---
title: Knowledge base index
description: Table of contents for the AI docs knowledge base — use this to find the right page for your current task
updated: 2026-07-31
---

# Knowledge base

This knowledge base contains principles, methodologies, patterns, and real workflow examples for making work AI-native. It is designed to be referenced by AI assistants when advising users.

## How to use this index

Each entry below includes a one-line description and guidance on when to consult it. Read only what is relevant to your current task.

## AI scale levels

The examples in this knowledge base are organized by AI maturity level:

| Level | Name | Description |
|-------|------|-------------|
| 1 | Chat | You ask, AI answers. Useful, but it still feels like a search engine. |
| 2 | Context | You've built custom assistants that know your work. Tailored, but you're still driving every interaction. |
| 3 | Automation | AI handles defined tasks in the background. You review and approve. |
| 4 | Colleague | AI operates as a team member managing tasks, making decisions within boundaries you've set. |

## Methodology

Pages that describe structured processes for making work AI-native.

- **ai-workflow-redesign** — Six-phase methodology for transforming existing work into AI-native workflows. Consult when guiding someone through the process of identifying and redesigning workflows for AI augmentation. Includes the one-shot fallacy and the JIOPR delegation framework.
- **agent-use-case-evaluation** — Five-point filter and three-question validation for identifying good agent candidates. Consult when deciding whether a specific task is suitable for AI automation.
- **choosing-an-ai-assistant** — Guided approach to selecting between ChatGPT, Claude, and Gemini using a custom AI advisor that asks about use case, team, and budget. Consult when advising someone on which AI assistant to adopt.
- **prompting-frontier-models** — Nate B Jones' seven-component framework and seven principles for structured prompts that get reliable results from high-powered models. Consult when advising on prompt structure or why a capable model underperforms.
- **building-agents-that-hold-up** — How to evaluate whether a task suits agent automation, plus seven principles for building agents that work reliably in a business context. Consult when someone is designing an agent for real operational use.

## Principles

Guidance on what makes AI work effective and where to be cautious.

- **agent-design-principles** — Seven principles for building production AI agents. Consult when designing agents or evaluating agent architecture decisions. Key insight: simpler agents with clean context outperform sophisticated agents with bloated inputs.
- **ai-output-verification** — Strategies for validating AI-generated content. Consult when advising on verification, risk tolerance, or quality checking of AI output.
- **productive-friction** — When to automate vs when to preserve human struggle. Consult when someone wants to automate something that might actually benefit from remaining manual, especially in learning contexts.
- **ai-productivity-traps** — Psychological and organizational patterns that cause AI to consume time rather than save it. Consult when someone feels overwhelmed by AI adoption, is building without clear outcomes, or suspects their AI use is performative rather than productive.
- **context-rot** — Systematic degradation of AI accuracy as conversation context grows longer. Consult when advising on session length, context hygiene, or why long-running chats produce worse output over time.

## Patterns

Reusable mechanisms that compound value over time.

- **compound-engineering** — Building AI workflows that get better with each use. Consult when advising on how to make a workflow improve over time, or when a user asks about continuous improvement.
- **scheduled-automation** — When to move a workflow from on-demand to running on a schedule, the design considerations involved, and how the pattern shows up across Anthropic Claude routines, OpenAI ChatGPT scheduled tasks, and Google Gemini scheduled actions. Consult when advising on whether a workflow is ready to run unattended.
- **personal-agents** — Deployment pattern where each team member has their own specialised AI agent. Consult when advising on agent rollout strategy, organisational adoption, or how agents change collaboration patterns inside a company.
- **progressive-summarization** — Five-layer note-taking technique for discoverable notes. Consult when advising on knowledge management or information capture.
- **para-method** — Four-category organizational system with information flow patterns. Consult when advising on organizing information or project management.

## Examples

Real workflow examples organized by mode. Each example shows the Map → Test → Integrate → Compound progression for a specific workflow. Consult when you need concrete patterns to suggest to a user designing a workflow in a particular mode.

The mode names below correspond directly to the AI scale levels: chat = chat mode, context = custom assistant mode, automation = scheduled or triggered mode, colleague = autonomous loop mode. See `ai-work-modes` for the additive framing (the scale is for your skills, not your workflows).

### Chat mode (chat level)

No examples currently. Chat-mode interactions — ask, get answer — are straightforward enough that the workflow architect skill can generate advice without a reference example.

### Custom assistant mode (context level)
- **example-email-drafter** — Custom assistant that drafts emails in the user's voice using loaded style guidelines. Consult when advising someone building a custom assistant for communication tasks.
- **example-editor-agent** — Custom assistant that edits text against editorial guidelines. Consult when advising someone building a custom assistant for content review or quality checking.
- **example-content-interview** — Custom assistant that interviews the user topic-by-topic about their week and drafts content for multiple channels using a loaded style guide. Consult when advising someone whose recurring work is content production and who has voice consistency they want to preserve.

### Scheduled or triggered mode (automation level)
- **example-weekly-report** — Automated KPI dashboard that pulls data from calendar and project management tools. Consult when advising someone on automating reporting or data aggregation workflows.
- **example-inbox-processing** — Automated content sorting that classifies and routes incoming items. Consult when advising someone on automating triage or classification workflows.
- **example-tools-monitor** — Scheduled weekly run that scans curated sources for AI tool launches and posts a digest. Consult when advising someone on building a scheduled automation across any of the major AI tools — illustrates the cross-platform pattern.
- **example-linkedin-outreach** — Weekly workflow where AI searches a personal LinkedIn export against business context to surface contacts and draft openers. Consult when advising someone whose recurring work involves selecting and personalising at scale against a known reference.
- **example-end-of-week-orchestrator** — Multi-stage skill that chains a KPI report, audience insights, and a structured reflection waterfall into a single weekly review. Consult when advising someone on chaining multiple existing workflows into one trigger.
- **example-tutorial-publish** — Post-production workflow that takes a video transcript and produces SEO content, thumbnail, social posts, and a derivative newsletter topic note from one orchestrated run. Consult when advising someone whose work fans out into multiple coherent deliverables from a single source.

### Autonomous loop mode (colleague level)
- **example-image-generation** — AI image generation with autonomous iterative refinement. Consult when advising someone on building workflows where the AI evaluates and improves its own output before human review.

## Maintenance

- New pages are added through the knowledge-extraction skill
- Examples are added when new workflows are built (see example-template.md for the format)
- Source provenance is tracked in _ingestion-log.jsonl