So far it was pieces. Now comes the system.

You walked the full prompt path:

• Track 1. a well-structured prompt.
• Track 2. a well-sustained conversation.
• Track 3. a well-defined toolkit so the model can act.

Each piece, on its own, solves a bounded scenario. In real life, problems are uglier: classifying 50 different input types, chaining 3-4 steps where each one touches data from the previous one, doing retrieval before generating, measuring whether all that works when you can no longer mentally test every case.

For that you need something different: designing the system.

A multi-step system is several LLM (and non-LLM) pieces orchestrated where the output of one feeds the next. Designing one well means knowing where to split the problem, how to route inputs, when to stop a loop, and how to evaluate the full result. not each piece in isolation.

In this track:

• Orbit teaches you to design the flow: where to split the problem, what runs in a chain, what branches, when the loop has to cut.
• Forge brings the toolkit (what you learned in Track 3, now orchestrated).
• Echo teaches you to evaluate (how to know your system works).
• Atlas signs the capstone when your full system passes the evals.

It's the last track of the onboarding. When you close it, you're a systems engineer with LLMs, not a user.