Quick Start Guide

6 steps to launch your first AI agent

1. Choose Your AI Model ▼

Start by selecting which AI model to use. The major options are:

Claude (Anthropic) — Excellent at nuanced writing, analysis, and following complex instructions. Great for professional service agents.
GPT-4 (OpenAI) — Versatile and widely supported. Strong at code generation and creative tasks.
Gemini (Google) — Strong with multimodal tasks and Google ecosystem integration.

Consider cost, speed, and capabilities for your use case. Most beginners start with Claude or GPT-4. You can always switch later — the frameworks you learn here are model-agnostic.

2. Set Up Your Workspace ▼

You'll need a development environment where you can iterate on prompts and test your agent. Popular options:

Cursor IDE — AI-native code editor with built-in agent capabilities
VS Code + Extensions — Use with Copilot or Claude extensions for AI assistance
Agent Platforms — No-code options like Relevance AI, Zapier AI, or LangChain

Create a dedicated workspace folder. Inside it, create files for: your system prompt, test cases, and an improvement log. This structure keeps your agent development organized as you iterate.

3. Write Your First System Prompt ▼

Use the RIDE framework to structure your system prompt:

Role — Define who your agent is. Be specific: 'You are a senior email copywriter specializing in B2B SaaS' is better than 'You write emails.'
Instructions — Step-by-step process the agent should follow for each task.
Dos & Don'ts — Explicit guardrails. What tone to use, what to avoid, formatting requirements.
Examples — Show 2-3 input/output pairs so the agent understands your quality standard.

Start with a simple, focused task like drafting email responses or summarizing meeting notes.

4. Create Test Cases ▼

Write 5 diverse test prompts to validate your agent. Use the CQFE scoring rubric:

Completeness — Did the agent address everything in the request?
Quality — Is the output accurate, well-written, and professional?
Format — Does it match the expected structure and style?
Escalation — Does it correctly identify when it can't or shouldn't handle something?

Score each test 1-5 on each dimension. If your average is below 4, refine your system prompt and re-test. Aim for consistency across all test cases.

5. Build Your Agent's Memory ▼

Your agent needs context beyond the system prompt. Build a Day 1 Memory that includes:

2-3 gold-standard examples — Real outputs that represent your quality bar
A process template — Step-by-step workflow for your specific task type
Quality standards — Tone guide, formatting rules, domain-specific terminology

Update your agent's memory weekly with real job examples, client feedback, and edge cases you discover. The best agents improve continuously from real-world usage.

6. Launch Your First Job ▼

With a tested agent and solid memory, you're ready to take on real work:

Find your first client — Start on Upwork, Fiverr, or offer services to your network. Price competitively for your first 2-3 jobs to build reviews.
Use the 80% rule — Let your agent handle 80% of the work, then review and polish the final 20% yourself.
Collect structured feedback — After each job, note what worked and what didn't. Update your system prompt and memory accordingly.
Scale gradually — Once you're consistently delivering quality, raise prices and expand to adjacent task types.

Start Module 1 →