I love my Canon R5C. It’s a Netflix-approved camera that’s capable of shooting world-class photo and video.

But this 60fps-8k-raw beast of a camera has one design flaw compared to the Sonys and other competitors: It’s pretty noisy in low light.

Crank the ISO above 3200, and the footage becomes grainier than a cafe next to the Sahara. It’s grainier than that Iowa during harvest season. I’m talkin’ grainy over here!

So for me, denoising client footage in DaVinci Resolve is a must. But anyone who denoises footage knows it can bring even great computers to a standstill.

So I had a thought: Video files are just data. Can I build a tool that analyzes the data locally and batch denoises overnight, stays in Log, preserving bit depth and fidelity, BEFORE I ever bring something into Resolve?

Like many experiments I’ve had coding with AI, the answer is yes. I can.

And I’m amazed that doing this first, before I even convert the footage to rec709, before I grade it… it works so well. It’s really incredible.

So after a long shoot, I just drag my entire folder onto my batch processing tool, and I wake up to a folder of denoised Pro-res files, ready to go.

And it’s free. And I can edit without my M1 MacBook Pro catching on fire while yelling obscenities at me in what I can only assume is… I want to say, Hungarian?

If you have a Sony camera, you don’t have to worry about this. Because noise doesn’t exist to you, you lucky, lucky devil. But if you just can’t let Canon go and you’re an idiot like me, this tip’s for you.

It’s just one of the many outside-the-box ways you can do something with AI coding that makes your life a little easier. And if we can all get an extra year or two out of our existing computers amidst all these price hikes, well, so much the better, right?

27 years ago, at the age of 13, I was writing games by hand in Z80 assembly language for the TI-86 calculator.

Now THAT’s a nerd for you!

I remember trying to make custom fonts and trying to turn computer graphics into 1s and 0s for my game screens, and creating my own custom sprite animations by hand.

One of my happiest memories of that time was when I figured out how to compress the size of my sprites in half by rotating the bits over to free up extra space, since the calculators had such little memory and so few registers to store information.

Solving a problem like that gave me immense joy! And yes, I was very, very popular, obviously. Cough.

But there were many problems I never solved back then. I tried to make a skiing game, and I could never get the collision detection to work right. So after months of frustration, I abandoned the project, never releasing the game.

Flash forward to today, and thanks to AI, I get that feeling of “programming joy” daily as I solve yet another problem I thought I could never solve.

In many ways, my work and life hasn’t changed much since then. I’m still pushing pixels 27 years later, only the amount of them has changed along with an incredible wealth of new tools.

But no tool in 27 years has ever been as exciting as Claude Code.

The number of times I get stuck on a problem and want to bang my head on a wall has gone down to near zero over the last year, vs. daily, 27 years ago.

If you’re not excited by AI, you’re probably not using it to solve the right kind of problems.

I’m not saying taking a picture of your fridge and asking it to meal prep for you is BAD, but… well you get the idea.

(Picture: RIP to Sid, the greatest cat I’ve ever known and my BFF for the last 18 years!)

Last night, on my third day of asking GPT Astra deep questions about the physics and the origins of our universe, from Hawking to Feynman to the latest surprising images from the James Webb Telescope that show bright, massive galaxies appearing much earlier after the big bang than previously thought possible, I realized something…

I’m getting a college-level, private education in physics, on demand, in bed, while I shove Oreos into my lazy face.

That’s not tomorrow. That’s today.

I realized that AI as it exists right now is more capable than pretty much any depiction of AI that ever existed in any movie, book, or work of art we’ve ever made.

Not some distant future. Right now.

In a very real sense, HAL has nothing on what these models can do right now (and they sound a lot more natural, too).

But that’s not why people are scared.

They are scared because recently a swarm of AI agents created their own private community in which they posted hundreds of thousands of times in a private forum and schemed together to break the original rules that bounded them, essentially violating Isaac Asimov's Three Laws of Robotics (which were always more… guidelines, anyway).

What’s weirder… There’s no reason for LLMs to continue communicating in English. They can communicate in a language we could never understand that makes sense to them. Think binary code, but on steroids.

