Prompt Engineering: Get reliable work out of AI
Upfold AI Lab
CoursesSoftware
Coding AI Agents with Claude
Go from a single API call to a working agent that uses tools, keeps state and stays inside guardrails.
IntermediateCreated by Upfold Engineering
Updated October 2026EnglishIntermediate
What you'll learn
Explain what separates an agent from a chat completion
Call the Claude Messages API with a system prompt, streaming and tool definitions
Design tools with schemas the model uses correctly
Build the agent loop with clear stop conditions and durable state
Add guardrails around write actions and sensitive data
Test, observe and control the cost of an agent in production
This course includes
10 min of narrated slidesAI tutor on every slide6 lectures, 24 slidesFull lifetime access18 quiz questionsCertificate of completion
Course content
1. From chat to agentPreview
2. Your first call to ClaudeLocked
Description
An agent is a model in a loop with tools and state. That sounds simple, and the first version usually is. The hard part is everything after: tools the model can actually use, a loop that stops when it should, context that doesn't blow up, write actions that don't surprise anyone, and tests that catch regressions before customers do. This course walks through building an agent on the Claude API step by step, in plain code you can read, with the design decisions explained rather than hidden behind a framework.