The previous project, Illuminate the Dark Dungeon, took longer than expected because it included login, data management, in-app purchases, game logic, and app release review responses. In particular, the Google Play Store release process involved continued reviews related to impersonation and copyright, which made me feel that continuing to build only apps was not the best way to quickly implement and validate ideas.

So I decided to make the next project a website rather than a mobile app. A website can be accessed without installation, and it can be checked immediately after deployment without app store review. In addition, because it can be registered with Google Search Console to check indexing and exposure, I judged it to be more suitable for quickly building a product and testing user response within a short period.

Another focus was AI-generated content and ad monetization. With AI, content that researches and summarizes news or issues can be produced relatively quickly. However, search traffic or ad revenue does not automatically happen just because AI generates content. Therefore, in this project, I wanted to build an actual AI-based content website and check what conditions and limitations appear during search indexing and ad review.

For the development tool, I used Claude Code. Previously, I had tried vibe coding mainly with Cursor, but when sessions became longer or the work scope became broader, the result often became unstable. So I wanted to see whether using Claude Code subagents to divide different types of work, such as planning, implementation, UI, and content processing, would lead to more stable results. [Anthropic - Claude Code Docs, Subagents]

Setting the Development Goal

The goal of this project was not to build a feature-heavy web service, but to create an AI-based issue content website within a short period and validate it in a real web environment.

I excluded features that would expand the implementation scope, such as sign-up and payment. Instead, I focused on a basic content delivery structure that allowed users to browse content and read individual issues.

During development, I used Claude Code for vibe coding. I wanted to check whether a workflow that divides roles using subagents could also be applied stably to real web development.

After deployment, I checked whether content pages could be indexed by search engines through Google Search Console and a sitemap. I also attempted Google AdSense and Ezoic reviews to see whether an AI-based content website could become a target for ad monetization.

System Architecture

"simnabuleo" was structured by separating the public website accessed by users from an internal admin website used to write and manage content. The public website is where users browse and read issue-based content, while the internal admin website is used as a management tool to generate, organize, and reflect AI-based content on the public site.

Core Implementation Flow

Content Generation and Publishing Structure

The content generation system of "simnabuleo" was built around Claude Code. Instead of simply asking one AI to write an article, I divided the roles required for content generation and assigned subagents to each role, so the system worked like a team of AI agents specialized in content creation.

This agent team divided work into roles such as finding recent issues, researching the background of those issues, summarizing key points, and organizing their meaning and outlook. Through this, I aimed to keep the flow and quality of the content more stable than generating everything at once with a single prompt.

The content automation was implemented using an open-source plugin that can automatically run and schedule Claude Code. I did not create a new automation logic through separate code. Instead, I configured Claude Code to run at scheduled times and set up the prompts and workflows it would use when running.

In other words, the content generation structure of "simnabuleo" was closer to prompt-based automation than code-based automation. With predefined prompts and subagent role structures, issue content was generated, then reviewed and edited in the internal content management website before being published to the public website.

Search Indexing Response

After publishing AI content on the public website, "simnabuleo" configured a sitemap so each content page could be discovered by search engines. Then I registered the site in Google Search Console and submitted the sitemap to check whether the published content URLs were included as indexing targets.

Through this process, I confirmed that simply deploying a website does not guarantee search visibility. The website needed a structure that allowed search engines to discover pages, and it was important to check whether each piece of content was provided through an independent URL, whether it was correctly reflected in the sitemap, and whether its indexing status could be checked in Search Console.

Therefore, the focus of this section is not to emphasize search traffic performance, but to show that I built a structure that allowed search engines to recognize an AI-content website and verified the actual indexing process.

Ad Monetization Validation

"simnabuleo" attempted reviews by Google AdSense and Ezoic to check whether an AI-based content website could become eligible for ad network review. The purpose was not to generate actual revenue, but to verify whether the public website could be expanded into an ad monetization structure.

For this, I reviewed elements required for ad review, such as content detail pages, category structure, basic site information, and public URL structure. Simply generating content automatically and publishing it on a website was not enough. I learned that readable content quality and site credibility were also necessary.

As a result, this project progressed to the stage of attempting ad monetization. I learned that turning an AI-content website into a monetizable product requires not only content production automation but also content quality control and site credibility.

How Claude Code Was Used

In this project, I used Claude Code not merely as a code generation tool, but as a tool for dividing and managing the web development workflow. In my previous Cursor-based vibe coding attempts, when the work scope became broader, planning, UI, content structure, and code modifications were mixed into one flow, which sometimes made the result unstable.

To reduce this issue, I used Claude Code subagents in "simnabuleo". I separated tasks with different characteristics, such as planning, implementation, UI composition, and content processing, by role and had each task reviewed from an independent perspective. This helped me organize public website development, internal content management website construction, and the content generation automation flow step by step instead of handling everything at once.

In particular, for the content generation area, I built a team of AI agents specialized in content creation inside Claude Code. Each subagent was assigned a role such as issue selection, research, or article writing, and I used an open-source plugin that automatically runs and schedules Claude Code so predefined prompts could be executed repeatedly. Rather than writing separate automation code, the key experiment of this project was automating the content generation flow through prompts and role structures.