🪜 5 projects that make you an AI builder (no degree)

"AI engineer" sounds like it needs a CS degree. For most practical purposes it doesn't — it needs a portfolio of things you've actually built. These five projects, in order, take you from prompting to shipping real AI-powered tools. Each one teaches the next one's prerequisite.

Project 1: A specialist chat assistant

Build: a Claude Project or custom GPT with deep instructions for one niche job (quoting your services, answering your industry's FAQs). You learn: system prompts, context, and how instructions change behavior — the foundation of everything after.

Project 2: A working web tool

Build: a single-page calculator or checklist app for your trade, built by describing it to Claude and iterating (full method here). You learn: spec-writing, iteration discipline, and shipping something a stranger can use.

Project 3: An automation with AI in the middle

Build: a pipeline where something happens automatically — form submission → AI drafts the reply → it lands in your inbox for approval (Zapier/Make + an AI step). You learn: triggers, human-in-the-loop design, and where automation breaks.

Project 4: A data project

Build: feed a real dataset (your sales, reviews, timesheets) into AI on a schedule and produce a recurring insight report (reports guide). You learn: structured data, verification habits, and the difference between analysis and hallucination.

Project 5: An agent with guardrails

Build: a Claude agent that does a multi-step job — research, draft, file — with approval gates before anything irreversible (starter guide). You learn: tool use, scoping permissions, and safe autonomy — the current frontier.

Make the portfolio visible