PowerPoint via Whatsapp

I have been experimenting with OpenClaw, a 24/7 personal AI assistant. It is an open source project that was created over the past month, created almost by accident as a hobby project by a developer. Late last year, developers discovered that using an AI coding tool for non-coding applications really works, especially if you can give it access to local files (rather than chatting). Next step: keep the AI agent running 24/7. Next step, find a way to let the AI agent keep some context about you. (Context memory is a big issue in AI, it fills up at some time, so storing bits of context for future reference enables persistent awareness). Next step, implement recurring instructions. Next, make the AI assistant available from any channel, including Whatsapp. This, combined with full system access, creates an incredibly powerful AI assistant (that can also be incredibly destructive). “It’s 2AM, I have a 12 hour to do list and full shell access, lovin’ it”. What can possibly go wrong.

So it requires some technical knowledge to run this thing in a vaguely responsible way. I put it on AWS EC2 in a completely isolated virtual network, with an access tunnel that only I control, plus a number of prompt injection filters running on yet another server. (OpenClaw’s founder has joined OpenAI exactly for that reason, to get the financial resources to fix the complexity and security issues)

Working with it, enables me to get a feel for the future. Direct instructions to applications (code editors, presentation design software, spreadsheets, food takout apps) will become increasingly irrelevant.

We will move from well-designed, beautiful, easy to understand, visual user interfaces that humans can interact with directly to clear and transparent API endpoints that an agent can call on behalf of a human.

Interesting times.

AI images are the new stock images

People are starting to develop a pretty good sense of whether an image “smells like AI”. The audience will notice when you use them just as visual fillers like you used to use stock images. (“Life-style conscious gen-z person working from the local cafe”).

Completely obvious AI renderings are comparable to clip art of the earlly 2000s. Hyper realistic AI images are similar to stock images, “something is off here”.

Apply the same approach to AI images as you did to stock images.

"Files are so back"

The past decade with spend a lot of energy moving all our files into the cloud, and accessing them through a browser-based application. That is going to reverse in the era of AI.

A good old desktop folder is the perfect context for an AI agent. All information in one place. Different formats, different applications, all accessible and editable.

I started experimenting with a spreadsheet/presentation workflow that mirrors the way I write code. Pointing Claude Code at a directory with files and let it orchestrate analysis and design under my supervision.

The "wait for AI" workflow

As AI gets more and more capable and I use it for more and more things in my daily workflow, it creates a new problem: waiting for AI. Most tasks now take 1-3 minutes for which you have to wait for the result. Most takes would have taken you much much more than that if you were to do them yourself, so a big productivity gain, but, 1-3 minutes is too short to go and do something else, so you wait, get distracted, break your flow, and find yourself getting back to what you were working on 30 minutes later. “Ah small mistake, let’s do fix that. Click. 1-3 minutes wait.

I have not found a solution yet, and am experimenting:

• Running lots of agents in parallel, so it is like a plate-on-stick catching game

• Running the tasks in parallel of an activity that is easy to interrupt (watching a tutorial video)

• Writing a quick blog post…

• Etc.

A new world

Claude plug-in for PowerPoint

This has gone under the radar screen for me. The Claude plugin for PowerPoint is available for Claude Max, Team, and Enterprise users. It links to your existing Claude account (to burn tokens) and does exactly what you expect it to do. More information here.

On the "danger" of AI

Recently, people start getting really concerned about autonomous AI doing its thing with your passwords and credit cards. But I think that not that much has changed since the first emergence of computers. Computers trade billions in stocks and currencies, run nuclear powerpoint plants.

The difference, these systems are carefully monitored and are predictable and well-understood. It is not about the intelligence of the software, it is about the level of control you give it. People wanting “to push the boundaries” and see how far AI can go seem to be putting a 10 year with Microsoft Flight Simulator experience at the controls of a real 747.

I have started to use AI tools heavily in my day-to-day work. Not for email and calendar management, but for coding and document production. All seems reasonably under control, things being sandboxed on my own machine. “Reasonably” What is not, is the chaotic user interface when it comes to giving permissions to things on your file system. Many of these questions will not be understood by the IT layman.

Everyone is raving about Anthropic’s Cowork tool to automate knowledge work. Many incumbent database/information providers and consultancy firm’s stock prices get hammered. The real revolution for Anthropic might be an acquisition that did not get a lot of press. In December, they acquired Bun, a software development platform. My sense is that Anthropic is developing a trusted platform that can build and run code safely on your computer without the constant flow of questions whether it is OK to install package x, y, and z. I think AI models will become a commodity, the real winner in the next generation of computing will be the player that offers this trusted platform that can keep AI in check.

Let’s read this post in 2036 again and see what happened :-)

Pondering SlideMagic 5.0

I had a backlog of feature improvement for SlideMagic in the back of my mind for a long time (SlideMagic 4.0). Started working on it, but never pushed them to a final end product. I am actually glad I didn’t. The path I was going down on was a “hand-held” guidance of AI combined with a dramatically improved graphical user interface. The development of AI models have made this approach obsolete.

i am starting now to tweak and customize the recently released knowledge workflow productivity tools so they work for me. Strangely enough, my decision to build SlideMagic 2.0 as a desktop app with local file access, will come in handy now to build a tool that wraps the raw engine and makes the underlying engine accessible to the average user.

Let’s see where this goes.

AI Mental Drift

AI is very poor at navigating context history. “Go back to the previous version”, “change this”, “replace that”, will almost always lead to confused, diluted responses. You need to “freeze” progress in your work. You got to a good version. Asked for an improvement but did not get it: open a new chat windows and start fresh without the last message exchange.

