← Sticker Wall AGENT · DETAIL

AI Workforce Outsourcing Agent: Replicating Cognition's Human-in-the-loop Model for $50k/Month Revenue

Workflow: Receive development or maintenance tickets from corporate clients daily, breaking down requirements into sub-tasks execu

AGENT

Key Fields

FIELD STAMPS

IndustryAI / LLM

RegionChina（中国大陆）

ScaleSME

ChannelOnline

🔧 Workflow

Receive development or maintenance tickets from corporate clients daily, breaking down requirements into sub-tasks executable by Devin. The AI agent autonomously writes code, runs tests, and submits PRs. A human lead performs spot checks, fixes critical defects, and delivers the work, billing by milestone. Reusable solutions are documented as SOPs. Weekly reviews of delivery data are used to optimize quality control rules and the toolchain, creating a compounding loop of efficiency.

🛠 Setup Requirements

Requires foundational engineering skills such as git, CI/CD, and API debugging, along with setting up and running Devin sample projects. Prepare a quality control checklist and contract templates. Select 1-2 vertical industries to gather real-world requirements; cold start typically takes 2-3 weeks. If operating as a team, configure simple subcontracting settlement and client communication SOPs to ensure consistent delivery quality.

🧰 Toolchain

🔧 Devin
🔧 GitHub
🔧 Linear
🔧 Notion
🔧 Feishu Base

💰 Revenue

① Corporate Level: Cognition charges enterprise clients annual fees (primary revenue) via B2B subscriptions/seats, with an ARR of $492M. With a $26B valuation, the P/S ratio is 53x (based on public reports and third-party analysis); specific business segment breakdowns are not public. ② Replicator Level: Individuals using Devin for software development and maintenance outsourcing charge by project milestones. With an average ticket size of 30k-150k RMB per project and 3-5 concurrent projects, monthly revenue is approximately 50k RMB, potentially exceeding 100k RMB at scale. Tool costs account for 5%-10% (based on practitioner reports, unverified). ③ Value-Added: Selling monthly labor packages to recurring clients, charging for output rather than hours. Pricing and volume are undisclosed. ④ Market Opportunity: Capital market data indicates a proposed valuation of ~$47B and ~$1B in funding as of September 2026 (media reports, unverified); revenue composition is not public.

💸 Cost

Fixed monthly costs for Devin accounts, code hosting, and API calls are approximately 6,000-8,000 RMB. After amortizing by project, tool costs represent about 5%-10% of revenue, with the remainder being human labor costs for quality assurance.

⏱ Time Investment

Approximately 6-8 hours daily. AI agents handle coding, while human QA and client communication take 2-3 hours. The initial setup phase requires an extra 2-3 hours for requirement decomposition and tool debugging, which can be reduced to 1 hour of daily maintenance once stable.

🚀 Getting Started

Start by listing 'AI-driven software development' services on platforms like Zhubajie, Taobao Services, or Upwork. Use Devin to complete 2-3 small projects to establish a delivery loop. Once the QA process is validated, transition to selling monthly labor packages to recurring clients to scale. During the cold start phase, prioritize low-risk projects with clear structures and acceptance criteria, such as landing pages or data reports.

🔑 Keys to Success

✅ Charge for output rather than billable hours
✅ Build client trust through human QA and accountability
✅ Compound growth through vertical-specific SOPs
✅ Control the delivery cadence of human-AI collaboration

⚠️ 风险

⚠️ AI-generated code may contain hidden defects, requiring rigorous testing and human oversight to prevent rework from eroding profits.
⚠️ Dependency on a single AI programming provider carries risks of price and capability volatility; maintain alternative toolchains like Claude Code.
⚠️ Client expectations for AI delivery can be volatile; early failures carry high reputation repair costs compared to traditional outsourcing.

📌 Real Cases

📌 36Kr reported that Cognition's three-founder team earns ~10 million RMB daily with $492M in annual revenue, validating the business model of selling AI labor.
📌 AgentScout's deep dive highlights that Cognition built a $26B autonomous engineering empire in 24 months, suggesting early individual adopters can capture the same demand wave.
📌 Luna reported that Cognition's proposed valuation reached $47B in September 2026, up from $25B in April, reflecting the continued heating up of the AI agent workforce sector.

Sources