In 2024 I stood in front of senior officials from more than 25 ministries of the Bangladesh government. Forest Department, primary education, the state broadcaster. Over the programme, 300+ officers came through those rooms.
Here is the confession first. I walked in expecting a hard day. I expected resistance, "we have procedures", the usual. What I got was quieter and more uncomfortable. A large share of the room had never opened a chatbot. And within the first half hour, several of them had drafted a departmental letter that would normally take half a day.
One officer looked up and said, half joking, "So what have we been doing all this time?"
I told him what I still believe: "Sir, you are not late. If you start today you are still in the first five percent of this country."
That day taught me something the corporate rooms had hidden. The gap between businesses is not which AI tool they bought. It is what level the AI is operating at inside the business. I now put every company I meet on a four-level scale, and I have not found one that does not fit.
The four levels
- Level 0. You have heard of AI. Somebody on the team "uses ChatGPT". Nothing in the business changed.
- Level 1. You paste a problem into a chat window and copy the answer back out. Emails, summaries, a caption. The work is the same; you have a faster search bar.
- Level 2. You have saved prompts, briefs and workflows. The same job runs the same way every time, and a colleague can run it without you.
- Level 3. An agent takes a task, decides its own steps, uses tools and finishes while you sleep. You see the result, not the process.
Most people who tell me they are "using AI" are describing Level 1. That is not AI in your business. That is you doing the same work with a faster typist.
Why Level 1 is a trap
Level 1 feels like progress because it is fast. But nothing accumulates. The prompt you wrote yesterday is gone. The tone your colleague got is different from yours. Nobody can hand the work over, because the work lives in a chat history.
The test I give every room: AI has not saved you any time until it can finish something without asking you a question. If every output still needs you to read it, fix it and paste it somewhere, you have moved the typing, not the job.
What Level 2 looks like on a Tuesday
Level 2 is where the return starts, and it is not glamorous. In my own company the first Level 2 win was a saved brief for replying to course enquiries: role, context, the price sheet, three rules on tone, a stop condition. Anyone on the team pastes it, adds the message, and the reply comes out in our voice. Same input, same output, no matter who is on shift.
That is the whole difference. A Level 1 output belongs to a person. A Level 2 output belongs to the business.
The officers who impressed me most in those ministry rooms went straight here. After the sessions, several built small knowledge bases from their own circulars and reports, so that a question returns a cited answer instead of a long hunt through files. Not magic. Management.
The Level 3 test
Level 3 is the one everybody wants to talk about and the one almost nobody has. Here is the test I use, and I use it on my own operations before I use it on anyone else's.
- Name one task that happens the same way every week.
- Write the trigger: when X happens.
- Write the steps: then Y, then Z.
- Decide where a human must look: money, a promise to a customer, anything legal.
- Let it run for a week and read the log, not the confirmation toast.
If you cannot write step 3, you are not ready for Level 3, and that is fine. Go back to Level 2 and save the brief. If you can write all five, you have an automation, and you are the person who designed it. More than 50 of these run my businesses today. Not one of them started as a tool. Each one started as a written step list.
The tool will change again next quarter
This is the part I say at the start of every session. The tool changes every quarter. The levels do not. A company at Level 2 with a boring tool beats a company at Level 1 with the newest model, every time, because the first one has a system and the second one has a subscription.
So the question is not "which AI should we buy". The question is "which level are we honestly at, and what is the one thing that moves us up one level".
Practice
What to do on Monday
- Ask five people in your team how they used AI last week. Place each answer on the four levels. Do not argue with the answers.
- Take the most common Level 1 task and write it as a brief: role, context, rules, output format. Save it where the team can find it.
- Run the saved brief three times with three different people. If the outputs match, you have your first Level 2 job.
- Write the "when X, then Y, then Z" for one weekly task and put it on your wall, even if you build nothing yet.
Where is your company honestly sitting today, and what is the one job that would move it up a level?