AI agent frameworks
Four technical parts turn a model into a working agent: context engineering, memory, tool use, and orchestration. The Holistic Agent Canvas organises these — and sixteen other elements — into one design and audit framework.
Context engineering
Deciding what information reaches the model at each step — the actual lever for reliable behaviour, not prompt wording.
Agent memory
What the agent retains across steps or sessions, so it doesn't repeat itself or lose the thread.
Tool use
How the agent takes an action outside the conversation — booking a slot, sending an email, updating a record.
Orchestration
How steps or multiple agents are coordinated toward one outcome.
The Holistic Agent Canvas
A 20-element framework across five planes — Purpose, Cognition, Action, Governance, and Evolution — for designing an agent before it's built.
View the Canvas →