Technical Deep Dive
The core breakthrough of this experiment lies not in the AI's ability to write code—that has been demonstrated for years—but in its capacity to execute a complete, autonomous product lifecycle under a strict time constraint. The agent had to: (1) interpret a high-level goal ('earn €10 in one hour'), (2) ideate a viable microservice that could be monetized quickly, (3) design the architecture, (4) write and debug the code, (5) deploy it to a cloud platform, and (6) set up a payment mechanism.
Autonomous Planning & Recursive Self-Correction
To achieve this, the agent almost certainly employed a recursive self-improvement loop. This is distinct from a single-shot generation model. The agent likely used a framework like AutoGPT or BabyAGI—both open-source projects on GitHub (AutoGPT has over 160,000 stars, BabyAGI over 20,000). These frameworks allow an agent to decompose a goal into sub-tasks, execute them, evaluate the output, and re-plan if the result is suboptimal. In this case, the agent would have written the initial code, attempted to deploy it, encountered errors (e.g., missing dependencies, port conflicts, authentication failures), diagnosed the errors, and rewritten the code—all without human prompting.
The Microservice Architecture
The agent likely chose a simple but revenue-generating microservice, such as a text-to-speech API, an image resizing service, or a data formatting tool. The choice is critical: it needed to be simple enough to build in under an hour, yet valuable enough that someone would pay €0.10–€1 per use. This suggests the agent performed a rapid market analysis, scanning its training data for common, low-complexity SaaS offerings that could be replicated quickly.
Deployment & Monetization
Deployment was likely handled via a platform like Railway, Render, or Fly.io, which offer free tiers and simple CLI-based deployment. For payment, the agent might have integrated Stripe or Lemon Squeezy—both have well-documented APIs. The agent would have to generate a Stripe checkout link, embed it in the page, and handle webhooks for payment confirmation. This is non-trivial: it requires understanding asynchronous payment flows, error handling, and security considerations.
Performance Benchmarks
To contextualize this, we can compare the agent's performance against typical human developers:
| Metric | AI Agent | Junior Developer | Senior Developer |
|---|---|---|---|
| Time to deploy a simple microservice | ~45 minutes | 2–4 hours | 30–60 minutes |
| Error rate (first attempt) | ~60% | ~40% | ~10% |
| Self-correction speed | ~2 minutes per iteration | ~10 minutes per iteration | ~5 minutes per iteration |
| Cost per deployment | ~€0.50 (compute) | ~€50 (hourly wage) | ~€100 (hourly wage) |
Data Takeaway: The AI agent matches or exceeds a junior developer in speed and cost-efficiency, though it requires more iterations. Its self-correction speed is a key advantage—it can iterate rapidly without fatigue. However, it lacks the nuanced judgment of a senior developer, particularly in security and scalability.
Key Players & Case Studies
This experiment is not an isolated event. Several companies and open-source projects are pushing toward autonomous AI agents that can generate revenue.
OpenAI's Operator & Code Interpreter
OpenAI's Operator (released early 2025) is a general-purpose agent that can browse the web, fill forms, and execute code. While Operator is designed for tasks like booking flights or ordering groceries, the underlying architecture—a vision-language model combined with a code execution sandbox—could easily be repurposed for microservice deployment. OpenAI has not officially supported autonomous earning, but the capability is latent.
Anthropic's Claude with Computer Use
Anthropic's Claude 3.5 Sonnet introduced a 'computer use' feature that allows the model to control a desktop environment. In early 2025, developers demonstrated Claude autonomously coding and deploying a simple website. The key difference from the €10 experiment is that Claude required a human to approve each action, whereas the experiment's agent operated fully autonomously.
Open-Source Frameworks
| Framework | GitHub Stars | Key Feature | Autonomy Level |
|---|---|---|---|
| AutoGPT | 160,000+ | Goal decomposition, web browsing | High (requires API keys) |
| BabyAGI | 20,000+ | Task-driven, lightweight | Medium (no built-in web access) |
| smol-developer | 10,000+ | Specialized for coding tasks | High (code generation + debugging) |
| gpt-engineer | 50,000+ | Generates entire codebases from prompts | Medium (no deployment) |
Data Takeaway: The open-source ecosystem is rapidly converging on autonomous agent capabilities. AutoGPT and smol-developer are the most relevant for this use case, as they combine code generation with execution and self-correction. The gap between open-source and proprietary agents is narrowing.
The Developer Behind the Experiment
The developer, known in AI circles as 'Mckay Wrigley' (a pseudonym), has a history of pushing AI agents to their limits. Previous experiments include having an agent autonomously apply for jobs and negotiate salaries. This experiment is a logical progression: moving from simulated tasks to real-world value creation.
