Using full fidelity artifacts to iterate on ideas with Claude

I’m currently working on two presentations — one for an upcoming seminar and one for a lightning talk, both on formal verification. (Shout out to the folks at Antithesis for the BugBash invite!)

Since working daily with AI, I’ve realized something about myself: I can think much better when things are presented in a visually compelling way. For example, brainstorming with Claude on my presentations inside a Markdown file may be helpful to get a broad perspective on the direction we’re going, but to really iterate, I need to see the fully-designed slides. Maybe a transition between two related topics seems fine in a bullet list, but when I see it laid out visually, it falls flat.

Generating well-designed artifacts is pretty easy now, so instead of refining a full landing page, or marketing copy, or presentation purely in text, I have Claude generate dozens of versions of the visual container that will hold the content. When I can see how both form and function work together immediately, the iteration process goes much more smoothly.

As an example, the lightning talk will be 10 mins. The outline seemed reasonably short. But when it was translated into slides, I saw that there was a bit too much detail that 1) distracted from the main point; and 2) would clearly extend the talk to at least 15-20 mins. Had I iterated until polish on an outline, I wouldn’t have realized this until much later. I also do this extensively for landing pages. I simply cannot imagine inside my mind how content will look on a finished landing page. This was a huge problem for me before AI. Content that might sound good in black and white Arial, doesn’t really land when it’s next to a particular image or spread out across a screen. So, I iterate on the content and the design simultaneously.

It probably doesn’t help that I have aphantasia — functionally no ability to “see” things in my mind’s eye. I can’t “see” a red triangle in my mind, no matter how hard I try. I’m not even sure what people mean by that. I think entirely in terms of concepts: my mind can imagine the concept of a red triangle, but the image itself never manifests.

When I was head of web at a science communications firm, I worked with incredibly talented designers, writers, and strategists who worked diligently to understand a client’s needs, preferences, and expectations. After iterating on content and requirements extensively, the first round of design presented (a page of a website, an illustration, a diagram, etc) contained about 3-4 versions carefully crafted to reflect the client’s wishes.

In a scenario like this, where you have talented/dedicated people who can hold your content and design requirements in mind as they create and iterate, you don’t really need more than 3 or 4 options to nail down a direction. And I saw it happen over and over, where the first set of options included a near-perfect implementation of what would later become the final product.

But when it’s just me and my Claude — who, respectfully, could not care less about impressing me — volume is the name of the game. I need it to accidentally throw out an option that I can use as a starting point. Most of the versions I get after an initial quick brainstorming session are super shallow and complete garbage. But there is invariably a direction that catches my attention and pulls me to iterate on top of it.

It still takes a LOT of iteration. I usually do 80-90% of my own writing, and once I have a fully fleshed out artifact in front of me, I can provide decisive feedback on the design itself.

Even with all this iteration, sometimes I still think that it would be nice to send the “final” product over to a team of seasoned, empathetic professionals who can add a distinct layer of polish and intention that no AI (at least for now) can match.

Anyway. Back to my slides.