THE SPINE
Eighteen concepts
One canonical vocabulary runs under all three pillars. Every page on the site maps to exactly one of these concepts. Get the spine right and every layer above it inherits the rigour.
01
Model
The reasoning core — a commodity, selected by tier not vendor
02
Harness
Everything around the model — guides and sensors
03
Tools
External capabilities the agent invokes
04
Context Engineering
Designing what enters the window, and when
05
Context Management
Runtime budgeting, compaction, eviction
06
Memory
Persistent state — episodic, semantic, procedural
07
Retrieval / RAG
Grounding in external knowledge
08
Skills
Portable, reusable, packaged capability
09
Workflows
Orchestration along predefined code paths
10
Agents
Systems that direct their own process
11
Handoffs
One-way transfer of execution between agents
12
Multi-Agent Orchestration
Supervisor, swarm, mesh, hierarchy
13
Steering
Run-time control — mechanistic and operational
14
Evals & Observability
The quality and reliability layer
15
Guardrails & Safety
Input filtering, output validation, injection defence
16
AFK & Autonomous Agents
Long-running agents with no human in the loop
17
Code & Doc Indexing
Specialised retrieval for software agents
18
Standards & Interop
MCP, Agent Skills, A2A, OpenTelemetry