Almost Timely News: 🗞️ Demonstrating the Art of the Possible in AI

Almost Timely News: 🗞️ Demonstrating the Art of the Possible in AI (2026-01-18) :: View in Browser

The Big Plug

🚨 Take this one question survey to tell me and the Trust Insights team what webinar/training content you’d like us to make this year!

Content Authenticity Statement

100% of this week’s newsletter content was originated by me, the human, and 80% of it was edited from my original voice recordings by Anthropic Claude Opus 4.5. Learn why this kind of disclosure is a good idea and might be required for anyone doing business in any capacity with the EU in the near future.

Watch This Newsletter On YouTube 📺

Click here for the video 📺 version of this newsletter on YouTube »

Click here for an MP3 audio 🎧 only version »

What’s On My Mind: Demonstrating the Art of the Possible in AI

“It’s not who you are underneath, it’s what you do that defines you.” - Batman

One of the most common refrains I hear on a weekly basis is that you feel overwhelmed by the pace of change with AI. Even OpenAI cofounder Andrej Karpathy was recently posting about how far behind he was in terms of what capabilities are now available. If one of the leading minds in the field feels overwhelmed, it’s no wonder the rest of us sometimes struggle to keep up.

So this week, let’s talk about changing focus so you can feel less overwhelmed AND get more things done.

I’ve shared in the past - and we have an entire course about - the seven use case categories of generative AI: extraction, classification, summarization, rewriting, synthesis, question answering, and generation. Those are still important and valid. But here’s the thing: those are process-based categories. They describe how AI works, not what you get out of it.

And each use case category has outcomes, but those outcomes aren’t necessarily the finished products that make people go “wow.” Showing someone a data extract? That’s not really a wow, right? They’re like, “Okay, that’s cool, you got data out of data, but what do I do with it?” It’s one step in a recipe, not the finished dish. You’re not doing just an extraction task in isolation - you’re doing an extraction task that leads to something else, something meaningful.

So I want to talk about how we get to those finished products. How do we get to the wow? How do we get to the art of the possible? What are the outcomes that make people sit up and pay attention?

Think about what outcomes you want with AI, what the tools can do. I’m intentionally ignoring the bigger picture around people and process because that’s what my cofounder and CEO Katie Robbert focuses on. Broadly, our outcomes are almost always about either saving resources (time, money, effort, materials) or creating resources.

When I do live demos at conferences and workshops, I make it look like a magic show. There’s a lot of prep work to make sure I have good prompts and good systems in place, but once those systems are ready, it looks easy. The outputs are pretty terrific, and people can see that it doesn’t take long to do. That’s what creates the wow - watching something happen that they didn’t think was possible.

I’ve developed a framework for thinking about AI outcomes that I call CRAFT: Create, Refine, Arrange, Flip, and Tackle. Let me walk you through each one with examples of what “wow” looks like.

LetterCategoryDefinitionCCreateBuild net new deliverables from ideasRRefineExtend, improve, and modernize existing workAArrangeOrganize chaos, strategize, bring orderFFlipTransform content across domainsTTackleTake on tasks and solve real problems

Part 1: Create - Building What Didn’t Exist Before

Create is about making stuff that simply didn’t exist before. Bringing ideas to life. My number one use case outcome here? Software.

I’m talking about building things that are outside your skill set. I use tools like Suno for music generation because I have zero musical capabilities. Zero. I wrote a trashy romance novel a few weeks back in my newsletter because it illustrated AI’s capabilities perfectly - I can take an idea and bring that idea to life. That’s the heart and soul of Create.

If you’ve got an idea and you know the vocabulary to ask AI for - or you have the meta-prompting skills to get AI to reflect on what skills and background it needs - you can bring any idea to life within what AI is capable of producing.

Here’s the thing though. A lot of people use generative AI for mundane tasks: writing blog posts, drafting emails, that kind of thing. There’s nothing wrong with that - they’re practical and useful. But they don’t illustrate the art of the possible.

Nobody is wowed by another blog post.

Those are practical, tactical, useful things, but they’re not sexy. So what does wow people?

Making a highly styled infographic or interactive that uses your brand guidelines. Building a browser-based video game - I did that once for an architecture industry talk, taking a super boring RFP response and turning it into a little browser-based video game using the Phaser library. Something fun. Something light. Something that opens people’s minds to “I didn’t know I could make that.” Anytime you make a video game, people are wowed by it. That’s a pretty straightforward reaction.

Another example: writing an academic paper. I wanted to validate whether LinkedIn was biased in how it reports feed results. Instead of gathering anecdotal evidence or setting up badly controlled tests, I gathered all the technical papers and engineering blogs from LinkedIn’s engineering team, replicated their system as closely as possible, ran tests with actual platform data, and ended up with a statistically valid, rigorous paper published on Zenodo with a DOI number. People didn’t think “I could have generative AI help me build a full-fledged research paper.” But you can. That’s good quality research that frankly had not existed until that point.

And here’s what really drives home the Create category: when people see solutions that they didn’t know could be solved. Problem solving and creating things to solve problems is sort of AI’s magical power. The wow comes from watching an idea become real in minutes instead of months.

Part 2: Refine - Extending and Modernizing What Exists

Refine is about taking any existing thing and extending it. Improving it. Bringing it up to modern spec. I use Claude Code heavily for this.

Here’s a concrete example. There’s this piece of software called iMouseTrick - a high-speed clicker I use in video games. It was made in 2010. When I upgraded my Mac recently to Tahoe 26.2, it finally stopped working. Sixteen years is a good run. So I sat down with Claude Code and said, “Here’s what the software did” - took a screenshot of the interface - “here’s its purpose, let’s build this in Swift,” which is the modern language for macOS.

In 45 minutes, I had not just replaced that software but extended it, refined it.

I did a deep research project to gather best practices for Swift applications. I made it compliant with modern standards. I added new features: jitter in the timing between clicks so it wasn’t exactly the same interval, the ability to save preferences so you don’t have to redial settings every time - whatever the last settings were that you used, it remembered them. A click counter that showed how many clicks it had made. Little things, but all things that extended the application beyond what it was.

Another example: Google released a tool to show how query fan-out works, built in both React and Python with a front end and back end. I thought that was unnecessarily complex, so I had AI move it all into Gradio. But I didn’t stop there - I wanted it to replicate what AI Overviews and AI mode does, and what deep research actually does, as closely as possible. I gathered all the relevant papers, patents, and public statements from Google, figured out what was reusabl