Start Here: Let Your AI Agent Build It
3 min read
What this is
This series teaches you how to let an AI agent build a production-quality application from scratch — from your idea to a running, deployed product.
Unlike traditional tutorials where you follow along step by step, your AI agent does the building. You describe what you want, review its work, and make decisions. The agent handles the code, tests, deployment, and documentation.
What you'll have at the end
A working application with:
Clean architecture backed by documented decisions
Full test coverage
Production deployment
Screenshots and documentation
Every technical choice recorded and justified
How it works
This series has two layers:
These docs (for you) — each page explains what the agent will do in that step, what decisions you'll need to make, and what the output looks like. You can follow along, understand what's happening, and intervene when needed.
The agent skill file (for your AI) — a machine-optimised instruction set that your agent downloads and follows autonomously. It contains the same process as these docs, but written for an AI to execute: templates, verification commands, state tracking, and error recovery.
Getting started
If you're using OpenClaw
Tell your agent:
> Download the build workflow skill from the Agent Build Workflow SKILL.md and follow it. Build me [your idea].
Your agent will download the skill, read the instructions, and start executing the 13-step process. You can follow along with these docs to understand what it's doing at each step.
If you're using another AI coding tool
These docs describe the full process. You can adapt the steps for your tool — the methodology (front-load thinking, phase-by-phase build, test after every change) works regardless of what AI you're using.
The 13 steps
Step | What happens | Your role |
|---|---|---|
0. Initialise | Agent creates project folder and state tracking | None |
1. Idea Pack | Agent structures your idea into 10 sections | Review and approve |
2. PRD | Agent writes full spec with acceptance criteria | Review and approve |
3. ADRs | Agent makes and records tech decisions | Optionally override |
4. Implementation Plan | Agent plans the build phases | Review |
5. Build Phase 1 |
Philosophy
Front-load the thinking. Steps 1-4 happen before any code is written. This means when the agent starts building, it has a clear spec, documented decisions, and a phased plan. The result: fewer rewrites, fewer "why is this broken" moments, and code that actually works.
One phase at a time. The agent never tries to build everything at once. Each phase is built, tested, and committed before moving to the next. This keeps quality high and makes it easy to recover if something goes wrong.
Everything on disk. Decisions, plans, and specs are written to files — not held in chat memory. This means the agent can pick up where it left off, sub-agents can work in parallel, and you have a full audit trail.
Next: What the Idea Pack looks like →
Ready to build something?
Jiffi can help you go from idea to working prototype fast. Book a free call.