From vibe coding to vibe working: how to build a professional presentation with AI

I had to prepare a presentation for my colleagues at REDE Advokater about some basic concepts of artificial intelligence.

I had no slides, no finished outline and had not decided which program to use. I had an idea: explain AI by imagining that we were bringing a virtual colleague called Eva into the firm.

About an hour and a half later, that idea had become a presentation with a structure, design, images created for it, screenshots of the application, a short video and an online version that could be shared through a link.

I directed almost the entire process from my mobile phone.

It was not a matter of pressing a button and receiving a perfect presentation. It involved conversation, examining proposals, rejecting some, correcting others and deciding when the result was ready.

I think we can call this way of working vibe working.

1. First the idea; then the slides

Vibe coding popularised a way of building software through conversation and testing: explain what we want, receive a first version, try it out and keep making corrections.

That pattern can also help us prepare a presentation. But starting production before organising the ideas can produce twenty attractive slides that explain the important points badly.

So I did not start by asking for a PowerPoint. I started by explaining what I wanted my colleagues to understand.

Eva would help us distinguish the components of an AI system. The model would roughly represent her intellectual capacity. The context would be her worktable. Accessible documents and sources would form her files. The tools would be the means she could use, and permissions would be the keys determining where she could go and what she could do.

The professional would still define the assignment, review the results and make the important decisions.

The metaphor was still incomplete. ChatGPT helped me separate concepts I had initially mixed together: model, reasoning effort, context, files, tools, instructions and agents.

Only when the explanation started to make sense did we move on to design.

An initial instruction could be:

I need to explain [topic] to [audience] in [time]. By the end, I want them to understand [idea] and be able to do [action]. This is my initial proposal: […]. Before creating slides, help me organise the explanation and point out what is confusing.

2. The tools I used

The main tool was ChatGPT: discussing the idea, structuring the presentation, reviewing concepts, preparing slides and generating and integrating materials.

The REDE Advokater website served as a visual reference. Instead of describing colours and style from scratch, I pointed to an existing representation of the firm's identity.

I used Gemini to generate a short video of Eva moving and waving.

Google Slides ended up being the final format. Initially, I expected to receive a PowerPoint, but then I clarified another need: I wanted to open the presentation online and share it through a link. I had a Google account and discovered during the process that its presentation tool solved that problem.

I did not need to know every application before starting. I needed to know what I wanted to achieve.

That does not mean every account can reproduce exactly the same process. It is worth checking which features you have available, which applications are connected and which steps you will need to carry out yourself.

3. Building by looking and correcting

Once the structure was approved, the first draft appeared.

Then came ordinary instructions:

“I don't like these images.”

“I want something more realistic.”

“Make Eva a hologram.”

“I like this version, but bring back the style of the previous one.”

There was no perfect prompt capable of describing the final result in advance. Each proposal helped me understand more clearly what I wanted.

That is exactly what happened with Eva. One version had an expression I liked; another fitted the presentation better. We kept making corrections until we found an image that worked. I had not imagined it precisely: I recognised it as it began to emerge.

But the presentation could not remain just a metaphor. If we explained that users could choose a model or adjust its reasoning effort, it made sense to show where they could do that. So we included real screenshots of the application.

First the concept. Then the button.

We also used approximate page counts to explain quantities expressed in tokens, making their limitations clear: there is no exact equivalence, and a context window is not permanent memory.

Finally, I wanted Eva to wave. I generated the video with Gemini and we incorporated it into the presentation.

The focus of the work was not a particular application. It was the presentation.

4. From the phone, but with review

Proposals, changes and the integration of materials took place over approximately an hour and a half. That is the time this experience took, not a promise for every presentation.

I was not sitting at a computer throughout.

I would receive a notification, open the result, review it and reply. Then I could carry on with something else while the work progressed.

The phone was my directing station. That does not mean everything ran inside the phone, but that I could intervene without remaining in front of the computer carrying out each operation.

My attention focused on decisions.

But delegating production does not remove the need for review. A presentation can look excellent and contain an incorrect figure. A metaphor can be memorable and explain the concept badly.

That is why I like this rule:

Vibe to build; method to finish.

Before using the presentation, we had to review the explanations, check figures and sources, read the whole thing and verify that images, video and links worked.

A critical review can also be assigned to another AI. Not to start an endless ping-pong between models, but to obtain observations that we will then assess ourselves.

The final decision remains human: am I willing to use this work as my own?

5. A recipe to try yourself

Choose a short internal presentation on a topic you know well. To start with, use public or synthetic information.

1. Define the result: who it is for, how long it will last and what it should achieve.
2. Explain your idea: even if it is still disorganised.
3. Review the structure: before producing all the slides.
4. Ask for a draft: use it to discover what works and what is missing.
5. Correct with specific instructions: text, images, order and level of detail.
6. Check before delivering: content, sources, readability and functionality.

Measure the time spent making corrections too, not just the speed of the first draft.

While preparing that presentation, I realised that the process itself illustrated what I wanted to explain: no longer carrying out every task personally does not mean no longer working.

My role had been to define, select, correct and approve.

Perhaps that is the essence of professional vibe working: making use of AI's production capacity without delegating judgement or responsibility for the result.