Decision path
AI Agent Stack
Use this path when you are deciding how much autonomy an AI product should have.
Step 1
Workflow
Use a predictable sequence when the product path is known.
Step 2Tool Calling
Let the model request external capabilities while the application stays in charge.
Step 3Function Calling
Use structured function arguments for narrow, known tools.
Step 4Agent
Add an agent loop when the next step depends on uncertain observations.
Step 5MCP
Consider a protocol layer when reusable tool and context servers matter.
Step 6Guardrails
Constrain permissions, validation, fallbacks, and review.
Step 7Evaluation
Test realistic tasks before trusting autonomy.