How I code with AI agents, without being 'technical'

What It Covers

Ben Tosel's guide demonstrates how non-technical people can build functional software using AI coding agents through CLI terminals, having spent 3 billion tokens across four months building apps, tools, and features without traditional programming knowledge.

Key Questions Answered

• •CLI over web interfaces: Use terminal-based tools like Factory's Droid with Opus 4.5 on high autonomy instead of web versions of Claude or Lovable. CLI provides more capability, visibility into agent work, and better control over the coding process for non-technical builders.
• •Agents.md configuration file: Create a dedicated agents.md file in your repos folder with explicit setup instructions, GitHub commit rules, testing requirements, and workflow preferences. This instruction manual follows you across projects, ensuring consistent agent behavior and reducing repetitive configuration work.
• •VPS for persistent projects: Deploy a virtual private server with sync tools to keep local repos continuously updated and accessible from anywhere. This enables coding from mobile devices via Telegram bots or Slack channels, allowing project work during downtime without being at a desk.
• •Fail forward learning method: Build projects beyond your current capability level, then treat every bug or error as a learning opportunity rather than a failure. Clone open source projects, modify them, and ask the AI agent to explain concepts simply instead of following traditional coding tutorials.