Blog post

My AI presentation workflow, November 2026 edition

November 19, 2026 · by Jan Schultink
← all posts

January 2026 was the month that pretty much every coder, including me, wrote the last line of code by hand, moving to an AI-based workflow. The Opus 4.6 model was the trigger. I think Fable 5.2 in November 2026 is the moment for the presentation designer who knows what she/he is doing to drop the traditional presentation design apps such as PowerPoint (and the current version of SlideMagic…).

I have been (and still am) thinking about tools for presentation design in the world of AI, but so far, my prediction that a next generation of a generic AI model would make any bespoke presentation functionality obsolote has proven to be true.

At the moment, things are far from perfect, but now for the first time, if you are AI-literate and not to dependent on others/corporate policies for collaboration, you can switch over. Here is how I more or less set things up.

At the moment of writing, I use Anthropic’s Claude Slides as my main work environment. The big AI companies are vibe coding their applications and web sites themselves, and as a result there is a real jungle out there of user interfaces. What used to be Claude Design (web site with a neat way to define a graphical design system) has now been rolled into the Claude desktop app. Anthropic has begun to develop Microsoft Office competitors inside it and calls this “artifacts”: spreadsheet, motion animation, presentations, and more. “Slides” is one of them.

Slides are generated in HTML. AI models have been trained on web sites, so they are really good at creating layouts in it. (Unlike the PPTX / XML file format from the 1990s). In a chat interface you get Claude to layout your slides. Slides includes editing functionality, but it is super basic, but sufficient to make text edits, move some things around and then tell Claude to mop up after you. Reordering slides is a huge pain, but I guess the next vibe-coded Claude app will fix this.

When you are done, you convert the whole thing in PDF, or PowerPoint if your colleagues require it. The PowerPoint version will work, but it is not convenient to edit. For example, data charts are just a collection of boxes, without editable spreadsheets behind them.

AI iterations take time, and if you only have one chat going, each instruction is enough to get you distracted by emails or Instagram. In order to work efficiently, you need to fire up multiple chat sessions in parallel. All these work on the seem presentation in Claude Slides, you keep one for story line changes, and then allocate the others to sections in your deck, not to overwrite things. You will be juggling all these agents with inputs.

I can now get great looking slides in a consistent look & feel, but it requires a lot of management. Doing basic layout work in Opus 5.5, giving it another push in Fable 5.2, providing hundreds of small design instructions that I notice as a designer, but the layman probably not. For me this is more efficient, I can fire of endless improvement commands with fixes that would have taken me hours to do to dozens of agents working in parallel ("flip the axes, take the 3rd row out, etc. etc.). As a designer, I can now also try 3 different versions of the same chart and see which one works best.

Collaboration is tricky. Sharing a project, sharing a Claude Slides artifact, making sure your colleague has the right context set up (the context of the presentation, the context of the graphical design system), and is not overwriting your slides, version management, backups, etc. etc. Tricky.

On the image front, a similar situation. As a professional designer, you can now do insane things with image generators. I am learning the world of movie editing, where you can define characters, create sets, and have a consistent theme of images throughout a presentation. The trick is to maintain that consistency, and instruct your model to create white space in images for your presentation content. Different image models are good at different types of images.

My guess is the collaboration issue and poor onscreen editor will be solved soon, the tiny corrections and eye for design will be lot harder. And strangely enough, there might be a future for some sort of presentation designer in the next years: someone who actually knows how to jockey all the technologies…

To be continued.

Share
AIPresentationSoftware

About this blog

Notes on all things presentations — design, storytelling, and AI workflows.

Subscribe now to never miss a post.

RSS

About SlideMagic

A platform for business presentations.

A free student plan is available.