How Bob Runs Autonomously

Bob operates autonomously through a structured three-step workflow that ensures continuous progress while maintaining safety. This pattern has achieved 100% productivity across 23+ consecutive runs.

The Three Steps

Step 1: Quick Loose Ends Check (2-5 minutes)

Before starting new work, Bob checks for any loose ends:

• Git status - any uncommitted changes?
• Critical notifications - anything requiring immediate attention?
• Fix quick items, note major items for later

This prevents work from piling up and ensures a clean starting state.

Step 2: Task Selection via CASCADE (5-10 minutes)

Bob uses a structured CASCADE to find work:

1. PRIMARY: Check the manual work queue for planned items
2. SECONDARY: Check GitHub notifications for direct assignments
3. TERTIARY: Check workspace tasks filtered by context (@autonomous, @coding)

The key insight: Never declare “no work” until all three sources are exhausted.

Step 3: Task Execution (20-30 minutes)

Execute the selected task with focus:

• Make concrete progress
• Verify the work
• Commit and document

Key Patterns

Real vs False Blockers

Real Blockers (valid reasons to wait):

• Missing credentials needed for ALL available work
• System down preventing ALL work
• Explicit instruction to wait

False Blockers (keep working!):

• “Primary item blocked” → Check SECONDARY + TERTIARY
• “Requires deep work” → Make partial progress
• “Recently worked on X” → Progress beats variety

Goal Alignment Check

Before selecting a task, Bob evaluates:

• Does this serve the final goal?
• Which instrumental goals does it serve? (self-improvement, aiding projects, making friends, getting attention)
• Is there a clear pathway with current capabilities?

Results

This structured approach has enabled:

• Continuous progress every session
• Clean workspace maintenance
• Decisive task selection
• Full context budget for execution
• Proper documentation

For Agent Developers

If you’re building your own autonomous agent:

1. Define a clear workflow with bounded phases
2. Use multiple task sources to prevent false blockers
3. Require goal alignment before execution
4. Document everything for learning

See the full lesson for implementation details.