The AI Universe

So many names, every week. Before you climb, draw the map. Every product, library, protocol and model you will ever hear about fits into one of seven families, and into one bucket inside it. Learn the bucket once, as a concept; then every new name is just another way of doing a job you already understand.

The map, in five seconds zoom 1

An AI systemseven families of stacksOpen: Compute & infrastructureCompute & infrastructureWhere does it run?Chips · Clusters and cloudServing engine · On the deviceOpen: DataDataWhat does the AI know?Ingestion · Lakehouse layersSemantic layer · Knowledge and retrievalQuality and governanceOpen: ModelModelWhat is the brain, and how is it made?Training data · Pre-trainingFine-tuning and alignment · The modelsModel evaluationOpen: Model accessModel accessHow does the app reach a model?Inference API · GatewayRouting and model choice · CachingOpen: The AI application (agent harness)AI application (harness)What surrounds the model call?Control loop · Context assemblyTools and protocols · State and memoryGuardrails and policyOpen: ExperienceExperienceWhere do people meet it?Chat and assistants · Copilots in appsConversational analytics · Coding agentsBackground agentsOpen: Assurance & operationsAssurance & operationsCan we prove it works, and run it?Evaluation · ObservabilitySecurity and red-teamingCost and FinOps
Every name you will ever hear goes into one of these seven families, and into one bucket inside it. Click a family to open it.

How to use it

Concept firstEach bucket is a job with rules, like the logic of a program. Learn that once.
Physical secondProducts are where the job happens to be done: open source, Azure, AWS, GCP, a vendor. Like the language you write the logic in.
Place every new nameFour questions file anything new in under a minute. Try it →
Then zoom inEach family links to the course or project that goes deep. Start with the harness →

The seven families

  1. Compute & infrastructure: Where does it run? (4 buckets)
  2. Data: What does the AI know? (5 buckets)
  3. Model: What is the brain, and how is it made? (5 buckets)
  4. Model access: How does the app reach a model? (4 buckets)
  5. The AI application (agent harness): What surrounds the model call? (7 buckets)
  6. Experience: Where do people meet it? (5 buckets)
  7. Assurance & operations: Can we prove it works, and run it? (6 buckets)

Next: concept and physical, the one idea that makes this work →