01
Babysitter
A human must stay in the loop forever.
Someone has to catch every mistake by hand. The more it runs, the more people you need — so it never really scales.
A learning platform for multi-agent systems
Learn how agent systems actually work.
Here you'll learn how multi-agent systems are built, which pattern fits which use case, the pitfalls that break them — and how it all connects. From a single agent to a production system.
The ladderloop → system
Start at the bottom. Every rung sits inside the next.
The cost of skipping architecture
Three postures teams fall into when there's no guarantee.
When nothing checks an agent's output automatically, every system slides into one of these — and each one quietly gets more expensive.
02
Auditor
Every result needs post-hoc review.
Nothing guarantees the result, so a human re-reads everything before it ships. The cheap agent turns out to be the expensive one.
03
Prayer
Outputs accepted unverified.
You ship whatever it produces and hope. Fast and cheap — until it breaks in production, and no one can say why.
The Ladder
Four rungs, from a loop to a system.
A learning ladder, not a hierarchy. Pick a rung to see its patterns, anti-patterns, anatomy, and the frameworks it lives in.
Single Agent
One mind, looping over a task.
How a single agent reasons internally — thought, action, observation, reflection. The unit you'll keep returning to, even inside the bigger systems.
- ReAct
- Plan-and-Execute
- ReWOO
- Reflexion
- Tree of Thoughts
- +2 more
reason · act · observe
Workflow
Steps you wire in code. The LLM fills the slots.
Structure belongs in code. Workflow patterns are the deterministic skeleton: sequential, routing, parallel, loops. The model produces content; you control flow.
- Sequential
- Routing
- Parallel
- Loop
- Orchestrator-Workers
- +4 more
sequence · route · parallel
Multi-Agent
Specialists. Who decides who works next?
The point where coordination becomes the system's hardest problem. Supervisors, handoffs, swarms, blackboards — twelve canonical answers to the same question.
- Supervisor
- Hierarchical
- Magentic
- Agents-as-Tools
- Graph-based
- +7 more
supervise · hand off · swarm
Production
Memory, tools, runtime, governance, evals.
This is the floor every other rung was already standing on. We present L4 last because skipping it is the #1 reason agent prototypes die on contact with Tuesday — but conceptually, it is the foundation, not a final step.
- Conversational Memory
- Episodic Memory
- Semantic / Vector / Graph Memory
- Working Memory / Scratchpad
- Virtual Context Management
- +29 more
memory · tools · runtime
Where to next