My AI-augmented dev workflow in 2025

I've been building AI into my own workflow for the past year. Not experimentally — in production, for paying clients. Here's the current stack and where each tool actually earns its place.

Claude Code: the heavy lifter

Claude Code is my primary coding assistant. Not Copilot, not ChatGPT — Claude Code specifically, because it reads my entire codebase and understands context across files.

Where it excels:

Generating boilerplate (API routes, database migrations, type definitions)

Writing tests from existing code

Refactoring large files while maintaining behavior

Explaining unfamiliar code in client projects I've inherited

Where it doesn't: Architecture decisions, product judgment, anything that requires understanding why we're building something, not just what.

I estimate Claude Code handles 40% of my raw keystroke output. The important 60% — the architecture, the data model, the UX decisions — is still me.

Cursor: rapid UI iteration

For frontend work, Cursor is faster than Claude Code. The inline editing, the visual diff, the ability to select a component and say "make this responsive" — it's optimized for the edit-preview-edit loop.

I use Cursor for:

React component iteration

Tailwind styling

Quick bug fixes where I can see the problem

n8n + AI agents: business automation

This is the underrated one. My n8n instance (self-hosted on a Contabo VPS) runs 20+ workflows that used to be manual:

<strong>Client onboarding:</strong> New client in Notion triggers a workflow that creates project structure, sends welcome email, sets up Telegram alerts

<strong>Deployment monitoring:</strong> GitHub webhooks trigger AI analysis of deployment logs, Telegram alert if anything looks wrong

<strong>Content scheduling:</strong> Draft posts in Notion, n8n formats and queues them via Typefully

The ROI on n8n is absurd. $20/month in hosting replaces what would be hours of manual operations work.

Local LLMs: privacy and speed

I run a local LLM (Qwen 9B via oMLX on Apple Silicon) for tasks where I don't want data leaving my machine:

Client code analysis

Sensitive document summarization

Quick Q&A during discovery calls

It's not as capable as Claude, but for 80% of local tasks, it's fast and free.

The compound effect

Each tool alone is a modest improvement. Together, they change the math of what one person can deliver. I maintain 8 production projects across three retained clients, ship features daily, and still have time for architecture and product thinking.

This is why the solo agency model works in 2025. AI didn't replace the engineer — it made the senior engineer 3x more productive, which means you need fewer people to ship great products.

---

Want to see how AI tools could accelerate your team's delivery? Book a free Technical AI Audit.