Levels of practical AI

From copy and paste to autopilot.

Eleven levels, four phases. Every level multiplies what one person can do by about 1.6, and every level changes what the human does: from doing the work, to running the systems, to checking the work, to setting the course.

Indicative model. Real gains depend on the process and the team.

Phase 1 · Levels 0–3

Assistant · the author

You do the work, AI helps.

0×1 output

No AI

Everything by hand: searching, writing, copying, checking.

A proposal written from a blank page.

Youdo all of it

How to reach level 1+

Focus: One assistant, one daily chore. The goal isn’t to be impressed, it’s to find the first task AI does well enough for you.

→Pick one assistant (ChatGPT, Claude, Gemini or Copilot) and keep it open for a week
→Give it one daily chore: an email reply, a summary, a first draft
→Notice what it gets wrong as well as what it gets right

Watch out for: Trying it once on something hard, getting a weak answer, and deciding it isn’t for you.

You’re at level 1 when you reach for an assistant a few times a week without having to remember to.

1×1.6 output

Copy and paste

A basic chat window. Paste a question in, copy the answer out. Every chat starts from zero.

An email pasted in to sound more polite.

Youcarry every word in and out

How to reach level 2+

Focus: Better input, not more questions. Tell AI who it’s for, what the goal is and what good looks like, and attach the real material.

→Give context: who it’s for, the goal, and what good looks like
→Attach examples and files instead of describing them
→Save the three prompts you reuse most, with their context

Watch out for: Short, bare prompts that get generic answers, and then fixing those answers by hand.

You’re at level 2 when you reuse saved prompts every week and the first draft needs only light edits.

2×2.5 output

Better prompts

Context, role, examples and files go into the prompt. Prompts that work get saved and reused.

A saved prompt that turns call notes into a summary.

Youwrite the prompt, fix the output

How to reach level 3+

Focus: Stop repeating yourself. Put your templates, prices, tone and examples somewhere AI remembers them.

→Create a project, custom GPT or Gem for your main area of work
→Load it with your templates, prices, tone rules and past examples
→Write standing instructions once, so you stop repeating them

Watch out for: A notes file full of good prompts that you still paste in by hand, one chat at a time.

You’re at level 3 when the project or custom assistant for your main work already knows the context before you ask.

3×4 output

Personal assistant

Projects with company knowledge and standing instructions. AI knows the context before you ask.

A project that knows your prices, tone and templates.

Youstart every task by hand

How to reach level 4+

Focus: Fewer windows. Use the AI inside your email, documents, meetings and CRM, where the work already happens.

→Switch on the AI inside your email, documents and meetings
→Let a notetaker write meeting notes and follow-ups
→Start each task where the work lives, not in another window

Watch out for: A brilliant assistant in one tab, and a day still spent copying between it and everything else.

You’re at level 4 when replies, documents and meeting notes get their first draft inside the tool, with no copy and paste.

Phase 2 · Levels 4–6

Connected · the operator

You run the systems.

4×6 output

AI in your tools

AI built into email, documents, meetings, the CRM and the code editor. No more copying between windows.

Meeting notes and follow-ups written as the call ends.

Youtrigger it where you work

How to reach level 5+

Focus: From answers to actions. Let AI read and update a real system (CRM, drive, calendar), with you approving what it changes.

→Connect your assistant to one real system: CRM, drive or calendar
→Ask it to read and update records, not only to answer questions
→Decide what it may change alone and what needs your approval

Watch out for: Using every built-in AI button but never letting AI touch real data, so every task still ends with you.

You’re at level 5 when you ask AI to update records or files in a real system every week and it does, with your approval.

5×10 output

Connected to data

Through MCP and APIs, AI reads and writes your real systems: CRM, drive, calendar, databases. It acts, not only answers.

“Update every deal we discussed this week.” Done.

Youapprove what it changes

How to reach level 6+

Focus: Take yourself out of the trigger. Pick a process that always starts the same way and let it start on its own.

→Pick one repeating process with a clear trigger: a new email, form or file
→Build it as a workflow in n8n, Make or Zapier with one AI step
→Add error handling, so nothing breaks quietly

Watch out for: Waiting for a free week to automate, while doing the same five-minute task forty times a month.

