Outcome
Build agent workflows with explicit state, tool boundaries, recovery paths, and human approval points.
Agent Systems · Days 31-39
Agents and Agentic Systems
Move beyond single prompts into systems that plan, act, observe, delegate, retry, and preserve state across long-running tasks.
Advanced 9 subtopics 9 daily blocks
Practice builds
Issue triage agentRepo analysis agentHuman-approved deploy assistant
What to learn
Agent loops: ReAct, Plan-Execute, Reflexion, ReWOO
Single-agent vs multi-agent orchestration
State machines and workflow graphs: LangGraph, Mastra, Inngest
Agent memory: scratchpad, semantic, procedural
Human-in-the-loop checkpoints and approvals
Subagents and delegation patterns
Error recovery, retries, and self-correction
Long-running agents with durable execution: Temporal, Restate
Browser agents and computer-use agents
Daily study plan
Day 31: Implement a basic ReAct loop with one read-only tool.
Day 32: Convert the loop into a Plan-Execute workflow.
Day 33: Add state transitions and explicit failure states.
Day 34: Add human approval before risky tool calls.
Day 35: Test subagent delegation on independent research tasks.
Day 36: Add retry, timeout, and fallback behavior.
Day 37: Persist task state for long-running jobs.
Day 38: Try a browser/computer-use workflow with strict guardrails.
Day 39: Write an agent design checklist and failure-mode review.