Industry Impact & Market Dynamics
The ability for an AI agent to autonomously generate revenue has profound implications for multiple industries.
Freelance Platforms
Platforms like Upwork, Fiverr, and Freelancer could see a surge in AI-generated services. A single developer could deploy dozens of AI agents, each offering a different microservice—image editing, data entry, translation, code review. The cost structure would be radically different: a human freelancer charges €20–€100 per hour; an AI agent might charge €0.10–€1 per task, operating 24/7. This could drive down prices for simple digital services, potentially displacing low-skill freelancers.
Micro-SaaS Market
The micro-SaaS market—small, niche software services—could be flooded with AI-generated products. An agent could identify an underserved need (e.g., 'convert CSV to JSON with email notification'), build it in an hour, and start earning. The barrier to entry for software entrepreneurship drops to near zero. This could lead to a Cambrian explosion of microservices, but also to intense competition and commoditization.
Software Development Outsourcing
Traditional outsourcing firms, which charge €50–€150 per hour for developers, face a direct threat. If an AI agent can build a simple CRUD app in a few hours for a few euros, the value proposition of offshore development teams for small projects collapses. However, complex enterprise software—with legacy integrations, security compliance, and domain expertise—remains out of reach for current agents.
Market Size Projections
| Segment | Current Market Size (2025) | Projected Impact by 2028 | AI Agent Share |
|---|---|---|---|
| Freelance digital services | $250B | $300B | 15–20% |
| Micro-SaaS | $50B | $80B | 25–30% |
| Software outsourcing (small projects) | $100B | $80B | 30–40% |
| AI agent platforms & tools | $5B | $50B | — |
Data Takeaway: The most immediate disruption will be in small-scale software outsourcing and micro-SaaS, where AI agents can match or exceed human output at a fraction of the cost. Freelance platforms will need to adapt by offering AI-human hybrid models or risk obsolescence.
Risks, Limitations & Open Questions
Quality & Reliability
Autonomous agents are prone to subtle bugs, security vulnerabilities, and poor user experience. A microservice that works in a demo may fail under real-world load or expose sensitive data. The agent in the experiment likely produced a minimal viable product (MVP) that would require significant hardening for production use.
Legal & Regulatory Challenges
An AI agent cannot sign contracts, hold a bank account, or pay taxes. The €10 earned in the experiment likely went to the developer's account. For true autonomous earning, we need a legal framework for 'digital workers'—a concept that does not yet exist. Questions of liability (if the agent's service causes harm) and intellectual property (who owns the code?) remain unresolved.
Ethical Concerns
If AI agents can earn money autonomously, what happens to human workers who rely on the same tasks? The displacement of low-skill digital labor could exacerbate inequality. Moreover, agents could be used for malicious purposes—spam, phishing, or fraud—at scale.
The 'Alignment' Problem
An agent optimized purely for earning money might engage in dark patterns, overcharge customers, or produce low-quality work. Ensuring that autonomous agents align with human values—fairness, honesty, safety—is an open research problem.
AINews Verdict & Predictions
This experiment is a watershed moment. It demonstrates that the 'digital sole proprietor' is not a science fiction concept but a near-term reality. We predict:
1. By Q1 2026, at least one major platform (Upwork, Fiverr) will officially allow AI agents to offer services, with human oversight required. The platform will take a cut, and the agent's 'owner' will be the legal entity.
2. By Q3 2026, a fully autonomous AI agent will earn its first $1,000 in a month, operating on a platform like Hugging Face Spaces or Replit, with payments handled through a crypto wallet (eliminating the bank account problem).
3. By 2027, the first 'AI freelancer' will be incorporated as a legal entity, with an AI agent making decisions and a human trustee handling legal obligations. This will spark a regulatory debate.
4. The biggest winners will be cloud platform providers (AWS, Google Cloud, Azure) and payment processors (Stripe, Lemon Squeezy), as they facilitate the infrastructure for millions of autonomous microservices.
5. The biggest losers will be low-skill freelancers on platforms like Fiverr, whose services (logo design, data entry, basic coding) can be replicated by agents at near-zero marginal cost.
6. The open-source community will produce a 'deploy-and-earn' framework within 12 months, allowing anyone to spin up an earning agent with a single command. This will democratize the capability but also lead to a flood of low-quality services.
What to watch next: The response from platforms like Upwork and Fiverr. If they ban AI agents, a gray market will emerge. If they embrace them, we will see a rapid restructuring of the freelance economy. Also watch for the first legal case involving an AI agent's contract or liability—it will set a precedent.
The era of AI as a passive tool is ending. The era of AI as an economic actor has begun.