01
Orchestration
From Chatbot to Orchestrator
The five-phase evolution from single-prompt assistant to persistent operator - roles, routing, handoffs, and why multi-agent systems need an explicit boss.
Course
Building Production AI Systems:
From Demo to Autonomous Operations
A course built directly from the six scoped studies on the research page, plus two operator modules on governance and day-two operations. Same system, same evidence base, but organized like a practical syllabus instead of a paper queue.
Curriculum
Six studies, then the operating model.
The first six modules track the scoped study slate directly. The last two turn those findings into governance rules and day-two operating practice.
02
Ops Cost
The 40% Tax
How to measure maintenance overhead, self-generated work, and the hidden labor that makes "autonomous" systems far more expensive than the demo suggests.
03
Memory
Three-Layer Agent Memory
Core, archival, and procedural memory as separate layers with different budgets, retrieval rules, failure modes, and time horizons.
04
Accountability
Agent Accountability
Why tasks, owners, completion gates, and explicit review loops tell you more about system quality than traces alone ever will.
05
Topology
Hub-and-Spoke Beats Mesh
Where peer-to-peer agent messaging helps, where it fails, and why supervised topology wins once the system is large enough to hurt itself.
06
Post-Mortem
The Leni's World Post-Mortem
A real production miss that turned cross-model review into a permanent rule and made completion gates non-negotiable.
07
Governance
The Autonomy Ladder
Auto, Notify, Propose, Ask, Defer - a practical governance model for deciding what the system can do alone and what must return to a human.
08
Operations
Operating the Live System
What day-two really looks like: schedules, incident response, backlog pressure, worker hygiene, and continuous adjustment in a system that is already running.
Get notified
Be first to know when it launches.
The course is being built from the published papers and the live production system. Drop your email and I'll notify you when the first module is ready. No spam, no drip sequence - one email when it ships.
Before the course
Start with the papers and think pieces.
The course will eventually package the full operating model. Until then, the research page and writing section show the ideas in their current form.