customers to order the official version. The four-tier revenue model below is the full commercialization blueprint after the process is proven; in the execution phase, the early stage is all free, and pricing for custom demos will be discussed only when real demand arises.

The performance platform itself is free to watch (a traffic-driving product), with full commercialization achieved through four monetization layers:

| Revenue Source | Pricing | Trigger Point |

|---------|------|--------|

| ① Digital employee product conversion (main revenue) | L1 single-scenario package 3-8 万 (entry 5 万) → L2 complete platform 8-30 万 (desktop 10 万 / enterprise edition 20 万 tier) → L3 source code/private deployment 50 万 | Watch performance → get intrigued → trial calculation/booking/order |

| ② Enterprise display payment | Single display 500-2000 元; monthly package 5000-20000 元/month; annual brand cooperation 5-20 万/year | Enterprises want their intelligent agents to gain recommended slots and traffic |

| ③ Custom performance service | Free in early stage (negotiable if needed) | Enterprise submits scenario, dedicated performance within 48 hours; build trust for free first, discuss price when real demand arises |

| ④ Ecosystem transaction commission | After third-party agents integrate, 10-20% commission on transactions (MCP ecosystem stage) | Commission based on closed deals after platform opens |

Logic: The performance platform is the "showroom," converting traffic into leads; BossAgents official version is the "factory," converting leads into delivery and renewals. The showroom burns money, the factory collects and profits.

17 Relationship with BossAgents Official Version

| Dimension | Agent performance platform (lead generation) | BossAgents official version (conversion) |

|------|----------------------|--------------------------|

| Positioning | Watch performances, experience, get interested | Formal use, work, management |

| What users see | Zuobang Youbi brand | Customer's own brand |

| Data | Demo data / sample data | Real enterprise business data |

| Pricing | Free to watch | Starting at 50,000 (buyout) |

| Features | Core employee performances + trial | Full 70+ digital employees + multi-platform integration |

| Entry | Mini program "Zuobang Youbi Agent" / ylxt.chat | Customer-specific mini program / web / desktop version |

In one sentence: Users watch product performances in the showroom, then walk into the factory to place formal orders. Both share the same set of digital employees, PLM data foundation, and performance engine, with marginal cost near zero.

18 Cold Start Roadmap

Entry rhythm: The mini program goes live first to run performances and conversions, while the web version simultaneously handles backend management and operations (performance configuration, data dashboard, lead pool, quotation).

• Phase 0 · Already in place (available today): 70+ digital employees working online (ylxt.chat), Ai Tucao comic drama/talk show engine, 60 performance scripts, 68 video scripts, scenario programmatic specification v2. All assets are in place; what's missing is a product that strings them into a "watch → buy" journey.
• Phase 1 · Weeks 1-2: Launch the first 7 employee showcases: seven showcase scripts go live for content marketing/procurement/customer service/documentation/product/mall/health (the live-work engine connects to real interfaces), and build the Showcase Plaza + player + detail pages + inquiry entry.
• Phase 2 · Weeks 3-4: Showrooms + value calculator go live: scenario-based revamp of 8 major manufacturing showrooms, value calculator + automatic quotation sheet go live, and real work record cards start being generated.
• Phase 3 · Month 2: Business opportunity engine opens: enterprise-exclusive customized showcase service goes live (48h delivery, free in the early stage, pricing discussed when real demand exists), and the growth flywheel + leaderboard go live.
• Phase 4 · Month 3: Open MCP ecosystem: third-party agents can be integrated into showcases, publish the "Agent Showcase MCP Specification", and start transaction commission.

19 Risks and Countermeasures

20 Data Definitions and Source Notes

All external data in this plan follows the unified BossAgents definitions and has been verified online with sources retained:

• Number of SMEs: over 60 million (MIIT definition) — China News Service
• Bosch Shopfloor Agent: downtime recovery 3-5x faster than traditional methods — Bosch official website, Virtasant
• BlackLake Technology: Series D financing of nearly RMB 1 billion, revenue growth 60%+, serving nearly 40,000 factories — Tencent News · Forbes
• Gechuang Dongzhi "Octopus Brain": 200+ industrial AI projects delivered — Gechuang Dongzhi official website
• Number of digital employees: 70+ on duty (unified external definition, subject to the CMS demo call document 2026-08-25)
• Competitions: all six major competitions reached the finals (historical definition; whether to update to seven competitions is pending the boss's decision)
• Market gap comparison (Baidu Yijing / Ai Tucao / ReUp / AI Puppet Theater) is an industry research observation, with no public link source, and is not cited as a definitive conclusion. The development/launch status of the showcase platform is subject to the actual schedule.

Decided on 08-29: Entry strategy — first mini-program, second web page (more powerful, also supports backend management); early free — showcases and customized showcases are all free in the early stage, with the goal of promoting the ZuoBang YouBi agent and building trust; the "7-day free trial" account is open by default; customized showcases are priced when real demand exists.

Pending decision: ① Number of competitions (six or seven).

Agent Showcase Platform · Detailed Design Plan v1 ｜ BossAgents Project Team ｜ 2026-08-29

Supporting plans: Agent Showcase · Enterprise Demo Business Plan v1, Enterprise Intelligent Transformation Training and Delivery Plan v2