Chapter 6. A person as a system
Practicum page for Chapter 6 of When Intelligence Became Cheap. Version: 2026-06-11. Working Practicum address:
https://cheap-intelligence.vercel.app/.
Why this page exists
The same AI produces different results because people build different systems around it. A harness is an agent's working environment: rules, context, tools, a loop, memory, helpers, verification, a sandbox, and reusable skills. A personal harness applies this idea to one person's work and long-term goals.
This page turns the chapter into action. In one hour, you will have a map of your system with one visible gap you can address.
What to do in one hour
- Complete a Task-to-Outcome Map for the 3 main tasks in your week.
- Use the Personal Harness Builder. Its four steps give you a snapshot of your current system.
- Choose one repeatable skill from the Personal Harness Builder.
- For a layer-by-layer tool map, open Tools Map by Layer.
- For examples of harness levels, open Examples by Level.
The order matters. Map the tasks first, then build the harness. Without a Task-to-Outcome Map, you do not know which layer to build first.
Who this is for
| Role | Main question | Artifact |
|---|---|---|
| Specialist | Where does my time go to output that is getting cheaper, and where do I create real value? | Task-to-Outcome Map + Harness Builder |
| Manager | Which harness layers are weak, and how does that affect decision speed? | Harness Builder |
| Founder | How can I build a system that moves faster with a smaller team? | all three artifacts |
How to use this without fooling yourself
- Diagnose the system, not whether you are "good at using AI."
- The Knowledge layer does not mean "I have notes." It means "I can hand over useful context in 2 minutes."
- The Trust layer does not mean "people respect me." It means "people know that I checked the work."
- The Outcome layer does not mean "I did a good job." It means "this is what changed."
- If the map shows that 70% of your time goes to tasks in the "AI can do this" column, that is not a failure. It is your starting point.
If you want the technical background
You do not need these materials to use the chapter. They show how engineers use the word harness, not
as a clever prompt but as the system around a model.
- OpenAI: Harness engineering: leveraging Codex in an agent-first world. This is a clear example of how an agent's result depends on the working loop around it: tasks, local checks, review, and another iteration.
- Anthropic: Effective harnesses for long-running agents. This article covers a basic problem in long-running work: an agent needs memory and state to carry across sessions.
- Anthropic: Harness design for long-running application development. This is an application-development example based on task breakdown, artifacts, handoff, and verification.
- WalkingLabs: learn-harness-engineering. This technical course and repository is for readers who want to build a similar system themselves.
- OpenAI: Agent Skills in Codex. This source uses
skillsas a format for repeatable agent workflows. - Anthropic: Agent Skills. This source describes skills as modular agent capabilities loaded for a task.
A reader does not need to copy an engineering harness exactly. Carry over the principle instead. A model alone is not a system. Context, tools, memory, verification, a sandbox, skills, and a goal create the system.
Examples of harness levels
Fast-changing examples live in a separate map: Examples by Level.
It covers levels from an empty chat and Projects in ChatGPT or Claude to skills, local agents, Hermes, OpenClaw, and catalogs of agent templates. The map does not rank services. It helps you see which next step makes sense now.
Related materials
- Task-to-Outcome Map, a map of tasks and outcomes
- Personal Harness Builder, a 4-step worksheet
- Tools Map by Layer, tools organized by harness layer
- Examples by Level, examples from Projects to OpenClaw and Hermes
- Chapter 5: My usefulness level, a related diagnosis