Module 9: From Toy to System•Lesson 2 of 6
Production vs Experiments
Production vs Experiments
Know the difference. Treat them differently.
Experiment Mode
Goal: Learn, try things, break stuff safely
Characteristics:
- Local machine
- Frequent changes
- Okay to fail
- No one depends on it
- Easy rollback (git reset)
Mindset:
"Let's see what happens if..."
Example tasks:
- Trying new prompts
- Testing sub-agent patterns
- Exploring tool capabilities
- Building proof of concepts
Production Mode
Goal: Reliable, consistent, valuable
Characteristics:
- Always-on server
- Stable configuration
- Failures have consequences
- Others may depend on it
- Changes are deliberate
Mindset:
"This needs to work every time."
Example tasks:
- Daily briefings
- Email processing
- Monitoring and alerts
- Business-critical automation
The Hybrid Approach
Run both:
Production (VPS) Experiment (Local)
├── Stable config ├── Latest changes
├── Proven prompts ├── New ideas
├── Critical tasks ├── Testing
└── Monitored └── Breaking stuffTest in experiment. Promote to production.
Promotion Checklist
Before moving something to production:
- Tested: Does it work reliably? (5+ successful runs)
- Documented: Can you explain what it does?
- Recoverable: What if it fails? Can you roll back?
- Monitored: Will you know if it breaks?
- Bounded: What's the worst it can do?
Anti-Patterns
❌ Experimenting in Production
"I'll just try this quick change..." Production breaks
Fix: Separate environments
❌ Never Shipping to Production
"It's not quite ready yet..." Stays in experiment forever
Fix: Ship when "good enough", iterate
❌ No Rollback Plan
"I'll update the config and..." Can't undo when it breaks
Fix: Version control everything. Always.
Version Control for Agents
Your entire workspace should be in git:
cd workspace
git init
git add -A
git commit -m "Initial agent setup"
# After changes
git add -A
git commit -m "Added memory rules"
# To rollback
git log --oneline
git checkout <commit-hash>Every change is recoverable.