Find the smaller problem
Remove accidental complexity until the true constraint becomes visible.
02 / Systems
How I design systems.
A set of checks, diagrams, and examples I use when reviewing AI architecture.
Constraint → Choice → Evidence
Operating model
Start with the job.
First define what the user needs to finish. Then design the workflow, feedback, and failure paths around the model. The model is one part of the system.
ObserveUsers, constraints, failure cost
FrameSmallest valuable system boundary
DesignTradeoffs, controls, recovery paths
LearnProduction evidence and evaluation
How I approach a system.
Name the tradeoff
State what the design optimizes, what it gives up, and what would change the decision.
Design the failure path
Treat evaluation, observability, recovery, and human judgment as part of the system.
Earn complexity
Introduce a new component only when its operational cost is lower than the problem it removes.
Topic guide
Read the systems as connected problems.
These articles form two paths: deciding where AI belongs, and making it dependable after it reaches production.
Reliable AI products
Start with product fit, keep semantic judgments bounded, and make uncertainty visible to code and people.
Production LLM systems
Treat routing, admission control, scheduling, streaming, and recovery as one operating system for inference.