Framework

The AI Adoption Ladder

The AI Adoption Ladder is a four-level model for how far a team has actually got with AI: L1 Chat, L2 Systematisation, L3 Automation and L4 AI-native. Most teams believe they are “doing AI” because they use ChatGPT. That is Level 1 — and there are three levels above it.

Developed by Michal Juhas, AI implementation consultant in Bratislava, Slovakia, across 30+ AI projects. Used as the opening exercise in every workshop.

L1

Chat

People ask ChatGPT, Gemini or Claude a question and paste the answer back into their work.

This is where most departments are today. AI is a faster search box. Nothing is captured, nothing is shared, and the quality of the output depends entirely on who is typing.

You are here if

  • Everyone uses their own account, often a personal one
  • Nobody can tell you which prompts work
  • The same problem gets solved from scratch every time
  • No visibility into what data is being pasted where

Moving up

Write down the three prompts your team actually reuses and put them somewhere shared. That single act is the start of Level 2.

L2

Systematisation

Procedures stop living in one person’s head. Custom GPTs, saved prompts and a knowledge base the whole team can see.

The work becomes repeatable. A new joiner can produce the same quality output in week one as someone who has been there a year, because the method is written down rather than remembered.

You are here if

  • Shared prompt library or custom GPTs the team actually opens
  • A knowledge base the assistant can read from
  • Output quality no longer depends on who ran it
  • You can point at a process and say how AI is used in it

Moving up

Find the systematised step that runs most often and ask what would have to be true for it to run without anyone starting it.

L3

Automation

The procedure runs in the background and nobody has to trigger it. A person approves the result, not every step.

This is where the hours actually come back. The work happens on a schedule or on an event, and a human moves from doing the task to reviewing its output.

You are here if

  • Workflows fire on a schedule or a trigger, not a person
  • Humans review outcomes rather than performing steps
  • Systems are connected — data stops being retyped between tools
  • You can measure hours saved per week, not per demo

Moving up

Stop asking how to speed up the existing process. Ask what the process would look like if you designed it today, knowing agents can do most of it.

L4

AI-native

The process was not accelerated — it was redesigned. A substantial part of the work is done by agents.

The shape of the work changes. Instead of four people doing a process faster, one person supervises agents doing a process that could not have been staffed before. This is where the compounding advantage is.

You are here if

  • The process would not make sense without AI in it
  • Headcount is deployed on judgement, not throughput
  • Volume can grow ten-fold without proportional cost
  • New offerings exist that were previously uneconomic

Moving up

This is the target. The question here is which process to redesign next, not whether to.

The part most companies skip

TWO QUESTIONS WORTH MORE THAN THE TOOLING

Who takes the time you save?

If a task took ten hours and now takes two, somebody decides what happens to the other eight. Either the employee keeps them, or a supplier resells them to another client, or you convert them into something deliberately. In most companies nobody has decided yet.

Cutting costs is a one-time win

You can save once. “We have four accountants, let's go to two” — and that is the end of it. The far more interesting question is what you would want ten times more of. Enquiries? Proposals? Content? That is where growth compounds.

FAQ

COMMON QUESTIONS

What is the AI adoption ladder?

A four-level model for describing how far a team has actually got with AI. L1 Chat: people ask a chatbot and paste answers back into their work. L2 Systematisation: prompts and procedures become shared assets. L3 Automation: workflows run without anyone starting them. L4 AI-native: the process is redesigned so agents do a substantial part of it.

What level are most companies on?

Most departments are on Level 1. Software engineering is usually two or three levels ahead of the rest of the business — developers work with AI daily while sales, marketing, finance and operations are still using a plain chat window.

Do you have to go through every level in order?

In practice yes. Automation built on procedures that were never written down tends to automate the wrong thing, and redesigning a process you have not yet systematised means redesigning something nobody can describe.

Why is cutting costs the wrong goal?

Cost cutting is a one-time win. You can save a salary once, and then it is over. The more interesting question is what you would want ten times more of — enquiries, proposals, content — because that is the growth that compounds.

Who takes the time that AI saves?

If a task took ten hours and now takes two, somebody decides what happens to the other eight. Either the employee keeps them, or a supplier resells them to another client, or you deliberately convert them into something. In most companies nobody has decided yet.

How do I find out which level my team is on?

Read the signals listed under each level above and find the highest one where every signal is true of your team. If you want an outside read, Michal Juhas runs this assessment as part of workshops, on-site across Slovakia, Czechia, Poland and Austria, or remotely.

Which level is your team on?

The first thing we do in a workshop is find out — honestly, per department. Book a short call and we'll sanity-check it together.

Book a Call