You’re at level 6 when at least one process runs on its own every week: a trigger fires, AI does its step, and you only see the exceptions.

6×16 output

Automated workflows

Processes start on their own. A trigger fires, then AI extracts, classifies, drafts and routes. Nobody presses a button.

Every inbound quote read, priced and answered in minutes.

Youdesign the flow, handle exceptions

How to reach level 7+

Focus: From steps to goals. Describe the outcome and what done looks like, and let an agent plan the steps.

→Give an agent a goal instead of steps: research, a document, a piece of code
→Hand it the tools it needs and a clear definition of done
→Review the result, then widen what it may do on its own

Watch out for: Wiring every step by hand, so each new job means building another workflow.

You’re at level 7 when you give an agent a goal every week and it hands back finished work that you only review.

Phase 3 · Levels 7–9

Autonomous · the reviewer

You check the work.

7×25 output

Agents

You give a goal, not steps. An agent plans, uses tools and hands back finished work: research, code, documents.

An agent researches a prospect and drafts the outreach.

Youset the goal, review the result

How to reach level 8+

Focus: From one agent to a crew. Split the work into independent tasks, and add a check before anything reaches you.

→Split a big job into independent tasks
→Run two or three agents in parallel, each on its own task
→Add a verification step that checks their work before you do

Watch out for: Supervising one agent at a time, so your review time becomes the ceiling.

You’re at level 8 when several agents work at once every week and a check step catches their mistakes before you do.

8×40 output

Parallel agents

Several specialised agents work at once, each on its own task, while a verification agent checks their output.

Five coding agents close five tickets in one afternoon.

Youassign the work, merge the results

How to reach level 9+

Focus: Stop being the dispatcher and the checker: work comes in from a queue, automated tests judge it, and a report comes out every morning.

→Turn your manual checks into automated tests and quality bars
→Let the system pick up work from a queue or ticket list on its own
→Get a report every morning of what was done and what failed

Watch out for: Assigning every task and reviewing every result yourself, even though the agents are faster than you.

You’re at level 9 when work is picked up from a queue, checked automatically and reported back without you starting it.

9×63 output

Self-running system

The system finds its own work, tests itself, fixes what fails and reports. Automated checks replace manual ones.

Tests fail overnight. By morning there is a ticket and a fix.

Youset the quality bar, review exceptions

How to reach level 10+

Focus: Guardrails over tasks. Set the goals, the quality bar and the risk line, then review samples and exceptions.

→Move whole functions onto the system, not single tasks
→Set goals, guardrails and a risk line for what needs a human
→Review samples and exceptions instead of every output

Watch out for: Automating tasks one by one inside a function that is still designed around people doing them.

You’re at level 10 when a whole function runs on the system and people review only samples, exceptions and decisions above the risk line.

Phase 4 · Level 10

Autopilot · the navigator

You set the course.

10×100 output

Autopilot

AI runs the operation. The system finds the work, does it, checks it and reports back, day and night.

A whole function runs on its own; people review samples and exceptions.

Youset the course, decide, orchestrate

How to hold level 10+

Focus: Keep raising the quality bar, move the risk line on purpose, and spread the model to the teams furthest behind.

→Keep raising the quality bar and moving the risk line deliberately
→Spread the model to the next team or function
→Teach others to navigate: people are now the bottleneck

Watch out for: A system nobody else can steer, next to teams still working at level one or two.

You’re holding level 10 when the system keeps improving without you and the next team has started its own climb.

How the human role changes

Your time moves from doing to directing.

The hours don’t disappear. They move up: from producing the work, to checking it, to deciding what work should exist at all.

You todayLevel 0 · No AI · Author

90%

At level 3Level 3 · Personal assistant · Author

70%

18%

12%

At level 5Level 5 · Connected to data · Operator

48%

31%

21%

At level 7Level 7 · Agents · Reviewer

25%

43%

32%

At level 10Level 10 · Autopilot · Navigator

40%

60%

Doing the workReviewing resultsDeciding and orchestrating

Share of working time per level. Indicative model.

Find your level. Then climb one at a time.

Take the test→