What AI Changes (and Doesn't Change) at Work

Between the headlines that say AI will take every job and the colleagues who say it is a toy, there is a much more useful middle ground. In this lesson you will learn to look at your job as a bundle of tasks, see where AI genuinely helps and where it gets in the way, and start the personal file that will become your adoption plan.

You already know how to write a structured prompt and how to check an answer. This course takes those skills into the real world: your inbox, your spreadsheets, your meetings, your customers. Before touching a single tool, we need a clear mental model of what AI actually changes at work.

Meet the people you will follow

To keep things concrete, four people will accompany you throughout the course. They are fictional, and any resemblance to real people is a coincidence.

• Camille, marketing manager in a small company that sells office furniture.
• Marc, a self-employed plumber who runs his business alone, with an occasional apprentice.
• Nadia, HR officer in a company of about 80 employees.
• Julien, an accountant in a small accounting firm.

You will see them make good decisions, and a few mistakes that are very common. You will also build, chapter after chapter, your own AI adoption plan: a short file with your task inventory, your key prompts, your usage rules, your indicators and a 30/60/90-day plan.

Jobs are bundles of tasks

When people say "AI will replace my job", they are treating a job as one single block. In reality, a job is a bundle of tasks of very different natures. Take Camille. In one week she writes a newsletter, answers suppliers, prepares a presentation for the management committee, negotiates a price with an agency, coaches a junior colleague, reviews a spreadsheet of campaign results, and decides which event to sponsor.

Some of those tasks are mostly about producing text from known information. Others are about judgment, relationships, accountability or physical presence. Generative AI is good at the first kind and much weaker at the second. That is why the useful question is never "Will AI do my job?" but "Which of my tasks can AI help with, and how?"

Think of it like a power tool. A cordless drill did not replace carpenters; it changed which parts of the work are slow and which are fast, and it rewarded people who learned to use it safely. AI is a tool of the same family, though one that speaks, which makes it easier to over-trust.

Where AI helps well

AI assistants tend to be useful when the task has most of these characteristics:

• It is made of language or structured information: drafting, rephrasing, summarizing, classifying, translating, extracting.
• A first draft is valuable: starting from a blank page is the slow part, and you can improve a draft quickly.
• It is repetitive with variations: similar emails, similar reports, similar customer questions.
• You can check the result: you know the subject well enough to spot an error in a minute.
• The cost of a small mistake is low, or a mistake will be caught before it matters.

Where AI helps badly, or dangerously

• Tasks needing exact facts you cannot check: legal references, prices, technical standards, figures.
• Decisions about people or money where an unexplained or biased answer could hurt someone.
• Anything involving confidential data in a tool your organization has not approved.
• Tasks that depend on being there: reading a room, calming an upset customer, inspecting a boiler.
• Situations where accountability must be clear: signatures, certifications, official statements.

Notice that this list is about the task and its context, not about the person or the sector. The same person can use AI wisely for one task in the morning and should not use it at all for another in the afternoon.

Complementarity: who does what

The most productive way to think about AI at work is as a division of labor between you and the tool.

The AI is good at	You are essential for
Producing a fast first draft	Knowing what the draft is for and who will read it
Reformulating, shortening, adapting a tone	Deciding what you really want to say
Listing options and angles	Choosing, with context the AI does not have
Never getting tired of repetitive formats	Spotting what is wrong, odd or risky
Working at any hour	Taking responsibility for the result

The last row matters most. Responsibility stays with you. If an AI-written email contains a wrong delivery date, your customer will not blame the tool; they will blame you.

Two myths to leave behind

Myth 1: "It will replace everything, so why bother?"

Nobody can predict precisely how occupations will evolve, and you should be wary of anyone who sells certainty, optimistic or pessimistic. What is observable today is simpler: tasks are changing faster than jobs. People who understand which tasks to delegate, and how to supervise the result, are in a stronger position than those who ignore the tool or those who use it blindly.

Myth 2: "It's useless, it makes things up."

It is true that AI makes mistakes, as you saw in the prompting course. But an unreliable tool can still be valuable if you use it on the right tasks and check the output. A calculator that sometimes mistypes digits would be dangerous for accounting, yet a colleague who proposes ten possible subject lines for a newsletter is useful even if three are bad. Fit the task to the tool's profile.

A first case: Camille's Monday

Camille spends her Monday mornings answering about fifteen emails from suppliers and customers. Most are similar: delivery dates, quotes, small complaints. She starts to use an approved assistant to produce draft replies, but she makes a rule for herself: the AI writes the draft; I read it, correct the facts and send it. After a few weeks she notices she is faster on routine messages and has more energy for the three emails that really need thought. She did not "automate her job"; she changed the way one task is done.

Compare this with an imaginary colleague who pastes a customer's full complaint, with name, address and order number, into a free personal chatbot account, and sends back the answer without reading it. Same tool, same task, very different risk.

A ready-to-use prompt: your first reflection

You can use an assistant to help you think about your own work. Use only general descriptions, never confidential details.

You are a work organization coach. I work as [job title] in [type of organization, e.g. a small company in retail].
Here is a list of my usual tasks in a typical week, described in general terms:
[list 8 to 12 tasks]

For each task, tell me in a table:
1. Whether it is mostly language/information work, judgment/relationship work, or physical/on-site work.
2. How AI could realistically help (draft, summarize, classify, brainstorm, none).
3. The main risk of using AI for it (accuracy, confidentiality, fairness, accountability).
Do not invent details about my company. If you lack information, ask me up to 3 questions first.

Bad use versus good use

Bad use	Good use
"Write my quarterly report" with no context, then send what comes out.	Give the objectives and your notes, ask for a structure and a draft, check every figure, rewrite in your own words where needed.
Refusing to try anything "because it makes mistakes".	Testing on low-stakes tasks with a verification step, then deciding on facts.
Using a personal account for work documents because it is quicker.	Using the tools your organization has approved, or asking for approval first.

Your turn: split your own job into tasks

Exercise. Write down ten tasks you did in the last two weeks, as precisely as you can ("answered supplier emails", not "communication"). Next to each, mark L (language or information work), J (judgment or relationship) or P (physical or on-site). Then circle the three tasks where you think AI could help most and the one task where you would never use it, with one sentence explaining why.

See a solution

Marc the plumber might write: "wrote quotes (L)", "diagnosed a leak at a customer's home (P and J)", "replied to customer messages (L)", "ordered supplies (L)", "explained a repair to an anxious customer (J)", "wrote a worksite summary at the end of the day (L)". He circles quotes, messages and worksite summaries as good candidates. He would never use AI for the diagnosis or for deciding whether an installation is safe, because it depends on being on site and on his professional responsibility.

Your own list will differ. What matters is that you end up with a mix: most people find that many of their tasks are neither obvious candidates nor forbidden, which is exactly why the next lesson teaches you to audit them properly.