What is agentic AI, really?
October 14, 2026 · Anil Jose · #agentic-ai #fundamentals
Most "AI agents" in demos are chatbots with a longer prompt. A real agent does three things a chatbot does not:
- It decides what to do next. The model picks a step instead of following a fixed script.
- It uses tools. It calls APIs, queries databases, runs code or searches documents.
- It loops. It looks at the result of each step and decides whether it is done.
The four building blocks
| Block | What it does | Example |
|---|---|---|
| Model | Reasons and chooses the next action | Claude on AWS Bedrock |
| Tools | Lets the agent act on the world | Search a knowledge base, create a ticket |
| Memory | Keeps context across steps and sessions | Conversation history in DynamoDB |
| Guardrails | Limits what it may do | Allowed tools, spend caps, human approval |
When not to use an agent
If the steps are known in advance, write a workflow. A fixed pipeline is cheaper, faster and easier to test. Reach for an agent when the path depends on what the agent finds along the way, such as triaging an incident or researching a customer question across many sources.
Where to start
Start with one tool and one narrow job. Measure whether it gets the job right before you add a second tool.