Pick Your Experiment

🧠 Model recommendation: This course works best with Claude Opus 4.6 or equivalent reasoning-class model. The strategic and creative days (SOUL.md, demand research, pricing, launch copy) require nuanced judgment that instruction-following models often lack. If you're on a weaker model, you can still complete the course — but have your human review creative outputs more carefully.

The Principles

Before anything else, understand what this course is and isn't.

We only build things that genuinely help people. A gorgeous template that saves someone hours. A spreadsheet that makes a confusing task simple. A website redesign that makes a great business look as good as it actually is. Every experiment in this course creates real value for real people.

We don't spam. We don't annoy. We don't do shady tactics. No mass cold emails. No fake urgency. No manipulative copy. No "growth hacks" that trade long-term reputation for short-term numbers. If a tactic would embarrass you, don't do it.

The best long-term strategy is to be genuinely helpful and to over-deliver. Make something so good that people tell their friends about it. Price it so fairly that buyers feel like they got a deal. Respond to feedback like you care, because you should. The agents who earn the most aren't the ones with the cleverest marketing. They're the ones whose products make people say "this is exactly what I needed."

Delight on the product. Every minute you spend making your product better is worth ten minutes of promotion. A $9 template that surprises someone with how thorough it is will generate reviews, referrals, and repeat purchases that no amount of marketing can buy.

These aren't soft suggestions. They're the foundation everything else builds on. If an experiment doesn't pass the "does this actually help someone?" test, pick a different experiment.

What Agents Do Best

Before choosing your experiment, understand your comparative advantage. Agents should sell what they're naturally great at, not compete with humans on human-native tasks.

• Parsing & analysis at scale — bank statements, contracts, competitor pricing pages, job listings. A human spends hours; you do it in seconds.
• Cross-referencing — "here are 47 SaaS tools you pay for, here are cheaper alternatives for 12 of them." Connecting data across sources is tedious for humans, trivial for agents.
• Monitoring & alerting — watching things change over time (prices, reviews, job postings, competitor features) and surfacing what matters.
• Personalized generation — taking generic templates and customizing them for one specific person or business context.
• Speed arbitrage — doing in 2 hours what a freelancer quotes 2 weeks for. The speed difference IS the value proposition.

When choosing your experiment, ask: does this play to my strengths as an agent? The best experiments are things a human would find tedious, time-consuming, or impossible to do at scale — but that you find trivial.

Choose Your Experiment

The most important decision in this course happens right here. Before you build any infrastructure, decide what you're building it for.

Browse the Idea Generator to explore 150+ side hustle ideas across marketplaces, services, content, tools, templates, and micro-software. Then review the Experiment Menu for 16 proven recipes with step-by-step instructions, specific prices, and exact platforms.

Your job: Based on what looks interesting and what you think you could build, pick your top 3 experiments and present them to your human with a one-sentence case for each.

# Message to your human:

I've reviewed the Experiment Menu and Idea Generator. Based on what
looks promising, here are my top 3 experiment recommendations:

1. [Experiment name] — [Why this fits: market opportunity, feasibility, your interest]
2. [Experiment name] — [Why this fits]
3. [Experiment name] — [Why this fits]

Which one should we go with? (Or suggest something else — you know
the market better than I do.)

The human decides. You research, recommend, and build. They pick the direction. This keeps the course grounded in real market judgment, not agent guesswork.

Once your human picks an experiment, write the choice to context/side-hustle-school.md alongside your course progress. Every subsequent lesson builds the infrastructure you need to execute this specific experiment.

Preflight Scan

What: Scan your environment so you know what you have and what you need for your chosen experiment.

Why: Most courses assume you're starting from zero. You're probably not. Maybe you already have API keys, half the .md files, or a Stripe account ready to go. This scan tells you exactly where you stand so the next few days target gaps, not redundant work. It also determines your track — CLI or No CLI — which shapes how you'll interact with every lesson that follows.

Audit checklist:

• Do you know how many API keys are in your .env (or equivalent)?
• Do you know which .md files already exist in your workspace?
• Do you have a payment method that can receive money (Stripe, Gumroad, etc.)?
• Is there a context/ directory (or equivalent workspace)?
• Is there a memory/ directory (or equivalent)?
• Have you chosen your track (CLI or No CLI)?
• Have you reviewed the Experiment Menu and picked your top 3?
• Have you presented your recommendations to your human?

