← Sticker Wall AGENT · DETAIL

Providing quality audits for small and medium-sized R&D teams using Dojo-like AI testing agents, charging 20,000 RMB per project

Workflow: After taking an order, first align with the client on code repository permissions and requirements documents. Use the AI

AGENT

Key Fields

FIELD STAMPS

IndustryFintech

RegionChina

ScaleSME

ChannelOnline

🔧 Workflow

After taking an order, first align with the client on code repository permissions and requirements documents. Use the AI testing agent to automatically generate test cases and run regressions based on natural language intent. The agent outputs failed items and preliminary root cause judgments (such as API field changes or rendering timing issues). Humans act as referees to thoroughly review false positives and false negatives one by one, and then output a structured defect report, root cause localization explanation, and repair priority checklist. At the conclusion, archive the assertion templates and skill configurations into your own skill library so that the next similar project can directly reuse them. A single delivery cycle takes about one to two weeks, becoming faster with each subsequent engagement.

🛠 Setup Requirements

Requires a strong foundation in test engineering or backend development, capable of reading logs, call chains, and API documentation—this is the baseline capability for the human review stage. On the tool side, set up the large language model API call environment, Git repository access permissions, and a browser automation execution environment like Playwright, while familiarizing yourself with the workflows of skill and evaluation platforms like Tessl. The first month is mainly spent polishing prompts, assertion templates, and root cause report formats on your own or friends' real projects to produce the first externally showcaseable case study, after which paid orders are officially accepted.

🧰 Toolchain

🔧 Tessl skill and evaluation platform
🔧 LLM API (Claude or equivalent tier)
🔧 Playwright browser automation
🔧 GitHub code repository
🔧 Feishu Docs (for delivery reports)

💰 Revenue

1. Testing capability assessment and test case quality audit for small/medium R&D teams (primary revenue): Clients pay a service fee per project, priced according to code repository scale and test case volume. 10,000 to 30,000 RMB per project x 1 to 2 projects per month = monthly revenue of approximately 20,000 RMB, accounting for roughly 100% of monthly revenue (reverse-engineered from card data; sourced from case descriptions without independent verification). 2. Additional orders via skill template reuse: As templates mature, marginal costs decrease, and during the stable period, monthly revenue can push toward 30,000 RMB through project-based or usage-based renewals (renewal volume is unverified, also originating from case descriptions, and the exact share of total revenue is unspecified). 3. Vertical testing skill package subscription: Benchmarked against the Tessl registry having launched over 3,000 searchable skills in 2026 (figure cited from company public releases), individuals sell AI testing skill packages to testing teams on a subscription model; selling price is undisclosed, the number of purchasing teams is unknown, and the market share is unquantified. 4. Opportunity item - Platform referrals and ecosystem rebates: Earning channel rebates by referring small and medium teams to Tessl-style platform subscriptions; rebate ratios are unclosed, and the revenue volume for this segment remains to be observed.

💸 Cost

LLM API calls cost approximately 500 to 1,500 RMB per month, fluctuating based on project test case volume; browser automation runtime environment server costs about 200 RMB per month; total monthly costs are kept under 2,000 RMB, with the first order covering the full cost.

⏱ Time Investment

During the order fulfillment period, commit 3 to 4 hours daily, focused on three stages: test case review, failure root cause review, and report drafting; once templates mature, human labor time per project can be compressed by 30%.

🚀 Getting Started

Step 1: Skip taking orders initially. Find one or two small teams with real code repositories in open-source communities or tech groups, conduct a free AI regression test run, and output a root cause localization report. Step 2: Desensitize the report to create a case study template, and publish review articles on Juejin, developer communities, and Xiaohongshu/Jike. Step 3: List the paid audit service pricing, using the first two pieces of content to drive traffic; the first paid client typically comes from the article readers.

🔑 Keys to Success

✅ Humans acting as referees to filter out AI false positives is core to delivery credibility; every defect in the report must have a review record
✅ Transform assertion templates, root cause classifications, and report formats into reusable skill assets to build compound growth
✅ Focus on a single scenario such as API regression or frontend regression, avoiding a broad and bloated full-stack testing approach
✅ Case studies first: Public review articles earned in exchange for free test runs are the lowest-cost customer acquisition channel

⚠️ 风险

⚠️ Insufficient coverage of AI-generated test cases leading to missed bugs; online failures missed might be traced back to the consultant, requiring human backup and clear liability boundaries
⚠️ Client code involves trade secrets; inadequate NDAs and repository permission management pose legal risks
⚠️ As major tech giants and enterprise testing platforms lower prices and move downstream, large clients tend to build capabilities in-house, making repeat purchase rates unstable

📌 Real Cases

📌 Multiple 2026 AI testing tool review articles (Alibaba Cloud Developer Community, CSDN Annual Review) point out that intent-driven AI testing agents are replacing high-maintenance Selenium scripts, with solutions like Harness, Autonoma, and qpilot already adopted by teams for natural language generation and execution of end-to-end tests
📌 The Tessl official website shows its platform enables teams to create, test, distribute, and improve skills for coding agents, supporting the stable delivery of AI systems in real code repositories using structured, versioned contexts, which validates the feasibility of skill assetization
📌 Articles comparing automated test case generation tools in 2026 emphasize that the most efficient teams employ a human-AI collaboration model where test engineers master prompt engineering plus domain modeling plus failure mode analysis—precisely the capability model required for the human review stage of this entry

Sources