Agentic AI · Part 1 of 7
What Makes an Agent
Agency, workflows versus agents, and the pieces an agent is made of.
Ask a plain language model “what's 47 times 89, plus 12?” and it does one thing: it predicts the most likely next tokens given the question, the same next-token-by-next-token process covered in Understanding Transformers. If the arithmetic isn't well enough represented in what it learned, it can guess wrong, and it has no way to double-check, look anything up, or do anything besides emit text once and stop.
What turns it into an agent is agency: the ability to decide what to do next on its own and act on that decision, repeatedly, without a person approving each step. Reasoning is how it decides. Tools are how it acts. Put both around the same model in a loop and it can work toward a goal by itself.
The model itself doesn't change to make this possible. It's still only ever predicting the next token. What's different is what it's predicting into: a running transcript that carries the user's question along with the agent's own past reasoning and the results of anything it's already tried.
Workflows and agents
Anthropic's guide Building Effective Agents draws the line this way. Workflows are “systems where LLMs and tools are orchestrated through predefined code paths.” Agents are “systems where LLMs dynamically direct their own processes and tool usage.” A pipeline that always summarizes a ticket, then classifies it, then drafts a reply is a workflow, even though every step is a model call. The developer fixed the order. In an agent, the model chooses the next step.
The pieces of an agent
| Piece | What it does | Covered in |
|---|---|---|
| Loop | Calls the model, runs what it asks for, feeds the results back, and stops at a final answer | The Agent Loop |
| Tools | Functions the model can request, which the surrounding program runs | Tools and Tool Calling |
| Planning and reasoning | How the model decides what to do next | Planning and Reasoning |
| Memory | What carries across steps and across sessions | Memory and Context |
| Orchestration | Routing between several agents, and the frameworks that handle it | Multiple Agents and Frameworks |
| Production safeguards | Authorization, confirmation, auditing, limits, and sandboxing around the loop | AI Agents in Production |
When an agent is the right choice
The same guide advises “finding the simplest solution possible, and only increasing complexity when needed,” and notes that for many applications, optimizing single model calls with retrieval and in-context examples is usually enough. Agents suit open-ended problems where the number of steps can't be predicted. Their autonomy brings “higher costs, and the potential for compounding errors.”