Building AI agents for your business
One agent, one real recurring task, under control: tools, a human approval step, logs, cost limits and a test set. Built with the OpenAI Responses API or the Claude API, in plain code.
Bought it? Sign in with the email from your purchase. Terms · Right of withdrawal
You can read and run short JavaScript (Node.js 20 or newer) and use a terminal. You need an API account with OpenAI or Anthropic and a payment method. A task you do at least weekly, with example cases, makes the course concrete.
Self-study course: slides, hands-on tasks and templates, at your own pace. No graded tests, no individual tutoring.
Founders, operations people and developers who want an agent to take over a real recurring task, such as sorting an inbox and drafting replies or preparing quotes, without losing control over what it does.
At the end one agent does a real task of yours: with defined tools, read-only access where possible, a human approval before anything leaves the house, a log of every step, a hard cost limit per run and a test set you rerun before each change.
Contents
- 01 · Agent or workflow?
- 02 · Foundations
- Messages, structured outputs and store: false14 min
- Tool calling and the agent loop16 min
- 03 · Tools and data
- Defining tools: read-only first13 min
- Retrieval over your own documents14 min
- 04 · Safety
- Prompt injection and untrusted input13 min
- Human approval: the agent proposes, you decide14 min
- 05 · Measure and run
- Evals: a test set for your agent15 min
- Logs, cost limits, versions and monitoring14 min
- 06 · Law
- GDPR, processing agreements and the EU AI Act12 min
Material
- Agent spec template (DE + EN)
- Tool definition example (OpenAI and Claude format)
- Agent loop with approval, log and cost limit (Node.js)
- Eval test-case sheet (CSV)
- Eval runner script (Node.js)
Fabian Faraz Farid · All courses