Over the past few months, I’ve taught a few courses on AI and marketing, been on a bunch of industry panels, and started advising a startup in the space.
Which means I keep getting asked:
“How do you keep up with AI?”
But when I hear that question, I often can’t help but interpret it as, “Don’t you have anything better to be doing?”
Which is understandable.
So much is happening, so quickly, it seems impossible to stay up to speed without dedicating every spare second to learning.
Now, if you’ve read this newsletter before, you know I’ve got takes on AI and its (mis)use, so I’m not exactly an unbiased source.
But I tend to feel that when entrepreneurs ask me, it’s because they’re worried about “falling behind”.
They’re worried that they’ll either get replaced by AI themselves, or they’ll miss out on some important opportunity.
And so they try to keep up. They read newsletters, scroll LinkedIn, watch YouTubes, attend courses, even buy books about it.
But I’ve noticed that the people most anxious about not using AI enough are the ones using it the least.
Because my honest answer when I’m asked “How do you keep up with AI?” is “I couldn’t stop if I tried.”
I’m a nerd from all the way back.
As a kid, I saw that understanding computers afforded the people around me, even in my own family, career opportunities.
“Being good with computers is a useful skill” may in fact be the first thing I ever truly learned.
I was never taught that, I just saw it, and that got encoded somewhere, in a way that “playing sports might lead to a more active and healthy lifestyle” apparently never did.
So, to me, technology has always been more than interesting, it’s always seemed important.
And over the past several decades, I’ve found you can’t learn technology by watching other people use it, or just listening to other people talk about it.
Every pundit has their angle, every commentator their bias. Every company their bottom line (even me!).
So to actually understand technology over the decades, I’ve learned how to keep up with it. Which has always been to use it, whatever the particular technology is.
From typing out little stories on an I-literally-can’t-believe-I’m-saying-this electric typewriter as a wee toddler, to making websites in grade school in the 90s, to building agent loops via Apple Shortcuts in 2023, to creating custom iOS utilities using Claude Code this year... making stuff is how I’ve always learned to understand stuff.
And with LLMs and modern AI “agents”, you really can’t understand their utility (and their hazards and weaknesses) until you start to think in tokens and loops.
If you see AI as a chatbot, as a magic box you talk to, you’re missing essentially everything about it that makes it interesting and powerful.
But simply saying that doesn’t really mean anything to someone who’s never built anything with them before.
So do me a favour.
If you’re not sure what I mean by tokens and loops, but you’re interested in learning more about AI and have an existing account with a provider (like ChatGPT, Claude, or Gemini), go give it this prompt and read its output:
[What’s an LLM instruct model? And how can you demonstrate that you are one so a student of AI could get an intuitive understanding of how token streams, when looped back in on themselves, can simulate conversation?]
Seriously, go do that. It’ll only take a couple minutes.
...
Okay, that was cool, right?
Do you kind of sense it now, do you kind of see the loops in your mind?
What you might see next are two things:
- That the abilities of these tools are more constrained than you previously imagined
- And that you can do a lot more things with token streams and loops than you previously understood
Most importantly, I think once you get a sense for what’s happening behind the scenes, you become less awed by it. And that’s a good thing, because you can start to see how to practically and safely apply it.
You start to get a sense of the limitations of a next-token-prediction algorithm, and the potential of a natural language emitter to be able to control tools on your behalf.
I’m starting to think about it this way:
Did you see the movie Oppenheimer? Remember that line Niels Bohr asks Robert, when he’s talking about truly understanding physics?
He asks, “Can you hear the music?”
Me, I’m neither a physicist, a genius, or a musician, so in many ways I cannot “hear the music.”
But when it comes to Large Language Models, I can think in token loops.
And that’s been extremely helpful for my work and for so many interesting personal and professional projects.
Like I said above, I’ve been building custom agent loops since 2023, only a few months after ChatGPT came out. I started building persistent memory systems, multi-agent coordinators, and parallel tool calling techniques years ago, many of which are only becoming mainstream now.
And it never felt like I was struggling to keep up, it felt like I was struggling to spend as much time as I wanted to building these exciting projects.
So, frankly, I’m not here to try to convince anyone to use AI, or even to care about it.
I absolutely get why so many people hate it, and I agree with a lot of their reasons.
And I certainly understand that people feel forced to use it and ‘compulsion deadens enthusiasm,’ as they say.
But for the people who keep asking me how I keep up, this is my answer:
“Keeping up” is the easy part.
Finding the time for everything I want to do with that knowledge is the hard part.
So that’s your assignment today:
Find a project that gives you something to do, and you’ll find learning how to do it becomes the easy, natural part.
(Or just send me an email and we can talk through some potential projects you could try out.)
And you’ll never feel behind again.
Kelford Inc. is the marketing team that’s never at a loss for words. If you’re struggling with what to say and where to say it to attract ideal clients, we’ll show you the way.