AI Universe · Concept and physical
Concept and physical
A concept is the job and its rules. A physical thing is a product doing that job somewhere. Sorting is a concept; writing it in Java or Python is physical. Evaluation is a concept; doing it in LangSmith or DeepEval is physical.
The same idea, for a whole system
The agent harness is a concept too: a loop, context, tools, state, checks, durable waits and tracing around a model call. It can be built on any platform. We built it twice, the same concerns both times; only the bindings changed.
Why this matters when you learn
| When you learn… | …you can |
|---|---|
| the concept (what the job is, what can go wrong) | judge any product that claims to do it, in an hour |
| only a product | use that one product, until it changes or you move platform |
| the concept, then one product properly | switch to the next product by mapping, not relearning |
Five kinds of physical thing
| Kind | What it is | Examples |
|---|---|---|
| Implementation | does one concept's job | Langfuse does tracing; vLLM serves models; pgvector stores embeddings |
| Bundle / platform | packages many concepts under one name | Microsoft Foundry, Bedrock AgentCore, Databricks |
| Protocol / standard | an agreed way for two concepts to talk | MCP (agent ↔ tools), A2A (agent ↔ agent), OpenTelemetry (anything ↔ tracing) |
| Wrapper | a convenience layer over another implementation | LiteLLM over provider APIs; a chat app over a model |
| Model | the trained weights themselves | Claude, GPT, Gemini, Llama, Qwen, gpt-oss |
| Pattern | a way of combining concepts, not a product | RAG, ReAct, human-in-the-loop |
Next: place the new thing →