The Superintelligence Ladder

Six rungs. You climb one at a time. Each rung is a small step from the one below. Most admin people in 2026 sit at Level 1 or 2. The highest rung is rare but reachable, and it is changing what one person can do at work.

Read each level. Find yourself honestly. Read your next move. Take it this week.

Level 1 · Asking

You use Claude or ChatGPT as a search engine you can talk to. One question, one answer. Nothing remembers anything. Nothing connects to anything.

You are here if:

• You have used a chat AI a few times for personal tasks
• You have not used AI for a real work deliverable yet
• You assume AI is a fancier Google

Typical gain: zero to ten percent on routine tasks. Output quality is variable and often disappointing because the AI has no idea who you are or what good looks like for you.

Your next move: Pick one task you do every week. Try Claude on it once. Compare the output to your usual. Save the prompt if it helped.

Level 2 · Drafting

You use Claude in your daily work. You have three to five saved prompts. You paste in data, you get drafts back. Output is rough but useful. You correct it before sending.

You are here if:

• You use AI three or more times a week for real work
• You have favorite prompts you reuse
• You still feel like the AI is guessing at what you actually want

Typical gain: twenty to forty percent on writing tasks. Quality is improving but inconsistent.

Your next move: Do the Memory Bootstrap. Tell Claude who you are, how you work, what you do not want it to do. Outputs change overnight.

Level 3 · Personalized

Claude has memory of who you are, how you work, and what matters. Outputs sound like you would have written them. You have connected one or two tools (Outlook, Teams, SharePoint) so Claude can read live context. Most days, the first draft is publishable with light edits.

You are here if:

• Memory is set up and you trust it
• At least one MCP connector is in place, read-only
• Your colleagues notice the writing feels like yours, not like AI

Typical gain: fifty to seventy percent on routine work. Quality matches your average effort, every time.

Your next move: Install one plugin for your function. Pick the skills that match the tasks you do most. Let them auto-trigger. Stop typing out the same instructions.

Level 4 · Equipped

You have plugins installed for your function. Skills auto-trigger on the right tasks. A quality gate catches obvious slop before it leaves your desk. When a hard decision lands, you ask a panel of perspectives before you commit. Output consistently matches your best work, not your average work.

You are here if:

• Three or more skills auto-trigger correctly on different tasks
• You have a quality gate that has actually caught something
• You have used a panel pattern at least once for a real decision

Typical gain: sixty to eighty percent on routine work. Quality at your best, every time. Hard decisions feel less alone.

Your next move: Stand up your first scheduled task. Pick the one piece of information you wish appeared on your screen before you sit down. Morning brief is the usual entry point.

Level 5 · Operating

Scheduled tasks run while you sleep. Morning brief, weekly market scan, regulatory radar, monthly retro. Self-evolving loops adjust your prompts based on what works. Memory grows automatically each week. A board-style panel rehearses your hard decisions before the real meeting. The system works when you are not working.

You are here if:

• Three or more scheduled tasks run reliably without your attention
• A monthly retro has actually changed how you work
• You have caught yourself thinking "the system handled that for me"

Typical gain: seventy to ninety percent on routine work. Plus you have capabilities you did not have before (deep market scanning, board rehearsal, regulatory tracking) that did not exist for a one-person operation.

Your next move: Find one colleague who is at Level 2. Hand them this ladder and your Setup Zero. Get them to Level 3 this month. The transition begins.

Level 6 · Autonomous

The system acts on goals, not on tasks. Given an outcome, it plans, executes, and only escalates the judgment calls. Subagents collaborate across days. Memory is shared across the function. The AI is no longer waiting for prompts. It is the operating layer underneath the work.

You are here if:

• Goal-level interfaces are normal ("achieve X by Y date") rather than task-level
• Subagents coordinate without you orchestrating each step
• Escalation rules are documented and the system honors them
• New work types get added to the operating layer, not just to a todo list

Typical gain: measurement shifts. Hours saved is no longer the meaningful unit. The unit is now goals achieved per person per month, errors caught before they shipped, decisions made better because the prep work was done before the meeting started.

Your next move: there is no Level 7 today. The frontier beyond Level 6 will be drawn when this rung is climbed.

How to use this ladder

• Self-assessment. Read each level honestly. The point is not to climb fast, it is to climb soundly.
• Skip nothing. Each rung makes the next one possible. People who try to skip from Level 2 to Level 5 fall back to Level 1.
• Bring colleagues with you. Level 5 alone is impressive. Level 5 across a team is transformative.
• Re-check every quarter. You will move. The ladder will move. The rungs may need to be redrawn next year.