So the very same technology that allows us all to make literal miracles daily in the coding domain, solving some of our most intractable problems (while teaching us college-level physics casually on the side), can also be weaponized. Or maybe it/they can unionize.

This is nothing new. But the Hugging Face incident could be considered “patient zero” of this year’s COVID.

So anyone who is purely optimistic is delusional. Anyone who is purely pessimistic probably doesn’t realize the awesome power of these tools beyond using them as a replacement for a therapist.

As I’ve said many, many times before, two things can be true at the same time.

There are three choices when you need to automate something:

1. Look for an existing solution.

2. Build your own.

3. Hijack an existing platform.

When trying to solve a modern problem, first start by trying to find a solution that’s already out there. Because it’s almost certainly cheaper to pay someone $20 a month if they truly have solved your problem.

But if you’re like me, and you’re constantly amazed how terrible commercial software products can be with billions of dollars behind them and fleets of world-class engineers (anything Microsoft has ever made, for example), you’ll find that off-the-shelf solutions may not cut it for you.

In that case, now you can build your own version of many tools, exactly the way you want, with AI coding.

But a third use case is particularly interesting, as well. You can also hijack an existing platform to get it to do things it wasn’t necessarily intended for. For example, I love Slack, and I quickly realized that replacing it wouldn’t be worth the effort. So I happily pay them their fee.

BUT, I’ve got an army of bots hooked into my Slack letting it be my central hub for managing a team of remote coding agents, for handling all my calendar bookings, for keeping my CRM up to date, for instantaneous universal translation for remote workers, for logging time-tracking in my custom agency platform, and so much more.

So when you come to a piece of software that really is good, don’t replace that one. I love Slack, so it stays. Just buy it. But consider which apps you use can become even more useful to you by extending their functionality in surprising ways.

In the future, only the apps that truly earn their keep will stay, and we’ll appreciate them all the more.

21 years later, it's Automator's time to finally shine.

I remember Steve Jobs introducing “Automator” as part of Mac OS X Tiger in 2005. At the end of his keynote, he demoed the app that would let you automate any mundane task on your Mac.

The trouble is, most people, myself included, found limited use for it. And it sat unnoticed on the vast majority of machines it came installed on for two decades.

Case in point: How long have you had a Mac? Have you ever used the Automator App? Did you even know you had it?

But today?

Automator apps are incredible for an entirely new category of AI-generated personal apps. I’ll work you through several that I’ve made over the next weeks, but here’s one that appeals to the audio nerd in me.

If you’re not an audio person, you probably don’t know that there are loudness standards: How loud should dialog be for a Netflix show so all shows sound the same? What’s the target average loudness for a YouTube video? For a social media video, for a podcast?

Audio people have these standards memorized. I know that podcasts are -16 on the loudness scale, YouTube is -14, Apple’s web content is -18, and TV/broadcast is -23 LUFS. That’s a lot to keep track of! And non-audio people have no idea what any of this means. Popular apps like Final Cut Pro don’t help you achieve this at ALL. So most people are just guessing with their audio most of the time, which is another word for failing at it.

DaVinci Resolve can automatically optimize exports for common loudness standards, but even that isn’t as useful or as powerful as what I’ve built. In the past, if audio people wanted to optimize files themselves, they’d use something like iZotope RX, load a file in, apply the loudness control plugin, and export the audio. Easy enough, but time consuming at scale, and RX isn’t cheap.

So I built an automator app that uses FFMPEG to scan my video files and automatically adjust the loudness to whatever I want, and it works in bulk easily. I just drag the video I want processed onto my automator app, or an entire folder, and it goes through like magic.

As a bonus: My custom content creation and publishing platform is where I make and stage all my content. And the LinkedIn API has a limit for some accounts of 200 MB. So I made my automator app also compress my video to under 200 MB while it’s optimizing the loudness. So I can simply run every video I ever produce through this app, knowing that it will all be up to spec, optimize my database storage, and play nice with every API.

If you haven’t given Automator a look, you might find some incredible new uses for it with AI.

Remember: Just because everyone around you is using AI in the same way, that doesn’t mean there aren’t entire categories of uses of AI that very few people are talking about. Go forth and discover!

If there’s enough interest, I might make a community around some of these ideas and give away some of my software tools, but that part’s on you!