I Built, Shipped, and Marketed a Mobile Game in 14 Days Using Only AI. Here’s What Actually Happened.

Summary

The article recounts a 14-day challenge where one person built, shipped, and marketed a mobile game using extensive AI tooling. The author emphasizes that Model Context Protocol (MCP) was the single most important tool decision, enabling seamless integration with vendor data (e.g., AppsFlyer, BigQuery). An AI agent named CLAW, powered by OpenClaw, managed user acquisition via the AppsFlyer MCP, while Data Locker streamed raw data into BigQuery for granular analysis.

ROI 360 automatically unified revenue and cost data, allowing CLAW to optimize campaigns. Creative Optimization surfaced winning creatives through AI insights. The result: 5,563 installs at $0.39 eCPI on ~$2,200 spend.

Key takeaways: MCPs are non-negotiable for agentic workflows; a robust measurement stack (Data Locker, ROI 360, Creative Optimization) is essential even for solo operators; human+AI collaboration outperforms either alone. The project underscores that with the right tools and AI agents, rapid app development and efficient UA are achievable, though store approval remains a bottleneck.

Analyst Note

As someone who's been in ad ops for over a decade, this article hits home. The claim that MCPs are 'the quickest bang for your proverbial buck' is absolutely true in practice. I've seen teams waste weeks building custom API integrations; MCPs slash that to hours.

The $0.39 eCPI on $2,200 spend is impressive but not unheard of for hypercasual games with strong targeting. What's more striking is the integration of Data Locker and ROI 360 – many UA teams still rely on fragmented data sources, leading to blind spots. The author's use of CLAW to auto-optimize campaigns in real-time is a glimpse into the future; we're already seeing similar agentic approaches from rivals like Moloco.

However, the article glosses over the real-world challenges: ad network bias, creative fatigue, and the need for human intuition in budget allocation. The claim that 'output quality depends on infrastructure around the agent' is spot-on – I'd argue that prompt engineering is still undervalued. For UA teams, the actionable takeaway is to prioritize MCP adoption and unified measurement stacks; otherwise, they'll be left behind as AI agents become the norm.

The creative optimization insight is also underrated – most advertisers I work with still struggle with creative attribution. Overall, this is a solid proof of concept that validates what many of us suspect: solo UA with AI is viable, but it requires disciplined tool selection.