← Sticker Wall AGENT · DETAIL

Using open-source Devon to automatically fix GitHub issues and take freelance gigs, earning 15,000 RMB per month

Workflow: Every morning, screen GitHub and freelance platforms for bug-fix tickets with clear boundaries. Input the issue descript

AGENT

Key Fields

FIELD STAMPS

IndustryLocal Services

RegionGlobal

ScaleSME

ChannelOnline

🔧 Workflow

Every morning, screen GitHub and freelance platforms for bug-fix tickets with clear boundaries. Input the issue description and repository URL; Devon reads the code, modifies it, runs tests, and generates a patch locally in a sandbox. After human review, the PR is submitted or delivered to the client. Humans only handle three tasks: selecting tickets, reviewing code, and communicating for payment. Conduct a weekly review of ticket types where the agent failed, adding high-frequency failure patterns to a screening blacklist. This continuously improves ticket selection hit rates week over week, making the system run with less effort.

🛠 Setup Requirements

Requires intermediate-or-above programming skills to review the quality of code produced by the agent. Deploy Devon locally or on a cloud server (Python environment) and integrate mainstream LLM APIs. Setup takes about one to two weeks; start by practicing on your own open-source projects to verify repair success rates. Use a Docker sandbox to isolate the runtime environment and prevent the agent from accidentally deleting files, and pair it with an automated test script as an acceptance gate—all patches must pass tests before entering the human review stage.

🧰 Toolchain

🔧 Devon open-source agent framework
🔧 OpenAI or Anthropic LLM APIs
🔧 GitHub account and Git
🔧 Docker sandbox environment
🔧 Automated testing frameworks like Pytest
🔧 Freelance platforms such as CoderTool and Upwork

💰 Revenue

Around 15,000 RMB per month, settled per task. Simple issue fixes range from 300 to 800 RMB per ticket, while complex feature fixes exceed 2,000 RMB per ticket, completing 20 to 30 tickets monthly. Open-source project sponsor fixes and bounty issues account for about 30% of revenue, while enterprise outsourced bug backlog clearing accounts for 70%. Unit prices become more stable once repeat clients transition to monthly maintenance contracts.

💸 Cost

LLM API costs range from 500 to 1,500 RMB per month. Complex tickets utilize GPT-4 class models, while simple modifications switch to cheaper models to save tokens. Cloud servers cost about 200 RMB per month. The tool itself is open-source and free, while freelance platform commissions of about 10% to 20% are factored into customer acquisition costs.

⏱ Time Investment

2 to 3 hours per day, mainly spent on screening tickets, reviewing code, and communicating with clients. The agent runs multiple tickets concurrently in the background without occupying human time. Spend a concentrated half-day on weekends processing deliveries and payments. Once proficient, the average time from accepting a ticket to delivery is compressed to under 40 minutes per ticket.

🚀 Getting Started

Step one: Run Devon's official sample repository locally using Docker, have it fix a real open-source project's 'good first issue,' and log the success rate throughout. After successfully running three tickets, list performance-based bug-fixing services on platforms like CoderTool and Upwork. Keep early pricing low in exchange for reviews, display links to merged GitHub PRs on your homepage as trust credentials, and raise prices after obtaining ten positive reviews.

🔑 Keys to Success

✅ Only accept tickets with clear boundaries and automated tests to ensure agent success rates
✅ Humans must rigorously review every line of generated code before delivery to safeguard quality reputation
✅ Accumulate a library of repair cases to form reusable prompt and workflow assets
✅ Set test pass rates as a rigid delivery gate; for tickets that fall short, it is better to issue a refund than to deliver substandard work

⚠️ 风险

⚠️ Agent repair success rates are unstable; complex business logic still requires human rewriting, and actual output may fall below expectations
⚠️ Price competition in the gig market is fierce; commercial products like Devin dropping to $20 per month will attract more competitors
⚠️ Code generated by agents may introduce license pollution or security vulnerabilities, with individuals bearing the risk of post-delivery accountability
⚠️ Over-reliance on a single LLM API means price hikes or rate limits will directly drive up monthly costs and compress profit margins

📌 Real Cases

📌 The Devon open-source project gained over 3,400 stars on GitHub, positioned as an open-source alternative to Devin, enabling community users to autonomously complete code generation and issue fixing
📌 Cognition's Devin dramatically slashed its pricing from $500 per month to $20 per month in 2026 while supporting automated PR submissions, validating the commercial viability of AI-driven autonomous bug fixing for freelance work
📌 Multiple in-depth evaluations in 2026 compared Devin 2.0 with GitHub Copilot Agent Mode and Claude Code in parallel ticket processing, showing that autonomous bug fixing followed by automatic PR submission has become standard capability across all three tool categories, indicating that the supply side of the gig market is maturing rapidly

Sources