People14 Nov 2026

The seven levels of using AI at work

By Alan Goh · 3 min read

Using AI at work

Which level is your team on?

  • Seven levels, from ask to evolve
  • Levels 1 to 3 are things you do
  • Levels 4 to 7 are things you build
  • The move that matters is 3 to 4
Short answer

AI use at work grows through seven levels: ask, think, produce, systemize, replicate, govern, evolve. The first three are things you do, the next four are things you build, and most teams are on the first three. The useful next step is moving one task that worked from level 3 to level 4, so it runs again next month without being rebuilt.

Ask around your office this week and you will find AI already in use. Someone asks it questions instead of searching. Someone pastes a customer email in and asks for a reply. Someone drafted a quotation with it. So the question “are we using AI?” gets an honest yes, and the yes tells you nothing about what to do next.

In my classes I draw a ladder with seven levels. I designed it myself, from my own work and from what I see in training rooms, and I say so when I teach it. It is useful because it turns “are we using AI?” into a better question: which level are we on, and what does the next one look like?

  1. Ask. Using AI like a library or a search engine.
  2. Think. Brainstorming, arguing a decision out with it.
  3. Produce. Turning the decision into an actual file: a document, a schedule, a draft.
  4. Systemize. The same piece of work runs again next month without being rebuilt.
  5. Replicate. Someone else can run it, because the instructions are written down.
  6. Govern. Checks around it, so it cannot go wrong quietly.
  7. Evolve. It notices something has changed and improves itself.

Levels 1 to 3 are things you do. Levels 4 to 7 are things you build. That line matters, because everything on the doing side lives inside one person, and everything on the building side belongs to the company. When the person who is good with AI goes on leave, the doing stops. A built process keeps going.

The pancake test

Think of teaching someone to cook. Making a good pancake is level 3: you did the thing. Writing the recipe so it comes out right every time is level 4. Handing the recipe to someone else is level 5. Level 6 is tasting what they make now and then, because if you never check, two months later they will be making roti canai from your pancake recipe. Level 7 is the recipe itself noticing that tastes have changed.

Every rung has a plain household shape like that. Climbing the ladder takes method and patience. It does not take a technical department.

Finding your team’s level

Take the last piece of AI work your team was pleased with, and ask one question: if the person who did it left tomorrow, would it still happen next month? When the answer is yes, you are at level 4 or above. A result that leaves with the person is level 3, however impressive it looked.

Most teams I meet are on levels 1 to 3. That is where everyone starts, and it is already worth having. The move that changes the business is the one from 3 to 4, when one task that worked becomes a process the team owns.

Try it this week

Pick the one AI result your team liked most this month. Sit with the person who produced it and write down, in order, what they actually did: where the inputs come from, what they ask for, what they check before using the answer. That written page is the start of level 4. Next month, let someone else run the task from the page, and see what the page forgot to say.

In your department

  • Strategy & Planning: The ladder turns an AI plan into something you can sequence: place each department on its level, pick the one task worth moving from level 3 to level 4 this quarter, and review whether it climbed.
  • HR & Admin: Training requests get sharper when they name a rung. Rather than asking for “an AI course”, ask for everyone comfortable at levels 1 to 3, and one routine process your department repeats every month written down and running at level 4.

Want to try this with your team?

Tell me what your team is working on, and I'll suggest where to start. I read and reply to every message myself.

More notes

  • 02 Nov 2026

    AI can read the screenshots your customers send

    Feedback that arrives as phone screenshots can go to AI exactly as it is. AI reads the images, sorts faults from suggestions, and a person decides what changes. In one real round, three fixes were running on a test copy the same morning.

    Read it →
  • 12 Oct 2026

    Fix it before you automate it

    Automation runs whatever version of the task it is given. A quick test tells you whether a task is ready — and getting it ready is usually smaller work than people expect.

    Read it →

All the notes →