Introduction to Subagents
AdvancedCoordinators, context budgets and explicit state passing. Build a research-write-review pipeline with three specialised agents — and know when not to.
Program Fee
1-on-1 doubt clearing · WhatsApp group · Post-training Q&A
Request Syllabus / Call BackProgram Overview
Subagents solve two real problems — context exhaustion and task interference — and create several new ones. This day is honest about both. You begin by taking a task that genuinely fails in a single context and splitting it, so the motivation is felt rather than asserted. You then learn decomposition into truly independent pieces, writing a subagent brief that carries everything it needs and nothing it does not, and return contracts the coordinator can actually use. The core module is context: budgets across a hierarchy, explicit state passing versus shared memory and why explicit wins, what progressive summarisation destroys, and how to find which agent went wrong. You finish building a research-write-review pipeline — and learning when a subagent is the wrong answer.
What You'll Learn
- Explain the two problems subagents solve, and the cost of using them
- Recognise when a subagent is the wrong answer
- Decompose a task into genuinely independent pieces
- Write a subagent brief that carries everything needed and nothing more
- Design return contracts a coordinator can consume
- Choose between parallel and sequential execution
- Allocate context budgets across an agent hierarchy
- Pass state explicitly, and understand why shared memory fails
- Protect transactional facts from progressive summarisation
- Debug a multi-agent run and find the agent that went wrong
- Build specialist agents with review loops and escalation
Who Can Join
- •Developers building multi-agent systems
- •AI engineers and solution architects
- •Technical leads hitting context limits
- •Anyone whose single-agent solution has stopped scaling
PREREQUISITES
• Experience building with the Claude API or Claude Code
• Comfortable writing Python or TypeScript
Curriculum
→ The problem they solve
→ Coordinator and subagent
→ When to use them
→ The cost
→ Lab 0
Total: 7h across 4 modules
Tools & Technologies
Projects Included
- ▹ Lab 0 — A task that fails in one context, split across two agents
- ▹ Lab M01 — A real task decomposed into three subagent briefs
- ▹ Lab M02 — A coordinator with context budgets and explicit state passing
- ▹ Lab M03 — A research-write-review pipeline with three specialised subagents
Official Certifications
Assessment is continuous and practical: 4 Applied Learning Labs (60%) and an end-of-course practical assessment (40%). Passing mark 80%, plus completion of all 4 labs and full attendance. The CertTulen Course Completion Certificate is awarded on meeting that standard.
Questions? Contact us for a free counseling session.
Have questions about this course?
Send us your details — we'll respond within 24 hours.