Two Architectures: The 2x2 for Agent Design
Gartner puts enterprise-agent cancellation at 40%, and this course is about the decision most of those projects skipped: separating who's waiting for the result (a live human, or nobody) from who decides what happens next (the model, or your code) before picking an architecture. Cross those two questions and you get a 2x2 that tells you, before you write a line of code, which quadrant your project actually lives in and what that quadrant will demand of you. You'll place the Claw family and Rasa CALM honestly on that grid, decompose a real support-triage brief into its four architectural layers the way production systems actually split, and leave with the handoff design, hardening checklist, and eval framework the riskiest quadrant needs before it ships. If you've already taken Tools, Memory, and Multi-Agent Systems, this is the synthesis neither course taught on its own — the 2x2 itself. If you haven't, every idea here still stands on its own two feet.
MethodThis course teaches a discipline. The examples use today's tools; the method is meant to outlast them.
Rod Rivera
Professor
Desk
First lesson is free to preview.
What You'll Learn
- Separate the application axis (who's waiting for the result) from the implementation axis (who decides what happens next), and place any agent on the 2x2 that falls out of crossing them
- Name the five-part anatomy every agent shares — LLM Brain, Planning, Tools, Memory, the Loop — before any architecture-specific choice gets made
- Compare the Claw family (OpenClaw, NanoClaw, IronClaw, NemoClaw) against Rasa CALM honestly — what each quadrant anchor is built for, and what CALM structurally doesn't pretend to give you
- Decompose a single brief into its real architectural layers the way Sarah's triage system splits into four, and ask the two axis-questions separately per layer instead of once for the whole system
- Design the handoff between two quadrants explicitly — where state actually lives, what latency is visible, who's accountable — instead of discovering the gap in production
- State the structural argument for why Headless + Stochastic carries the sharpest risk profile, and the four-category hardening checklist (sandbox, budget, surface, observability) that quadrant needs before shipping
- Build a concrete eval suite starting point for any layer, and choose which of the three eval loops (offline, online, regression) a given deploy actually needs
- Explain why the organizational axis follows the same two axes as the technical one, and what the AI Product Engineer role actually spends its time on
Prerequisites
- Comfort with the idea of an LLM calling tools; this course does not re-teach tool-calling mechanics from scratch
- Helpful, not required: `Tools, Memory, and Multi-Agent Systems` — this course shares source lineage with that course and lessons 1-5 explicitly cross-link to its lesson 01 (the application axis) and lesson 05 (the implementation axis) rather than re-deriving them
- No prior Rasa or agent-framework experience required — both the Claw family and Rasa CALM are introduced from their architecture-level basics
Syllabus
Ready to keep going?
The lessons in Two Architectures: The 2x2 for Agent Design are open. The Prof Rod weekly lesson is the next thing, not a door.
