A Non-Developer's Guide to Using AI Agents: From Data Storage to Workflow Standardization

Habitualizing Data Storage and Security

When non-developers use AI agents to share local HTML tools, the limitations of static screens necessitate data storage and state management. To reduce the burden of operating a formal DB, Google Sheets was utilized as a lightweight shared data store, with deployment and execution automated via Apps Script and clasp. Additionally, security management became a natural work habit, such as using .env files and .gitignore to prevent token exposure during webhook integration.

Structuring Repetitive Tasks as Skills

To delegate tasks previously done manually, such as file organization or meeting minutes, to AI agents, clear specifications for inputs, outputs, and exception handling were required. The author defined and stored recurring judgment criteria as skills for Codex and Claude. In particular, the meeting minutes skill was specialized weekly through a feedback loop that identified improvements and updated the skill by comparing generated results with final versions. This process confirmed that understanding business structure is more important for performance improvement than AI's sentence generation capability.

Automating Data Analysis

For Google Analytics (GA) reporting tasks, MCP was used to collect data via API, maintain report structure through skills, and automate monthly comparative analysis. While data collection became easier, connecting the collected data to specific questions remained a human domain. The author emphasized that collaboration with AI agents goes beyond simple coding to structuring business inputs, outputs, security, and validation criteria, concluding that one must consider how work will be replaced and redefined in the future.