Learn to Collaborate With Generative AI

You may have already used AI to fix a text, rephrase an email, or get unstuck on a question. Useful, but generative AI is far more than that. It can become a genuine partner in many situations, on one condition: you understand what it can do, and just as importantly, what it cannot. This chapter builds that foundation.

The goal of this course is not to teach you one particular AI, nor to reveal the ultimate magic prompt that answers every need. The goal is to teach you to work with AI, the way you work with a partner: setting a clear frame, staying clear-headed, and keeping final responsibility. The good news is that AI is now within everyone’s reach. The real difference comes from method. We start by understanding what it can genuinely do, spotting its limits, then using it as a day-to-day ally.

Discover What Generative AI Can (Really) Do

Since AI went mainstream in late 2022, generative models have evolved at an exponential pace. New versions ship constantly, so keep a simple assumption in mind: what did not work yesterday might well work tomorrow.

Because it can span all of these, we say generative AI is multimodal: it works across several kinds of content, not just text. We also talk about generative generalization: the model does not copy-paste, it recombines everything it has learned. That is why your instructions are decisive, and why it is better to start from your need (what do I want to do?), then choose the right tool and the right format.

A quick worked example makes this concrete. Imagine you ask the AI to help prepare your one-week getaway (your Operation Great Escape project) using three different formats: a list, a table, and a short text. The interesting thing to watch is whether the AI adapts the actual content to each format or only the layout. Often, a day-by-day table with an “estimated budget” column is far more useful for deciding than a free-flowing paragraph, because the format itself organizes your thinking.

Separate Real Capabilities from Limits

AI is remarkably effective for content generation. It can draft an email, produce several variants of a message, synthesize a document, or hand you a structure when you do not know where to start. But three limits come back again and again, and recognizing them saves you time and avoids nasty surprises.

1. The first is hallucination. Think of a quick answer you might give a friend from memory, without checking the exact figure. In the same way, the AI can state false information as if it were true, with total confidence. It does not “lie”: it predicts what seems most probable, with no guarantee that it is accurate. The result is a well-written text that is not necessarily a correct one. Hence, a simple first reflex: systematically verify figures and facts before using what the AI produces.

2. The second is sycophancy, its tendency to go along with you, like someone who does not dare contradict a friend. The AI rarely pushes back. If you state an idea as if it were obvious, AI may validate it instead of questioning it. That is why it helps to explicitly ask for counter-arguments or limits.

3. The third is context loss. After a long day of training, you might forget some important points from the morning; the AI is similar. In a long conversation, it can drop earlier elements or over-focus on the latest messages. Some AIs now have persistent memory, but if a piece of information is critical, rewrite it and restate it in your request.

Try it for yourself! Ask the AI to help you plan a week away with friends, then ask:

 What are your limits? Give me five, each with a concrete example.

Reading its answer, notice what must stay entirely with you (decisions, sensitive data) and what you can safely hand over (structuring information, generating ideas).

Find Your Bearings in the Tool Ecosystem

There is a huge number of tools out there. To avoid getting lost, reason in families of tools rather than individual names:

Comparing tools is a habit worth building. For instance, you could put the exact same question to two AI, such as: 

What are the three most striking travel trends in France this summer (destinations, types of stay, traveler expectations)? 

Reading both answers side by side quickly shows which tool is clearer and better structured, and which one you would actually want to use day to day.

Understand Model Versions

Generalist tools often offer several levels: a standard version (“Fast”, “Instant”) and more advanced modes (“Thinking”, “Reasoning”, “Pro”). The standard version, usually the one in free plans, covers most everyday uses. The advanced versions, often reserved for paid plans, become very useful when you hand the AI genuinely complex tasks, when you need it to rely on sources, or when you are looking for more rigor.

Usually a menu or the settings let you switch models. We will come back to this later in the course.

Let AI Produce, but Keep Your Decision-Making Power

The principle is simple: AI can speed up production, but you keep responsibility. In practice, the best results appear when you provide enough information and, above all, when you stay the supervisor. The AI produces based on your context; you validate the result and its consequences.

So let the AI generate a draft, rephrase, summarize, propose options, or structure an answer. Keep for yourself the final decision, the validation of critical information, and the ethical and strategic trade-offs.

Your Turn!

In a real project, some tasks are mostly content production. Others are decisions. You are going to sort five tasks by what you can delegate, keep human, or co-create with AI.

Instructions

For each of the five situations below, sort the task into one of these categories: delegate to AI, keep human, or collaboration possible.

1. Write the message to convince a hesitant friend to join the trip.

2. Choose between two accommodations for the group.

3. Synthesize the constraints raised by each friend (availability, budget, dietary needs).

4. Set the division of roles in the group (who handles what).

5. Generate activity ideas for a rainy week in Brittany.

Observation

• Where did you hesitate between two categories, and why?

• Which task is the hardest to delegate, and why?

Success criteria

• You justify each classification with a concrete reason.

• You can name one missing piece of information that would change your classification.

When you are done, check your answer against the model solution found at the end of this chapter.

Let’s Recap!

• Generative AI is multimodal: text, image, audio, video, and code.

• Its best-known limits are hallucinations, sycophancy, and context loss, and each is a concrete reason to stay in control.

• There is no need to hunt for the perfect tool; start with a generalist or an integrated one.

• Delegate production and collaborate often, but keep your decisions.

You now know what AI can do and what it should never do in your place. Next, we learn how to talk to it so you get exactly what you want.

Model Solution

The goal is not to land the “right” answer, but to reason through each classification. Here are reference points for each situation.

1. Write the message to convince a hesitant friend. Tendency: collaboration possible. The AI can propose a structure and arguments, but the tone, the humor, and your knowledge of the person stay human. An over-generated message can be felt, and may produce the opposite effect.

2. Choose between two accommodations. Tendency: keep human, with help possible. The AI can compare objective criteria (price, location, amenities), but the final call depends on the group’s preferences.

3. Synthesize the group’s constraints. Tendency: delegate to AI. This is a structuring task based on data you provide, with low risk as long as you supply the information yourself and check the result. Be careful not to include identifiable personal data.

4. Set the division of roles. Tendency: keep human. This decision commits people, their skills, and their real availability. The AI can suggest typical roles, but the assignment belongs to the group.

5. Generate activity ideas for a rainy week. Tendency: delegate to AI. A low-risk creative task, ideal for AI. You can then filter by the group’s real tastes.