AI predictions for 2026: The flood is coming

The digital world will be transformed. Your day-to-day life will be basically the same.

The author generated the preceding image with the following prompt: “I'd like an image of five children playing on a playground, with at least one dangling upside down with the effects of gravity accurately depicted. Be attentive to physically possible and sensible construction: poles should extend all the way to the ground, openings in the playground structure should have an access point, etc.” (Kelsey Piper/Nano Banana Pro 3)

A good forecaster doesn’t start with the future; she starts with the past. I wanted to answer the question “What should we expect from AI in 2026?” and so I began to reflect on what had happened this last year. If you’re interested in doing some forecasting of your own, stop reading for a moment and jot down a few notes (or comment below) about how you think AI changed in the last year.

Most people can’t name a single thing that changed from the beginning to the end of the year, even while the technology improved massively.

The best widely available general purpose image model on Jan. 1, 2025 was Midjourney V 6.1. The best today, I’d argue, is Google’s Gemini 3 Pro Image Preview (which is branded as Nano Banana Pro, and available to test out for free). Here are some prompts that I think showcase the differences (copied these straight from my conversations with the chatbots so please excuse the typos):

“photorealistic image of four children sitting on a couch gathered around a laptop, watching something. some of them are sitting on the back of the couch so as to get a good view. their postures are varied and absurd but they’re all engrossed in what they’re watching”

Here’s a response from the version of Midjourney available Jan. 1, 2025: