Welcome to the AI For Humans newsletter!

2025 was quite the year for AI.

Around this time in 2024, we’d only just heard about OpenAI’s o3 model, Google and Gemini were still far behind and no one was really sure if they could take Grok seriously.

And now we have GPT-5.2, Gemini 3 Pro, Claude Opus 4.5, Grok 4, Sora 2, Nano Banana Pro, GPT-Image 1.5 and on and on and on.

Time moves very fast in this space and we figured this would be a good time to make some predictions for what we might see in AI for 2026.

First up…

Andrej Karpathy (former Tesla / OpenAI engineer & current AI educator) had another post go hugely viral over the weekend where he laid out how he’s feeling somewhat behind when it comes to the current state of AI advancements.

When someone so close to the edge is feeling behind, it gives us normies some sense of relief. After all, it’s extremely hard to keep up with the pace of change in AI.

In part, Karpathy’s tweet is about the current feeling of programmers around Anthropic’s new model Opus 4.5… there’s a rising sense that this model is just able to do stuff aka program entire apps and codebases on its own.

Yes, there were promises of this sort of thing before but now it’s finally happening.

But, as Karpathy’s tweet points out, along with the ability to get the machine to create all sorts of wonderful new things, new languages and toolsets are arising that will take a phalanx of new specialists to properly exploit.

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

So the prediction here is simple:

It’ll get ever easier to ‘vibecode’ your own apps. And personalized software aka something you make just for yourself to use will actually come to frutition.

But those doom-and-gloom stories of programming as a job going away? Not going to happen. In fact, I’m predicting we’ll see a boom in jobs for those that understand how to get these models to do exactly what they want.

Progammers (or whatever we’ll call them) will be in higher demand than ever before.

If you’re reading this newsletter, you’re pretty up-to-date on my feelings about AI video and how it is going to transform the world of entertainment. And, after the Disney / Sora deal, the entire world has woken up to this as a possibility.

But I saw a video over the weekend that made me realize we’re just now starting to see the sort of impact that putting these tools into the hands of creators will have more broadly.

Unlock a focused set of AI strategies built to streamline your work and maximize impact. This guide delivers the practical tactics and tools marketers need to start seeing results right away:

Stay ahead of the curve with these top strategies AI helped develop for marketers, built for real-world results.

Download the Free Report

And while this video is clearly a novelty and works off of a lot of existing IPs, you can get a sense of just how big this shift could look.

Ignore what many might think of IP theft here and just examine the quality of what you’re seeing. Not only do all the characters look nearly perfect, but the quality of the editing and the voice-over work here is top notch. AI video generators in the hands of truly capable technicians is a remarkable thing.

Now for my prediction:

I think 2026 will be the year that these sort of workflows will begin to filter deeply into the hands of some of the most capable storytellers alive. We’ll start to see not just incredible AI commericals but longer-form AI videos (made by humans!) that start to really move the needle on storytelling.

More specifically, I think 2026 will be the first year we see a break-out AI video created series that becomes a decent sized mainstream hit. It won’t be on a streaming service at first (prob too many issues with guilds and the like) but much like how self-publishing books has become a pathway towards mainstream publishing deals, an AI series will get picked-up for the masses.

And finally…

We try to focus at least a small part of AI For Humans each week on the advances in humanoid robotics, even though technically most people listening to us are more interested in what the latest model can do etc.

While the field of robotics at large has been around for some time (and we, as humans, have imagined what a future full of robots might look like forever), the AI boom of the last few years has not only advanced what’s possible with humanoids but helped infuse a level of capital investment never seen before.

While 2025 has been a mixed bag (terrifying Chinese terminators robots vs the semi-janky 1x home robot), the overall robotics market is set to explode even further in the coming decades.

Prediction: 2026 will be the first year that you and I see a humanoid robot in the wild.

It might be someone’s small toy or entertainment (like the now infamous NSFW Rizzbot) or perhaps for funsies in your office as part of a work event.

It will feel really weird at first and a total novelty. And it won’t work that well or do anything of consequence. But, much like that first time you saw a Waymo in action, it will feel like you’re suddenly living in the future.

And then you’ll start seeing them everywhere. And then? It will just be normal.

Yes, it’s probably five+ years until we really see robots everywhere but 2026 will be the year it all starts.

For now, we’ll have to settle for this:

That’s all for this week, have a wonderful New Year and we’ll see you soon.

-Gavin

Nano Banana Pro is one of those AI tools where it’s so easy to use, that you might think you don’t need to learn proper prompting techniques.

While it’s tempting to go about getting merely great results, I implore you to try a few specific JSON prompts and you’ll begin to understand just how powerful the tool can be with a little extra work under-the-hood.

Prolific AI creator (and now Google employee) fofr recently shared his JSON prompt for an image of a specific video game selection screen (prompt is here on GitHub).

I took that prompt, pasted it into Gemini 3 Pro and asked it to turn the characters into famous chefs and it just worked.

JSON Prompting (using JSON code rather than just words) allows you to get insanely granular when it comes to specific parts of the image and gives the model a much clearer blueprint to follow. It also allows you to specify words and phrases for particular details.

It’s actually quite easy, espeically when paired with one of the frontier LLMs.

Go try it and be amazed!