Part 3
AI
Useful AI needs the user’s trigger, intended progress, constraints, and evidence, not just chat history. This part follows that context into objects, permission, and the reusable components and trust tokens that make autonomy legible. It closes with memory, the handoff back to a person, and the discipline of finishing a job cleanly.
Tools in this part
- Job context · Fill-in card
Chapter 17 · AI Without Job Context Is Just Autocomplete
- After an escalation: “Summarize this customer” · Example
Chapter 17 · AI Without Job Context Is Just Autocomplete
- A context audit · Use it checklist
Chapter 17 · AI Without Job Context Is Just Autocomplete
- The three questions · Fill-in card
Chapter 18 · Generative Outcome-Oriented Design
- Scaling by accident vs. scaling by design · Fill-in table
Chapter 18 · Generative Outcome-Oriented Design
- The three questions for “Build a collaboration feature” · Example
Chapter 18 · Generative Outcome-Oriented Design
- A GOOD Iteration check · Use it checklist
Chapter 18 · Generative Outcome-Oriented Design
- Four modes · Fill-in card
Chapter 19 · Capability Isn’t Permission
- Four modes: a support team rolls out an agent · Example
Chapter 19 · Capability Isn’t Permission
- Consent scope · Fill-in card
Chapter 19 · Capability Isn’t Permission
- Check the boundaries for each agent action · Use it checklist
Chapter 19 · Capability Isn’t Permission
- Core component families · Fill-in table
Chapter 20 · The Agentic Design System
- Core component families: states a spinner and a toast can’t carry · Example
Chapter 20 · The Agentic Design System
- Semantic contract · Fill-in card
Chapter 20 · The Agentic Design System
- A system audit · Use it checklist
Chapter 20 · The Agentic Design System
- Trust token · Fill-in card
Chapter 21 · Trust Tokens: A Shared Grammar for Delegated Products
- Trust token: an external, high-risk action · Example
Chapter 21 · Trust Tokens: A Shared Grammar for Delegated Products
- Trust-token inventory · Fill-in table
Chapter 21 · Trust Tokens: A Shared Grammar for Delegated Products
- Start a trust-token inventory · Use it checklist
Chapter 21 · Trust Tokens: A Shared Grammar for Delegated Products
- Memory scope · Fill-in card
Chapter 22 · Memory Is a Permission Model
- Memory scope: a tone, a timezone, a recommendation · Example
Chapter 22 · Memory Is a Permission Model
- Memory provenance · Fill-in card
Chapter 22 · Memory Is a Permission Model
- A memory audit · Use it checklist
Chapter 22 · Memory Is a Permission Model
- Five pass-back questions · Fill-in card
Chapter 23 · Designing the Pass-Back
- Five pass-back questions: a rerouted invoice · Example
Chapter 23 · Designing the Pass-Back
- A pass-back checklist · Use it checklist
Chapter 23 · Designing the Pass-Back
- Convergence stages · Sequence map
Chapter 24 · AI That Knows When to Stop
- Convergence stages: the draft from the opening · Example
Chapter 24 · AI That Knows When to Stop
- Readiness check · Fill-in card
Chapter 24 · AI That Knows When to Stop
- A convergence audit · Use it checklist
Chapter 24 · AI That Knows When to Stop
Next part
About to give an agent more room to act?
Start with the checklist to spot what needs attention.
Run the self-checkBook a 30-minute call. Bring the agent and we’ll look at its job spec, permissions and pass-back.
Book a call