Who decides the next step
AI in an app comes in four levels: a raw LLM, an LLM integration (your code calls it for one job), a workflow (fixed steps, AI is one of them), and an agent (the model picks the steps).
Last updated
Analogy
An integration is calling Stripe to charge a card. A workflow is a hard-coded Express route. An agent is a junior dev with a ticket and the API docs: you don't list every step, they work it out.
In an agent
The more the model decides, the more agentic the system is. Interviewers love asking where your project sits on this scale.
Practice
Interview questions
- Agent vs workflow?
- A workflow runs steps you designed in advance, in a fixed order.
- An agent lets the model choose the next step in a loop, until the goal is met.
- Workflows are cheaper, faster and easier to test; agents handle open-ended tasks.
- Default to a workflow; use an agent when the path depends on what is found.
Questions people ask
What is the difference between an LLM integration, a workflow and an agent?
An integration calls the model for one job, like summarising a ticket. A workflow runs steps you fixed in advance, and the model is one of those steps. An agent lets the model decide which step comes next at runtime. The more the model decides, the more agentic the system is.
How do I explain where my project sits on the AI scale in an interview?
Name the level and say who decides the next step. For example: it is a workflow, because the steps are fixed and the model only classifies. Or: it is an agent, because the model chooses which tools to call based on what it finds. Then say why that level fits the problem.