Stopping Criteria

What is Stopping Criteria in Agent Fundamentals?

Stopping criteria is the set of rules that tells an AI agent:

"You have done enough. Stop working and return the result."

Without stopping criteria, an agent could keep thinking, planning, calling tools, and generating actions forever.

Why It Matters

An AI agent typically follows a loop:

User Request
      ↓
Planning
      ↓
Action
      ↓
Observation
      ↓
Reflection
      ↓
Need More Work?
      ↓
   Yes → Loop Again
   No  → Stop

The decision at "Need More Work?" is the stopping criteria.

---

Example 1: Web Search Agent

User:

Find the capital of France.

Agent:

Search web
   ↓
Found: Paris
   ↓
Answer user
   ↓
STOP

Stopping criterion:

Required information found.

---

Example 2: Coding Agent

User:

Fix my failing unit test.

Agent:

Read code
   ↓
Analyze error
   ↓
Modify code
   ↓
Run tests
   ↓
Tests pass?

If:

YES

Stop.

If:

NO

Continue debugging.

Stopping criterion:

All tests pass.

---

Example 3: Research Agent

User:

Research AI agents.

Agent:

Search source #1
Search source #2
Search source #3

Stopping criteria might be:

At least 10 reliable sources collected.

or

Research confidence > 90%.

Common Types of Stopping Criteria

1. Goal Completion

Most common.

Stop when the objective is achieved.

Example:

Task:
Book a hotel

Hotel booked
→ Stop

2. Maximum Iterations

Stop after a fixed number of loops.

Example:

Max Loops = 10

Agent:

Loop 1
Loop 2
...
Loop 10
STOP

Used to prevent infinite loops.

3. Time Limit

Stop after a certain amount of time.

Example:

Maximum Runtime = 5 minutes

After 5 minutes:

STOP

Useful for production systems.

4. Token Limit

For LLM-powered agents.

Example:

Max Tokens = 50,000

If the agent reaches the budget:

STOP

This controls cost.

5. Confidence Threshold

Stop when confidence is high enough.

Example:

Confidence = 95%

Agent decides:

Answer is reliable.
STOP

6. Error Threshold

Stop if too many errors occur.

Example:

5 consecutive failures

Then:

STOP
Report failure

7. Human Approval

Agent waits for user confirmation.

Example:

Generated deployment plan.

Then:

Approve?

If user approves:

Continue

Otherwise:

Stop or revise ```