5-stage map from chat user to AI-leader
Where 20+ years of experience start paying again, stage by stage.
If you are 20 years into a product, commercial or regulatory career and your AI use still starts with a blank chat window, you are standing on a stage of a path nobody showed you, comparing yourself to people 3 stages further and concluding you missed the train.
I started at stage 1, after 20+ years in commercial roles. Every stage above it turned out to be learnable. This letter is the map. It follows one person, me, through all 5 stages, with a look at each stage from the product management and the production and supply seat. Find your stage and take one move.
Stage 1: chat user
Every task starts in a blank chat. You re-explain your context every time, and the answers are useful in the moment and gone by tomorrow.
At this stage AI is an assistant: quick question, quick answer.
I felt I was a confident user of LLMs at that time and I used ChatGPT instead of Google: quick question, quick answer. And my questions were rather simple: find me the best running shoes for trails, or check my translation to German. I was satisfied with the experience but was annoyed that so many people described their experience with LLMs completely different. They talked about projects, memory, md files. I did not understand a word and honestly was ok with that: too much to do at work plus all hobbies plus children. Once I tried Claude and I was not impressed. The answer felt too generic and too neutral. I asked for an opinion and received a general well written answer, that pushed me to close the app and conclude that Claude is not for me.
What motivated me to come back was exactly that annoying question: why are other people so in love with Claude?
So I tried again. This time my first message was one sentence: I am an absolute beginner, without technical education, teach me.
Looking back, the generic answer that pushed me away had a simple cause: I gave the tool nothing to work with, so it could only answer in general.
The mistake here: treating every task as a fresh conversation. The re-explaining is the tax, and most people pay it for months without noticing it is optional. One chat per recurring topic already fixes most of it.
The moves: save the prompts that worked. Before asking, write one line on what good output looks like. Try Projects: it groups your chats by subject and cuts the need to re-explain context every time.
In your seat. In product management, this stage is pasting a competitor’s press release into the chat and asking for a summary before a meeting, then retyping your whole market context the next time you need one. In production and supply, it is pasting an email thread about a delayed shipment and asking what to answer the supplier.
Stage 2: structured collaborator
You reuse what worked. You can say what good output looks like before you ask.
At this stage AI is an analyst: it works on your documents and to your standards. You have spent a career telling good work from almost-good work, and this is the stage where that skill starts working on AI output.
The mistake here: collecting prompts instead of finishing one real task.
The moves: pick one recurring task from your real week. Write down the steps you repeat every time. Run it the same AI-supported way twice in a row, and note where the output missed your standard. But mapping your workflow first, you will be surprised to see that some steps are made on auto-pilot without even realizing that, or that something is even should and could be improved.
A real example from my own record: the sales trainer I built for my daughter’s small business. It drills customer objections, like “too expensive” and “we don’t need that.” It works because the structure is fixed and only the objection changes. That is stage 2 thinking packaged into a tool.
In your seat. For a product manager, this stage is running the same monthly competitor summary the same way twice, with a written instruction on what counts as relevant to your portfolio. For a P&S lead, it is handing over this month’s and last month’s delivery-performance table and asking for exceptions, using your own definition of what a material deviation is.
Stage 3: workflow builder
One piece of recurring work runs the same AI-supported way every week.
At this stage AI is a workflow: the repetition stops living in your head and starts living in the system.
What significantly improved my productivity are 2 basic things:
- E-mail triage twice daily with the simultaneous unsubscribe from spam e-mails and newsletters.
- Maintenance of all folders and desktop in a healthy clean state. That was always a huge energy drain, not being able to find the required document because you saved it under a strange name in a random folder. How I do it now: Claude uses one name pattern for all documents it generates, “Project name_status_date”. If those files are not in their respective folders but instead in downloads or at the desktop, a daily wrapup procedure moves them.
Once you are confident in the workflow and quality of the output, you can set triggers: either time-based (every day at 9:00 am) or event-based (every time I receive an e-mail from a customer).
The mistake here: adding a second workflow before the first survives a month.
The moves: connect AI to the files it needs, so you stop pasting them in by hand. Add the next workflow only when the first one runs without you thinking about it.
In your seat. In product management, this stage is a weekly market-signal brief that assembles itself from the same sources every Monday, filtered by your rules, waiting for you instead of being built by you. In production and supply, it is the weekly supply-risk review running on a trigger: the regional reports come in, your thresholds flag the exceptions, and a draft brief is ready before the meeting.
Stage 4: AI-enabled operator
AI is connected to your files, knowledge and rhythms. You manage systems, not chats.
At this stage AI is your operating model: the workflows have joined up, and your job has quietly changed from doing the work to reviewing it.
This is the target stage for an experienced professional, because it is the first stage where seniority starts paying again. Operators have spent careers reviewing work and retiring what stopped earning its place. Stage 4 is that exact job with a new workforce.
For me this stage is the system I described in an earlier letter as my AI chief of staff. From inside, it looks like this. The morning starts with a brief that assembles itself: what is waiting and what needs my decision. Ideas I capture by voice route themselves into an idea bank without me filing anything. The e-mail triage and the folder hygiene from stage 3 run as rhythms now, on their own schedule. Every working session ends with a wrapup that files the documents and writes the log, so the next session starts in under a minute instead of half an hour of remembering where I was. My own estimate, published earlier this summer, is that this saves me around 10 hours a week.
My job inside that system is the part that cannot be delegated: I read what it surfaces and I decide what ships.
The mistake here: measuring progress by how many tools you have tried. Progress at this stage is one system that runs a real piece of your work end to end.
The moves: review outputs on a schedule, and retire what you stopped using. Keep the judgment calls; delegate the repetition.
In your seat. In product management, this stage is the competitor monitor and the forecast commentary running as connected rhythms, while you spend your time on the decisions they surface. In production and supply, it is the S&OP preparation assembling itself across regions, so the meeting starts at the trade-offs instead of at the slide collection.
Stage 5: AI-enabled leader
Others adopt AI through paths you designed and standards you set.
At this stage you draw the map for other people. Every organization currently wants AI-fluent teams, and the person who has walked the path herself holds a map drawn from experience.
I am at the beginning of this stage myself, so I will keep it honest. My first adopter was my daughter: her business runs on paths I designed, a 50CHF voice bot that answers her calls and books into her calendar, a trainer that drills her sales objections. She never had to learn what an API key is. In public, the stage looks like the guides, the demo anyone can open and try, and this letter: documenting the path that worked is the first move of the stage, and you are reading me making it. At work it looks like my new title, a role whose whole job is this stage.
Being new here has one real benefit: I still remember exactly what stage 1 felt like, and the mistake below is the one I am most determined not to make.
The mistake here: leaders forgetting their own anxiety of 6 months ago. I wrote about this earlier this summer: everyone wants the team AI-native, and the honest question is whether a typical employee has been given any reason to adopt.
The moves: document the path that worked for you, including what good output means at each step. Then teach the stages themselves; any tool you name today will be replaced before the path is finished.
In your seat. In product management, this stage is writing the playbook your team uses to prepare launches with AI, with your standards written into it. In production and supply, it is mapping which planning steps stay human and which run on the system, and teaching your planners the stages, which outlast the tool of the month.
Where this leaves you
Will everything you spent decades becoming excellent at still matter? That is the question under all 5 stages, and the map’s answer is that experience is the only thing that transfers to every one of them. It becomes the standards at stage 2, the workflow rules at stage 3, the judgment at stage 4, and the paths you draw at stage 5.
Find your stage. Take one move this week.