Scaling agents horizontally in Agentic AI - Model Metrics & Evaluation

Metrics & Evaluation - Scaling agents horizontally

Which metric matters for scaling agents horizontally and WHY

When scaling agents horizontally, the key metrics to watch are throughput and latency. Throughput measures how many tasks or requests the system can handle per second. Latency measures how fast each task is completed. These metrics matter because adding more agents should increase throughput without making latency worse. Also, resource utilization helps check if agents are efficiently used. Monitoring error rates ensures quality does not drop as you add agents.

Confusion matrix or equivalent visualization

Throughput and Latency Example:

| Number of Agents | Throughput (tasks/sec) | Latency (ms/task) |
|-----------------|------------------------|-------------------|
| 1               | 100                    | 50                |
| 2               | 190                    | 52                |
| 4               | 370                    | 55                |
| 8               | 720                    | 60                |

This table shows throughput nearly doubling as agents double, while latency slightly increases.

Error Rate Example:

| Number of Agents | Total Tasks | Errors | Error Rate (%) |
|-----------------|-------------|--------|----------------|
| 1               | 1000        | 5      | 0.5            |
| 4               | 4000        | 20     | 0.5            |
| 8               | 8000        | 40     | 0.5            |

Error rate stays stable, showing quality is maintained.
    

Precision vs Recall tradeoff analogy for scaling agents

Think of precision as the quality of each agent's work and recall as how many tasks get done. When scaling horizontally, you want to increase recall (more tasks done) without losing precision (quality). If you add many agents but quality drops, it means precision suffers. If you keep quality high but throughput stays low, recall is low. The tradeoff is balancing speed and quality as you add agents.

For example, a customer support system adding more chat agents should handle more chats (higher recall) but still give correct answers (high precision). If agents rush and make mistakes, precision drops.

What "good" vs "bad" metric values look like for scaling agents horizontally

Good:

• Throughput increases close to linearly with number of agents.
• Latency increases only slightly or stays stable.
• Error rate remains low and stable.
• Resource utilization is balanced (agents are busy but not overloaded).

Bad:

• Throughput plateaus or grows very slowly despite adding agents.
• Latency increases sharply, causing delays.
• Error rate rises, showing quality loss.
• Some agents are idle while others are overloaded.

Common pitfalls when measuring scaling metrics

• Ignoring latency: Only tracking throughput can hide delays that frustrate users.
• Resource contention: Adding agents without enough CPU or memory causes slowdowns.
• Data leakage: Sharing state incorrectly between agents can cause errors.
• Overfitting to test load: Optimizing for a specific workload but failing in real use.
• Not measuring error rates: High throughput with many errors is useless.

Self-check question

Your system has 98% accuracy but only 12% recall on fraud detection when scaling agents horizontally. Is it good for production? Why or why not?

Answer: No, it is not good. The low recall means the system misses 88% of fraud cases, which is dangerous. Even with high accuracy, missing most frauds is unacceptable. You need to improve recall before production.

Key Result

Throughput and latency are key metrics; good scaling means throughput grows with agents while latency and error rates stay low.

Practice

1. What does scaling agents horizontally mean in agentic AI?

easy

A. Adding more agents to share and run tasks in parallel

B. Making one agent work faster by improving its code

C. Reducing the number of agents to save resources

D. Changing the task to fit a single agent's ability

Solution

1. Step 1: Understand the term 'scaling horizontally'

   Scaling horizontally means increasing the number of units (agents) to handle more work simultaneously.

2. Step 2: Apply to agentic AI context

   In agentic AI, this means adding more agents to share tasks and run them in parallel, speeding up processing.

