Great communicators win (AI)

A skill that is somewhat hard to quantify, but I found to be incredibly important in work and really sets apart good from great colleagues is the ability to communicate effectively.

In the age of AI, this is more true than ever before. With the latest model of Anthropic Claude Opus 4.7, it apparently tends to do a great job, but only if you communicate effectively at it: if you are precise with what you want and how you want it to be done. If not, there’s a good chance that the model will use more tokens (1.3x its own model card claims) than previous models.

In other words, the worse you communicate, the more effort the large language model has to do, which in turn costs more money. Interestingly, this is very similar to life before AI where you had to communicate with colleagues and if you are not clear about what you’re looking to communicate, your colleagues will have to do a lot of interpretation and they might not do the exact thing that you told them or asked them or otherwise communicated to do.

How to communicate with AI effectively

I’ve said this before, but I cannot recommend enough that you start actively using voice to text. There’s all sorts of software for this. I like to use Aqua, but there are many alternatives that might suit your needs better.

Using software like this allows you to press a button, talk into your microphone, and it will on the spot write whatever you are saying. Even correct little mistakes and do interpretation of the words that you’re using so that you are not having to go back and actively edit text. In fact, the words that you are reading right now are all narrated by me into my microphone and have received no editing whatsoever until now.

So when you are using AI, what you then do is you just talk into your microphone and describe whatever you wanted to do as clear as possible. But you don’t just describe what you wanted to do, you describe very clearly what is the outcome that you hope to achieve. And modern large language models have an enormous context window, so you can really yap on for quite a while and it will not be a problem.

An example

Let’s say that you want to make a website for yourself that shows off your art. AI is really good at this, and you can ask any to make a little website for you right now, and it will do just fine.

Here is what not to do:

AI, please make a website that showcases my art. [you attach your art here]

With this prompt, AI will do just fine. It will create your website with your art. Now, depending on which model and it might ask you some follow-up questions, but this is not a good start. If you don’t tell the painter which color to paint your room, she will choose one. The same goes for AI. It’ll probably be white and boring.

Here is a better example:

I want to make a website that showcases my art. I’ve attached the art, but what I want you to do is make it so that there’s a big screen that has brutalist elements where you showcase two or three pieces of my art at random to a visitor and then allow them to click on any piece and be guided into a gallery of all of my art where we put one in big and they can then click on subsequent pieces of art. It makes those big and gives a bit of an overview.

The website should be opinionated in design, so brutalistic. Make use of a monochromatic color scheme, but if there’s any highlights necessary, use a bright green. The background of the website should be light colored, white or just off-white. And the typography should be equally opinionated.

Please ask me questions about elements of this that aren’t clear or would need further specification

Much more clear to anyone reading what you want, and therefore to AI as well!