Breaking the 4th wall

One of the trickiest parts of teaching someone how to use AI is that it can do…well basically anything.

The dissonance comes from the fact that most AI companies position their products as a “tool” so as to not threaten you. It’s a tool for lawyers, not an AI lawyer.

But that locks people into traditional ways of thinking about tools. That they are built for a use case, and I need to have that use case to start.

But as we’ve seen, most AI products start you on a blank text input. So people get stuck, understandably so.

The key is to realize that AI is like clay. It’s moldable, and you can’t really think of it as a traditional software tool even though it’s just bits.

In our AI training and workshops, we emphasize this concept of breaking the 4th wall in AI and embracing the uncomfortable reality that what’s in front of them is way more capable than they can imagine.

Here are some stories that might help illustrate this further:

When the 4th wall gets broken

Using Claude to set up Claude

In one of our workshops, we helped 60 people onboard to the Claude platform at the same time. A big part of that was setting up everyone’s connectors, and unsurprisingly we quickly ran into issues. For some the process was seamless, but for many others the OAuth redirect URLs were breaking and people got lost in between screens.

After much individual debugging, it left the group in a weird partial state. How can we quickly check that everyone has the correct set up?

In a traditional software tool, this means fumbling around “Settings” until you see green checks everywhere. You can do this on Claude too, but we came up with a much more effective method on the spot. We asked everyone to type this into a new Claude chat:

Can you check that all of my connectors are working?

The primary result was that we got very quick answers without having to share screen again and show people where to click. But the more interesting secondary effect was that it showed everyone that Claude can see its own settings. That Claude can be used to set up Claude.

After realizing that, people started to try all sorts of things.

Prompting AI to prompt you

In a 1:1 session I had with a PM, the PM was complaining about how AI didn’t really work. It was quite vague, so I prodded further and was able to get him to reflect on all the features he wanted that were missing from a dashboard artifact he had created.

Funny enough, I often have Granola running while this is happening. So I decided to try something I hadn’t before and just prompt Granola to summarize what’s happened so far and formulate it as a follow-on prompt that we can give Claude.

I shared that prompt with him, we ran it, and it just worked. When he asked me how I typed 500 words in 2 mins, I shared my screen and showed him my set up. That was another “breaking the 4th wall” moment where he realized that I was just prompting him to pull the ideas out of his brain. And that he could do the exact same thing himself with AI.

“Wow nice okay yeah. It makes sense to have AI help create the prompts, because language is what it does best.”

Closing

In storytelling, breaking the 4th wall means shattering the invisible barrier between the performer and the audience. It applies here because thinking about AI as a tool creates that same invisible barrier between you and AI. And if the key to AI is constant exploration and tinkering, then unlocking your imagination around it is often the most important first step.