Welcome back to the AI For Humans newsletter!

On Friday, OpenAI announced GPT-5.6 Sol. Then it… didn't release it. (More on that below.)

The short version: the smartest AI models on earth are stuck in a weird kind of limbo right now, tangled up with US government and the ongoing Anthropic Mythos drama. Thankfully, Claude Fable 5 might finally be back next week!

So I could spend this whole issue retelling that saga. Blah, blah, blah.

But I'd rather just show you something useful.

Because while everyone's refreshing their feeds waiting for the frontier to reopen, I spent the weekend going deep on a new tool that significantly upgraded something I do often and I think it’ll help you too.

And it's something you can use right now, today, no government approval required.

Let's get into it.

Please click through to help AI For Humans grow!

If you’re a decision maker at your company, you need to be on the bleeding edge of, well, everything. But before you go signing up for seminars, conferences, lunch ‘n learns, and all that jazz, just know there’s a far better (and simpler) way: Subscribing to The Deep View.

This daily newsletter condenses everything you need to know about the latest and greatest AI developments into a 5-minute read. Squeeze it into your morning coffee break and before you know it, you’ll be an expert too.

Subscribe right here. It’s totally free, wildly informative, and trusted by 600,000+ readers at Google, Meta, Microsoft, and beyond.

I talk a lot about doing little experiments to figure out what AI can really do.

And I mean it: you can read every thread and watch every demo, but until you sit down and make the thing yourself, you don't actually get it.

So here's this weekend's experiment.

HeyGen just updated HyperFrames, their skill for AI coding agents, with a bunch of really useful sub-skills and presets.

The HyperFrames tl;dr?

An AI agent builds real videos and motion graphics by writing HTML, CSS, and JavaScript, then rendering them into an actual MP4 right on your machine. You describe what you want and the agent codes it into existence.

Sounds insanely nerdy I know, but this has real world practical use cases for almost anyone.

My first instinct was "oh cool, I can add visuals to my short-form videos." But most of you aren't posting to TikTok or Reels every week.

But I would assume nearly all of you have to make a presentation at some point.

A deck, a pitch, a report you're walking a room through. This is exactly the kind of thing that makes that stuff pop, the gap between a flat slide and something that looks like a real motion graphic.

Like most agentic AI stuff, Hyperframes does not work perfectly out of the box.

And weirdly, that's the part I find most interesting. A big skill all of us have to build as these agents become our co-workers is learning how to coax the thing in your head out of them. It's not always as easy as you'd want it to be.

But I got a motion graphic I was really happy with after about 30 minutes of back-and-forth with Claude Code. For my own work, that's a major, major step up.

Want to try it yourself? Quick version below.

1. Get a coding agent. For most of you, the easiest way in is the Claude or Codex desktop app. (If you're super nerdy, the terminal works too.) You'll need a paid account with Anthropic or OpenAI. One note for Claude: make sure you're on the "Coding" tab of the desktop app.

2. Ask it to install the skill. Say something like "can you find and install HeyGen's HyperFrames skill?" It should go out, find it, and set it up for you. If it needs a nudge, point it here: hyperframes.heygen.com

3. Use the HyperFrames design agent first. Expert hint: let it give your project a look before you start (even your own brand's look), so it's not guessing in the dark.

4. Give it something to build. My graphic came out of a stat I'd heard, that the movie Obsession (which I saw this weekend, and it's great) actually climbed at the box office in its first two weeks. I didn't have the numbers. I just told the agent to go find them.

5. Go back and forth. Tell it what to fix until you're happy with it.

That's really it. It can feel intimidating, but it's very simple once you start thinking of the agent as a little co-worker you send off to handle the thing you need.

It'll get frustrating when it doesn't do what you pictured (which, yeah, is why we need Fable 5 back, like now). But that back-and-forth is the skill. Practice and push through.

While the smartest models sit behind government letters and limited previews, the tools already on your laptop & in the cloud can do stuff that would've been impossible a year ago.

The promise of AI was never really about the biggest model. It's about you getting your work done faster and better than you could yesterday. That's available right now, and almost nobody is using it anywhere near its potential.

So don't sit around waiting for the frontier to reopen. Go make something this week. I want to see it.

-Gavin

This week on AI For Humans: Is This The AI Cold War?👇

Yes, I'm hijacking this section again to show off something I made. Sorry!

I've gotten a little obsessed with AI microdramas taking over TikTok and Chinese social media lately. So I'm challenging myself to make one over the next few weeks.

I really didn't want to make "Revenge Of The Pregnant Seahorse 2" (an actual microdrama series with millions of views), so I built my own look instead, a 90s anime vibe I'm pretty into.

Below is my "pilot," a two-minute video I put together in about six hours with a grab-bag of tools. You can read the full workflow in the thread:

If you want to follow along on this dumb, fun little adventure, I'll keep posting there. And don't worry, I'll do a full breakdown right here once the whole thing is done.