What You Will Learn
Three goals. First, you will be able to explain clearly why an AI agent is fundamentally different from a chatbot. Second, you will know the names and definitions of the five components every agent is built from. Third, you will apply that framework to a real-world scenario and see how it maps onto something concrete.
The Answer vs The Worker
What a chatbot does
A chatbot operates on a simple loop: one question in, one answer out, done. Ask it what time it is in Tokyo and it tells you. Each exchange is independent. It has no memory of what was said before unless explicitly shown the history. It does not take action in the world. It does not call APIs. It responds and stops.
What an agent does
Give an agent a goal — 'Book me a flight to Milan next Tuesday. Hotel near the centre under 150 euros. Add both to my calendar and send me a summary' — and something fundamentally different happens. The agent does not produce one response. It executes a sequence of actions: searching a flight booking site, searching hotel sites, writing to a calendar API, drafting and sending a summary email.
The agent works. One goal → many actions. Autonomous.
The Five Components of Every AI Agent
Every agent — simple or complex, no-code or full-code — is built from exactly five components.
| # | Component | Definition |
|---|---|---|
| 01 | Perception | How the agent receives information. Text, documents, API responses, live web feeds. If it is not in the perception layer, the agent is blind to it. |
| 02 | Planning | How the agent decides what to do. The LLM reasons through what actions are needed, in what order, and what to do if something fails. |
| 03 | Tool Use | The agent's hands. Web search, calculators, database queries, email APIs, code executors. The agent calls these tools and uses their results to take action in the world. |
| 04 | Memory | Short-term: the current session context. Long-term: a database the agent reads and writes across sessions, enabling it to remember over time. |
| 05 | Feedback Loop | The agent checks its own output before considering the task complete. This is what prevents silent failures. |
Worked Example: The School Flu Sentinel
White Box vs Black Box Agents
Not all agents work the same way under the hood. One of the most important distinctions in agent design — especially in scientific and medical domains — is whether the agent's reasoning is transparent and inspectable, or opaque and probabilistic.
Black Box Agents (LLM-based)
Most modern agents are built on large language models. When you ask the agent why it made a decision, it can explain its reasoning in natural language — but that explanation is itself generated by the model, not extracted from a verifiable internal process. You cannot trace the exact path from input to output. This makes LLM-based agents powerful but difficult to audit in safety-critical settings.
White Box Agents (Rule-based or Algorithmic)
A white box agent is one whose decision-making is fully transparent and inspectable. Every step of the reasoning can be traced, verified, and explained. Guido Putignano, bioERGOtech's director, described a real example from biomedical engineering research:
In this agent, you can audit every step. If the shear stress calculation is wrong, you can trace exactly which measurement was off, which formula was applied, and where the error occurred. This matters enormously in research and clinical contexts where results feed into publications or patient decisions.
📝 Your Lesson Submission
Share your reflection, ask a question, or leave a comment. Your mentor will respond.
You need to be logged in to submit your reflection.