3. Final Answer:

   Adding more agents to share and run tasks in parallel -> Option A

4. Quick Check:

   Scaling horizontally = Adding more agents [OK]

Hint: More agents working together means horizontal scaling [OK]

Common Mistakes:

• Confusing horizontal scaling with making one agent faster
• Thinking scaling means reducing agents
• Assuming scaling changes the task itself

2. Which of the following is the correct way to start multiple agents in parallel in Python?

easy

A. for agent in agents:
agent.start()

B. for agent in agents:
agent.run()

C. for agent in agents:
agent.execute()

D. for agent in agents:
agent.parallel()

Solution

1. Step 1: Identify the method to start agents in parallel

   In many agent frameworks, start() is used to begin an agent's process or thread asynchronously.

2. Step 2: Compare options

   run() usually runs synchronously blocking the loop, execute() and parallel() are not standard methods.

3. Final Answer:

   for agent in agents: agent.start() -> Option A

4. Quick Check:

   Use start() to launch agents in parallel [OK]

Hint: Use start() to run agents asynchronously [OK]

Common Mistakes:

• Using run() which blocks instead of start()
• Assuming execute() or parallel() are valid methods
• Not looping over all agents

3. Given this code snippet for scaling agents horizontally, what will be the output?

class Agent:
    def __init__(self, id):
        self.id = id
    def run(self):
        print(f"Agent {self.id} running")

agents = [Agent(i) for i in range(3)]
for agent in agents:
    agent.run()

medium

A. Agent 3 running

B. Agent running\nAgent running\nAgent running

C. No output, code has error

D. Agent 0 running\nAgent 1 running\nAgent 2 running

Solution

1. Step 1: Understand the Agent class and its run method

   The run method prints the agent's id with the message "Agent {id} running".

2. Step 2: Analyze the loop over agents

   There are 3 agents with ids 0, 1, 2. The loop calls run() on each, printing their messages in order.

3. Final Answer:

   Agent 0 running Agent 1 running Agent 2 running -> Option D

4. Quick Check:

   Each agent prints its id running [OK]

Hint: Each agent prints its id in order [OK]

Common Mistakes:

• Thinking all agents print the same message without id
• Assuming only one agent runs
• Believing code has syntax error

4. This code tries to scale agents horizontally but does not run agents in parallel. What is the error?

class Agent:
    def run(self):
        print("Running")

agents = [Agent() for _ in range(3)]
for agent in agents:
    agent.run()

medium

A. The list comprehension syntax is wrong

B. Agent class is missing an __init__ method

C. Agents are run sequentially, not in parallel

D. The run method should be named start

Solution

1. Step 1: Check how agents are executed

   The for loop calls run() on each agent one after another, so execution is sequential.

2. Step 2: Understand parallel execution requirement

   To scale horizontally, agents must run in parallel, e.g., using threads or async calls, not sequential calls.

3. Final Answer:

   Agents are run sequentially, not in parallel -> Option C

4. Quick Check:

   Sequential run ≠ horizontal scaling [OK]

Hint: Sequential calls don't scale horizontally [OK]

Common Mistakes:

• Thinking missing __init__ causes no parallelism
• Believing list comprehension is incorrect
• Assuming run must be renamed to start

5. You want to scale 5 agents horizontally to process independent tasks faster. Which approach best achieves this in Python?

hard

A. Run all agents sequentially in a single loop

B. Run each agent's task in a separate thread using threading.Thread

C. Use a single agent to process all tasks one by one

D. Run agents in a loop but wait for each to finish before starting next

Solution

1. Step 1: Understand the goal of horizontal scaling

   We want to run multiple agents at the same time to speed up processing independent tasks.

2. Step 2: Evaluate options for parallel execution

   Using threading.Thread runs agents concurrently, achieving horizontal scaling. Sequential loops or waiting block parallelism.

3. Final Answer:

   Run each agent's task in a separate thread using threading.Thread -> Option B

4. Quick Check:

   Threads enable parallel agent execution [OK]

Hint: Use threads to run agents in parallel [OK]

Common Mistakes:

• Running agents sequentially thinking it's parallel
• Using one agent for all tasks ignoring scaling
• Starting agents but waiting for each to finish before next