Welcome to the AI For Humans newsletter!

If the leaks are right (and the timing tracks), Anthropic's mysterious, "dangerous" Claude Mythos is dropping this week. Maybe GPT-5.6 too.

We don't know yet which version we're going to get. But the leaked SVG game controller builds making the rounds on X this weekend suggest we're in for a big upgrade.

How could this change the current AI world? A lot.

Let's get into it!

Please support AI For Humans by learning about our sponsors below:

Your browser gets you to a page. Norton Neo gets you to the answer. The first safe AI-native browser built by Norton moves with you from idea to action without slowing you down. Magic Box understands your intent before you finish typing. AI that works inside your flow, not beside it. No prompting. No copy-pasting. No switching apps.

Built-in AI, instantly and for free. Privacy handled by Norton. Built-in VPN and ad blocking protect you by default. No configuration. No extra apps. Nothing to think about.

Fast. Safe. Intelligent. That's Neo.

Download Norton Neo

Quick primer in case you haven't seen these before.

SVGs (scaleable vector graphics) are code-generated images. The model writes actual code, and that code renders the picture.

Which is why they're a really good benchmark for vision and reasoning at the same time. The model has to understand what the thing looks like, then write functional code to draw it.

Early versions of this (think Opus 4 or before) were rough. Blocky. Often wrong.

The Mythos leaks? Not blocky. Not wrong. Actually… really freaking insane.

These almost look like AI image generations they’re so good.

Same goes for the new wave of Minecraft builds people are posting:

Again, what’s cool about this is that it’s showing off how well the new model understands how to write the underlying code to get these sorts of things to build.

If you're not a coder, you might wonder why any of this matters.

It’s because of one VERY important reason:

The future of AI isn't really about you writing code. It's about AI agents writing code in the background to do normal stuff. Without you ever knowing.

A perfect example from Google I/O this year. Their new personal AI agent, Google Spark, wrote a tiny bit of code on the spot to spin up a personal fitness tracker.Right now, most people use AI in a chat window. You type something, it responds, you type something else.

Love this newsletter? Forward it to one curious friend. They can join in one click.

In the past, this kind of thing was unreliable. WAY more trouble than it was worth. And often wasted time. It wouldn’t do the thing you wanted it to do.

But with the coding jump Mythos is showing?

I suspect a lot of this stuff is just going to work.

And that's where it becomes a huge deal.

The big thing we don't know yet is how expensive Mythos is going to be. At first, almost certainly very.

These models always launch pricey, and Mythos has been under restricted access for a reason. Anthropic only just expanded it to 150 enterprise orgs last week.

But costs do come down. Capacity gets better. And once Mythos is sitting inside Claude Code as a default option, every human with a Pro plan is going to be running this stuff for hours a day.

Which means we're about to find out how good "good" really is.

-Gavin

This week on AI For Humans: Microsoft’s New AI Agent First Mentality👇

I mentioned Codex's /goal tool last week, and that I'd put 60+ hours into a project with it.

Well, I kept going. It's now 80+ hours in, and you can actually play it.

To be clear, this isn’t great yet (it may never be) but it’s getting pretty good! It’s also way too packed with stuff and none of the flows work that well.

The current build is here: bear-jump-port.vercel.app

Give it a shot and let me know what you think.

It’s not perfect but, again, this is just with steering from me and no real planning or coding at all. But it shows you the power of what these tools can do mostly on their own.