Conversational AI lets a person request help through a chat interface. A direct LLM chat generates text in response to a prompt. A tool-using assistant can also act on a request—for example, by searching the web or generating an image—with the user supervising its actions. section 1.1.1
Table 1.1 The four LLM interaction patterns covered in this book
| Pattern |
Approval model |
Autonomy level |
Typical use case |
Example platform |
|---|---|---|---|---|
| Direct LLM chat |
No tools invoked; the model only generates text |
None |
Q&A, drafting, brainstorming, learning |
Original ChatGPT (2022), early Claude (2023) |
| Tool-augmented LLM |
Implicit per call; a user prompt triggers a single tool |
Low |
Image generation, web search, simple lookups |
ChatGPT with DALL·E, Gemini with Google Search |
| Assistant |
Per-task approval; the user confirms each action |
Medium |
Pair programming, cited research, document editing |
GitHub Copilot Chat, Cursor, Claude in Excel |
| Agent |
Goal-level approval with gated high-stakes actions; the user sets the objective, and the agent decides the steps |
High |
Multistep research, repo-wide refactors, browser automation |
Claude Code, OpenAI Operator, Devin |
Compare four interaction patterns by tool use, approval, and autonomy. The useful question is not just “Can it use tools?” but “Who decides what happens next?” section 1.1.1
An assistant typically handles a request, uses a tool if needed, and stops. An agent can instead choose further steps, inspect tool results, and continue toward a larger goal. Permission prompts alone do not settle the distinction: agents may still require approval for high-stakes actions. section 3.2.4 section 1.1.1
From a reply to an action
Ask for an explanation, and direct chat can produce one without a tool. Ask for a current lookup, and a tool-augmented system may search. Ask for an edit that you confirm before it is made, and the interaction is assistant-like. A system that plans and carries out a multistep workflow moves toward the agent pattern. section 1.1.1
When choosing a pattern, match autonomy to the task. A supervised assistant suits bounded actions; a longer-running agent needs tighter tool controls and checkpoints because one mistaken tool call can affect later steps. section 1.1.1 section 3.2.4