I am sure it is a matter of months before this issue gets resolved in most AI applications.

When not to use AI...

Triggered by a post by Nate Jones on Substack (might be paywalled).

I am working on a next big update for SlideMagic, where I am trying to push AI-supported slide generation to the next level. Especially when it comes to layouts with data and text in them, AI is not there yet. The more I start experimenting, the more I realize that AI should not be used everywhere: hallucinations, unpredictable (non-repeatable) outcomes, long wait times, LLM model changes that can completely trip up the tool you have written on top of it.

Nate identifies 4 use cases:

1. Traditional data processing (adding up last month’s sales by product)

2. Traditional machine learning (predicting machine failure base on historical data)

3. Generative AI (summarizing stuff using a chatbot)

4. AI agents (planning your holiday by optimizing and matching lots of constraints)

Work in progress.

More Nano Banana

I worked with Google’s Nano Banana a bit more over the past days, and I think I understands what it is doing under the hood.

“Regular” imaging LLMs predict pixels, you give a prompt, the prompt gets translated into a series of tokens, and the model predicts the best matching pixels given the token input. A flat “soup of pixels” is the result. And because of that, it is hard to make small adjustments to an image, editing one particular aspect and leaving everything else as is.

I suspect Nano Banana works with layers. The model tries to understand what aspect of it refers to the bottom of the pile (the background) and what elements go on top. As a result, it is possible to make very precise edits to individual objects in the overall composition of the image.

In order to make a coherent image, the model needs to have a good understanding of the 3D perspective of the background, and all the objects above it. Like the example about the Porsche in a Dutch town in my previous post, the car gets rotated, and pasted back into the background image with the correct vanishing point in mind.

Vanishing point is preserved when making edits to the image

What the model cannot do is change camera position. view the entire image from a completely different angle. Zooming in and zooming out works. An example is the cover image of this post, where I took an image from the band of my son (Project71) and put them on a big stage. I could not get the model to produce a view from the audience given the image it already produced. (Starting from scratch with an explicit prompt for an audience view would have worked of course).

Note the small glitch in the keyboard of the synth

This is a limitation I can work with for the moment though.

PS. I work with Nano Banana via Google AI Studio, not via its own web site

Nano Banana

I just played around with Google’s “Nano Banana” AI image generator, and it is incredibly good and useful for presentation design.

Current AI image generators take a prompt and predict pixels. Ask for a modification, and a whole new bunch of pixels get generated, redoing the entire image. Nano Banana (we need a better/shorter name), seems to work with layers and objects, and keeps things consistent.

Below 2 quick examples:

“White Porsche in Hoogeveen”

“Turn it around”

Some observations:

• Super fast, the first image was an almost instant response

• Hyper realistic image, does not look cartoonish

• Correct text: the name of the cafe, the license plate, the branding of the car

• (That town looks Dutch, but it is not Hoogeveen)

• But most importantly: isolated editing, changing one thing and leave everything else the same

Photoshop, it was nice meeting you…

I will study the API structure of Nano Banana and see whether I can swap out the image generator in SlideMagic.

Impressive! You can try it out in Google AI Studio

SlideAudit - Academic research to improve slide layouts

I am following academic efforts to use LLMs to improve / automate slide design with great interest. Each takes a slightly different approach. SlideAudit was recently published by Zhuohao Jerry Zhang and others.

SlideAudit teach LLMs what good design is by teaching it rules and principles. A lot of effort goes into building a bank of slides, identifying design flaws for training, synthetically introducing slides, letting the model run and evaluate the results.

I think this approach can work well for publications that resemble print: designs with lots of text in smaller fonts, and images / graphics that are placed in some sort of grid. Books, magazines, newspapers, but also web sites.

Presentation slides are trickier. It is harder to describe what makes a slide a good slide. You know when it when you see a good one (or a bad one), but pinpointing and automating the steps to go from bad to good is tricky.

AI is good at reading data from charts

Need to make over a slide but don’t have access to the data in a graph? AI to the rescue. Upload a screen shot to an LLM and you get back pretty good estimates of the data values in the chart. It might not be scientifically 100% accurate, but good enough to recreate the graph in your own presentation software.

AI to clean up text tables

One of my big slide puzzles is usually a messy table of pros/cons of a product, or a competitive comparison. How to highlight the right dimensions. Make sure that one is not a sub point of another. Make sure that cells have short text in them. Make sure that text is roughly equally long in each cell.

GPT-5 is very good at this. Copy-paste the messy table into the interface, and the output is pretty useful. Something to integrate in SlideMagic at some stage.

No more disguised examples?

Many confidential presentations often use disguised case examples of clients, potential investments, drugs in development. AI has become incredibly good at uncovering even the most vague ones. Before sending your deck, double-check them in an LLM to see what comes up…

The end of the presentation?

Apologies for the click bait…

I have not been writing here for a while now as I am focusing on 9xc/9vc. But recently, as I am pushing further into the world of AI, all my pas experiences seems to be coming together: computer science, company analysis, presentation design, and hardcore biopharma science…So I might occasionally come back here.

The majority of “presentations” are documents that are used make decisions inside companies. They happen to have graphs and other visuals inside them, hence the word “presentation”. Most humans are not very skilled in writing a concise memo to make a point: hence bullet points in large font and visuals to the rescue.

AI could change that: boiling down these big slide decks to a few paragraphs with a decision that needs to be taken, pros/cons of alternatives, and the recommended way forward. No slides needed.