Progression map
You do not learn AI in one jump. You upgrade the environment around your work. Each level unlocks the next capability only when the current one is visible, usable, and inspectable.
The agent checks what exists now: files, dashboard, rules, tools, and checks.
The agent names the missing piece that prevents the next upgrade.
The agent gives one concrete task that moves the system forward.
Level 0
State: You use ChatGPT or Claude in the browser.
Unlocks: Better questions and fast drafts.
Blocked by: The AI cannot see your actual work, and every session resets.
Pass condition: You can name the project, goal, and repeated context you keep retyping.
Next upgrade
Write one plain project note.
Level 1
State: Your important project context lives in files, not only chats.
Unlocks: The agent can read the project instead of being briefed from zero.
Blocked by: There is no short view of what matters now.
Pass condition: An agent can open a project file and summarise what it is, current state, and next step.
Next upgrade
Create one dashboard.
Level 2
State: You have a control surface for active work.
Unlocks: The agent can see priorities, active projects, idle projects, and next actions.
Blocked by: The agent still does not know your operating rules.
Pass condition: An agent can tell what is active, what is blocked, and what should happen next.
Next upgrade
Write agent rules.
Level 3
State: You have instructions for how AI should behave in your workspace.
Unlocks: Safer delegation and fewer repeated corrections.
Blocked by: The agent can understand the workspace, but cannot do much inside it.
Pass condition: A new agent session knows where to look, what not to touch, when to ask, and how to verify.
Next upgrade
Connect tools carefully.
Level 4
State: The agent can inspect useful surfaces: files, browser, GitHub, or other work tools.
Unlocks: AI can act, not just suggest.
Blocked by: Work can be done, but quality still depends on vibes.
Pass condition: The agent can make one small change, inspect the result, and explain what changed.
Next upgrade
Add checks.
Level 5
State: You define what good means before trusting output.
Unlocks: More reliable work and clearer human review.
Blocked by: The process still depends on you remembering the steps.
Pass condition: The agent can compare its output against a checklist, screenshot, examples, or human approval step.
Next upgrade
Turn repeated work into recipes.
Level 6
State: Repeated work has recipes the agent can follow.
Unlocks: Repeatability across projects and sessions.
Blocked by: The pieces exist, but the system is not yet self-assessing.
Pass condition: The agent can run a known process without you explaining it again.
Next upgrade
Let the agent assess the whole setup.
Level 7
State: Dashboard, files, rules, tools, memory, workflows, and checks work together.
Unlocks: Compounding AI work.
Blocked by: The system now needs maintenance, not more theory.
Pass condition: The agent can enter the workspace, report actual state, propose the next upgrade, do one task, verify it, and leave useful memory behind.
Next upgrade
Keep improving the harness after each failure.
The useful product is not a quiz. It is an inspection. The agent connects to the workspace, looks at what is actually there, and reports progress in plain language.
Current level: 2 - Dashboard
What works:
- Active projects are listed.
- Each project has a next step.
- Idle work is separated.
What blocks Level 3:
- No agent rules file.
- No instructions for editing, publishing, or verification.
Next upgrade:
Create a short agent rules file with five operating rules.
Levels make an abstract system concrete. A designer-founder does not need to understand every AI term on day one. They need to know what they can do now, what is missing, and what capability unlocks next.
The progression also makes AI less mystical. Chat is level zero. Files create memory. Rules create safer delegation. Tools create action. Checks create reliability. The harness is what happens when those pieces start working together.