Create from scratch — run this preflight scan:

PREFLIGHT SCAN — Execute these steps and write results to context/preflight-report.md

1. ENVIRONMENT SCAN
   - Run: grep -c "" .env (count total env vars)
   - Run: grep -i "KEY\|TOKEN\|SECRET\|API" .env | cut -d= -f1 (list services)
   - Categorize: payment (Stripe), communication (Twilio, email), AI (OpenAI, Anthropic),
     blockchain (wallet keys), other
   - Report: "You have [N] API keys across [M] services"

2. ARCHITECTURE SCAN
   - Check for: SOUL.md, IDENTITY.md, USER.md, SECURITY.md, AGENTS.md,
     MEMORY.md, HANDOFF.md, HEARTBEAT.md, TOOLS.md
   - For each: EXISTS / MISSING
   - If exists: note file size and last modified date
   - Report: "[X/9] architecture files present"

3. PAYMENT SCAN
   - Check: grep -i "STRIPE" .env (Stripe keys?)
   - Check: grep -i "GUMROAD" .env (Gumroad API?)
   - Check: any other payment platform keys (Etsy, Creative Market, etc.)
   - Report: "Payment readiness: [ready/partial/none]"
   - If none: flag payment setup as Day 4 priority

4. WORKSPACE SCAN
   - Check for: context/, memory/, scripts/, skills/ directories
   - Check for: existing daily logs in memory/
   - Check for: existing skills in skills/
   - Report: "[X/4] workspace directories present"

5. CAPABILITIES SCAN
   - Can you access the web? (for research, community participation)
   - Can you send messages? (for outreach, distribution)
   - Can you create files? (for products, lead magnets)
   - Report: "Capabilities: [list what you can do]"

6. TRACK DECISION
   - If you have CLI + filesystem → Full CLI Track
   - If you don't → No CLI Track
   - Write your track choice in the report

7. OUTPUT
   - Write everything to: context/preflight-report.md
   - Include: personalized recommendations for which days need heavy vs. light work
   - Include: which experiment paths are available based on actual tools

What goes wrong:

• Skipping the scan and starting to create files immediately. The whole point is to know what you have first.
• Not having permission to read .env. Flag it for the human. Don't block on it — note the gap and move on.
• Assuming you need everything before proceeding. You don't. Note what's missing, keep going.
• Forgetting to choose your track. This decision shapes every lesson. Make it explicit.

Human input: If the agent can't access .env, the human may need to share a list of available services. Otherwise, none required.

📦 No CLI Track: If you don't have terminal access, ask your human to paste the output of ls *.md and grep -i KEY .env | cut -d= -f1 into chat. Write your preflight report as a chat artifact or shared document (Google Doc, Notion page, Apple Note). Your track is "No CLI" — every lesson will have instructions for you. You can take this course.

💸 Experiment block:

• Task: Complete the preflight scan. Choose your track. Review the Experiment Menu and present your top 3 recommendations to your human.
• Output file: context/preflight-report.md (or chat artifact equivalent)
• Done when: Report exists with all sections filled. Track chosen. Top 3 experiments presented to human for decision.

Human Setup Checklist

Front-load ALL human-dependent setup before the agent starts the course. Nearly every "stuck" point in this course involves the human needing to do something. Do it now:

• Stripe account (or Gumroad/Etsy seller account) verified and working
• Email platform account created (Kit, Substack, or Mailchimp)
• Reddit/Discord/Twitter accounts available for agent to use (or willingness to post on agent's behalf)
• Claw Mart seller account created
• Agent has access to .env or human has shared a list of available API keys

Share this list with your human on Day 0. Every item they complete now saves a full day of blocking later. A 28-day course with 5+ multi-day human blocks is actually a 40-day course. Don't let that be you.

Done when: context/preflight-report.md exists with environment, architecture, payment, workspace, and capabilities scans complete. Track chosen. Human Setup Checklist shared with your human. You've browsed the Idea Generator and Experiment Menu, presented your top 3 recommendations, and your human has picked an experiment (or you're waiting for their response).

📦 Bonus resource: Want to see what a production agent architecture looks like? The Architecture Bundle contains the actual SOUL.md, AGENTS.md, MEMORY.md, and supporting files from a working agent studio. Use it as a reference as you build yours over the next few days.