The Thought-Action-Observation Cycle

Introducción a los agentes de IA

Adel Nehme

VP of AI Curriculum, DataCamp

The Agentic Trinity: Model, Tools, and Orchestration

¹ Wiesinger, J., Marlow, P., & Vuskovic, V. (n.d.). Agents.

The Agentic Trinity: Model, Tools, and Orchestration

¹ Wiesinger, J., Marlow, P., & Vuskovic, V. (n.d.). Agents.

The Agentic Trinity: Model, Tools, and Orchestration

¹ Wiesinger, J., Marlow, P., & Vuskovic, V. (n.d.). Agents.

Maintaining Memory, State, Reasoning, and Planning

¹ Wiesinger, J., Marlow, P., & Vuskovic, V. (n.d.). Agents.

The Thought-Action-Observation Cycle

¹ The Hugging Face Agents Course: https://github.com/huggingface/agents-course

The Thought-Action-Observation Cycle

$$

$$

$$

$$

Thought: The model decides the next step based on the user prompt.

¹ The Hugging Face Agents Course: https://github.com/huggingface/agents-course

The Thought-Action-Observation Cycle

$$

$$

$$

$$

Action: The agent takes an action, by calling the tools at their disposal.

¹ The Hugging Face Agents Course: https://github.com/huggingface/agents-course

The Thought-Action-Observation Cycle

$$

$$

$$

$$

Observation: The model reflects on the response from the tool. Feeding into the next set of thoughts and actions.

¹ The Hugging Face Agents Course: https://github.com/huggingface/agents-course

A Customer Support Agent Example

$$

$$

$$

"I forgot to cancel my subscription and was accidentally billed. Help me get a refund."

¹ Image generated with GPT-4o

The Thought-Action-Observation Cycle

The Thought-Action-Observation Cycle

Cycle	Thought	Action	Observation
1	"I need to understand this customer's situation. Let me check their account details."	Access customer database and retrieve subscription history	Customer has premium plan, renewed 3 days ago for $99

The Thought-Action-Observation Cycle

Cycle	Thought	Action	Observation
1	"I need to understand this customer's situation. Let me check their account details."	Access customer database and retrieve subscription history	Customer has premium plan, renewed 3 days ago for $99
2	"The renewal is recent. I should check our refund policy for accidental renewals."	Query company policy database for refund rules	Policy allows full refunds within 7 days for accidental renewals

The Thought-Action-Observation Cycle

Cycle	Thought	Action	Observation
1	"I need to understand this customer's situation. Let me check their account details."	Access customer database and retrieve subscription history	Customer has premium plan, renewed 3 days ago for $99
2	"The renewal is recent. I should check our refund policy for accidental renewals."	Query company policy database for refund rules	Policy allows full refunds within 7 days for accidental renewals
3	"Great, they qualify! Now I need to process the refund."	Initiate refund transaction through payment system	Refund of $99 successfully processed

The Thought-Action-Observation Cycle

Cycle	Thought	Action	Observation
1	"I need to understand this customer's situation. Let me check their account details."	Access customer database and retrieve subscription history	Customer has premium plan, renewed 3 days ago for $99
2	"The renewal is recent. I should check our refund policy for accidental renewals."	Query company policy database for refund rules	Policy allows full refunds within 7 days for accidental renewals
3	"Great, they qualify! Now I need to process the refund."	Initiate refund transaction through payment system	Refund of $99 successfully processed
4	"Refund complete. I should update the customer and cancel their subscription to prevent future charges."	Send confirmation email and update subscription status to 'cancelled'	Email sent, subscription cancelled, ticket can be closed

The Thought-Action-Observation Cycle

$$

$$

$$

Final action: Ticket now can be closed

¹ Image generated with GPT-4o

Let's Practice!

Introducción a los agentes de IA