Ship
Traced agent
A real multi-step task, tools, budgets, a kill switch, a trace you can replay.
AI Agent Development
Build the loop yourself: tools, state, budgets, tracing — and the judgment to ship a pipeline instead when that is the senior call.
Created by Baljeet Dogra
Course Objectives
- Build plan → act → observe yourself before a framework hides the bug.
- Design tools with validation, idempotency and recoverable errors.
- Cap steps, cost and irreversible actions.
- Say “this should be a pipeline” and mean it.
What you will produce
Write
Decision note
Why agent, pipeline or one call — rejected options named.
6 weeks curriculum
Expand a part for the syllabus. Content stays searchable when closed.
01 Tools Week 1
Schemas, validation, idempotency, errors the model can use.
- Schemas
- Idempotency
- Errors
- Ergonomics
02 The loop Weeks 2–3
Build it yourself. Then a framework, and where it obscures bugs.
- ReAct
- State
- Frameworks
- Debug
03 Bounds Week 4
Budgets, approvals, memory, what belongs in a database.
- Budgets
- HITL
- Memory
- Kill switch
04 Multi-agent honesty Weeks 5–6
When extra agents make things worse. Capstone + decision note.
- Multi-agent
- Tracing
- Eval
- Capstone
Who this is for
Engineers past a single API call
You want agents that survive contact with tools and money.
LangGraph / LangChain users
You can follow a tutorial. You cannot debug a trace.
Prerequisites
- Python
- You have called an LLM API with tools or JSON
- Git and a terminal
Not a fit if
- Autonomous-agent hype with no budget
- A first programming course
Related: Hands-on LangGraph · Applied GenAI Engineering
Questions
LangGraph required?
You build the loop first. A framework is optional and examined, not assumed.
Multi-agent by default?
No. Multi-agent is a last resort you have to justify.
Ready to start?
Six weeks to a traced agent — or a pipeline you can defend. Create an account to enrol.
Enrol now