When Plans Break

Even the best plans encounter unexpected problems: APIs go down, data is missing, or assumptions prove wrong. Adaptive agents detect failures and create new plans on the fly, balancing between replanning too often (wasted time) and too late (cascading failures).

Interactive: Replanning Simulator

Watch how agents adapt when critical tasks fail

📋 Initial Plan Created

1. Fetch user data from API

2. Process user preferences

3. Generate recommendations

4. Send email notifications

Replanning Strategies

🔄

Full Replan

Discard old plan, create entirely new one

When: Critical assumptions violated

Cost: High (time to replan)

Benefit: Optimal new path

🎯

Partial Replan

Keep completed tasks, replan remaining ones

When: Failure in middle of plan

Cost: Medium (partial replanning)

Benefit: Preserve progress

🚨

Fallback Strategy

Switch to pre-defined backup plan

When: Common failure modes

Cost: Low (plan already exists)

Benefit: Fast recovery

When to Trigger Replanning

✅

Critical Task Fails

Task failure blocks all downstream tasks

✅

Assumption Violated

Core planning assumption no longer holds

✅

Resource Unavailable

Critical API, tool, or data source is down

❌

Minor Hiccup

Transient error - retry instead of replan

Plan and Execute

Your Progress