Section
AI / ML
Stack
Multi-Agent Systems · LLM Agents · Tool Use · Python
Category
Multi-Agent System · LLM Agents
Status
Core agent loop working, critic integration in progress
01 — Challenge
The core engineering challenge
Single-agent LLM systems hit a ceiling on complex multi-step tasks — decomposing work across specialised agents with explicit handoff contracts unlocks more reliable multi-step reasoning.
03 — Decisions
Engineering decisions and what they cost
01Each agent has a single defined role — no agent both plans and executes to prevent conflation.
02Inter-agent messages are typed and validated so the communication contract is explicit.
03The critic agent operates on outputs, not inputs — it cannot interfere with execution mid-task.
04 — Stack
Multi-Agent SystemsLLM AgentsTool UsePythonLangGraph