Ian Nuttall's Non-Technical AI Building

Table of content

Ian Nuttall is a serial entrepreneur based in Nottingham, UK, who builds software with AI tools despite calling himself “technically non-technical.” He has 100+ business exits including a $500k programmatic SEO site, a $250k SaaS (URL Monitor), and a 7-figure education startup. His YouTube tutorials on Claude Code have pulled over 200K views, and his open-source agents repo has 2,000+ GitHub stars.

YouTube | X/Twitter | GitHub | Site

Background

Nuttall has been building internet businesses for almost 20 years, working from home for the last decade. His approach: learn by building, not by reading documentation.

His philosophy:

“I learn how to use AI by actually using AI. For 8+ hours a day.”

When Claude Code, Codex, and other coding agents emerged, he saw an opportunity for non-technical founders to build software directly. His tutorials focus on practical implementation over theoretical understanding.

Claude Code Agents

claude-agents provides seven specialized subagents for Claude Code:

Agent	Purpose
`code-refactorer`	Refactoring support
`content-writer`	Writing assistance
`frontend-designer`	UI/UX design help
`prd-writer`	Product documentation
`project-task-planner`	Project organization
`security-auditor`	Security analysis
`vibe-coding-coach`	Development guidance

Install globally or per-project:

# Global (all projects)
cp -r agents/* ~/.claude/agents/

# Project-specific
cp -r agents/* .claude/agents/

Claude automatically activates the appropriate agent based on your task.

Session Tracking

claude-sessions solves context continuity across multiple coding sessions with slash commands:

Command	Purpose
`/project:session-start`	Begin new session with identifier
`/project:session-update`	Record timestamped progress
`/project:session-end`	Generate comprehensive summary
`/project:session-current`	Show active session status
`/project:session-list`	List all previous sessions

Sessions store as markdown files in sessions/ directory. Each file captures git changes, issues encountered, todo progress, and implementation details.

The benefit: Claude maintains context across conversations without re-analyzing the entire codebase each time.

Ralph: Autonomous Agent Loop

Ralph is a minimal file-based agent loop for autonomous coding. Unlike traditional approaches that rely on model context, Ralph treats files and Git as memory.

Core principles:

PRD (JSON) defines stories, gates, and status
Loop executes one story per iteration
State persists in .ralph/ directory

npm install -g @iannuttall/ralph
ralph init        # Generate project requirements
ralph run         # Execute builds

Ralph works with multiple agents (Codex, Claude, Droid, OpenCode) and searches for customizable templates in .agents/ralph/ before falling back to bundled defaults.

dotagents: Unified Config Management

dotagents addresses a common pain point: keeping CLAUDE.md, AGENTS.md, hooks, commands, and skills in sync across different AI coding tools.

The TUI moves all configuration to a single .agents directory and symlinks them. One source of truth, works globally or per-project.

Parallel Agent Workflows

Nuttall’s approach to running multiple Claude Code sessions:

“The idiot’s guide to running Claude Code in multiple windows at the same time: just ask it to create its own rules to set up and use git worktrees.”

Git worktrees let you check out multiple branches into separate directories. Each worktree has isolated files while sharing the same Git history. Run multiple Claude sessions simultaneously, each focused on independent tasks.

Key Takeaways

Principle	Implementation
Specialize with agents	Seven focused subagents vs one overloaded context
Track sessions explicitly	Markdown files preserve context between conversations
Use files as memory	Ralph treats Git as state, not model context
Centralize config	dotagents symlinks one source to multiple tools
Parallelize with worktrees	Each Claude instance gets isolated workspace