Just when you think you’ve wrapped your mind around computers that can put your dog in front of the Eiffel Tower or chatbots that act like your best friend (or lover), the AI behemoths surprise you with a fully AI-powered TikTok or the ability to virtually bring back your dead relatives.
I’ve worked in the AI space for 15 years. I served as an early beta tester for OpenAI in 2020, when I predicted that a little model called GPT-3 had world-changing potential.
It was later released as something called “ChatGPT”–perhaps you’ve heard of it?
I’ve also called several big AI trends correctly, including the rise of video generators and the “AI Wars” between Google and OpenAI.
Based on my experience, here are my six AI predictions for 2026 and beyond.
1. OpenAI goes garlic
In late 2025, Google’s Gemini model started to gain ground on OpenAI and its GPT-5.1 system.
That apparently really irked Sam Altman and the OpenAI team. Altman reportedly called a “code red,” directing staff to focus all their efforts on besting Google.
Rumor had it they were developing a new, fully re-trained thinking model, codenamed Garlic. When OpenAI did a surprise drop of a new GPT-5.2 model in mid December, lots of people thought it might be Garlic coming to market.
Based on my own testing, it’s not. Or at least, it’s not the complete model.
GPT-5.2 is indeed better than the previous model. It’s faster and more efficient, and makes fewer errors. It’s also notably better at scientific tasks, and practical ones like coding.
But now it appears there will be a second new OpenAI release, which I expect to come out in January. That’s most likely the full Garlic model.
This new model, I predict, will have a new knowledge cutoff sometime in 2025, a broader context window, and much better image generation capabilities. It will also be faster and more efficient to run, especially on “thinking” tasks.
2. Google’s Gemini continues its march toward domination
Whenever it finally arrives, Garlic will enter the world with plenty of competition.
Google was very slow to roll up to the generative AI table. For a company that’s been working in deep learning for decades and has some of the most intelligent people in the world working for it, that felt like a big miss.
Google had reportedly developed its own ChatGPT years before OpenAI, but chickened out on releasing it. In the beginning of the AI race, that allowed OpenAI to very loudly and publicly eat Google’s lunch.
The history of science, though, is littered with examples where early innovators weren’t the ones who successfully commercialized a new technology.
Just ask Joseph Swan, the true inventor of the lightbulb. You’ve never heard of him. But you do know Thomas Edison, who made the lightbulb a widely available technology—and did a great job promoting his invention (and himself) in the process.
Historically, first-mover advantage has proven surprisingly inconsequential in the tech space. And now that Google has woken up to the importance of AI, they’re aggressively building out their Gemini model and integrating it into almost all their products, including their core search experience.
Google has more data, more resources (including its own custom AI chips), more people, and a much broader reach than OpenAI.
In 2026, Google will continue to throw its weight around in generative AI. Gemini will go from being an also-ran to one of the most powerful models on the market. Because it will be broadly integrated into products that normal people use on a day-to-day basis, it will immediately have an audience in the billions.
The struggle now isn’t for newer companies like OpenAI to create the best product. It’s to create a product better than Google’s. That will be very hard in 2026 and beyond.
3. Chatbots become therapy (and a bit more)
Users have already realized that ChatGPT can take the place of a human therapist. In a recent poll by the Economist, 25% of people reported turning to chatbots for mental health support.
As cases of AI psychosis and alleged suicides demonstrate, this can go very badly. But for people who can’t afford any kind of psychological support—or simply don’t have access to it in their language or country—using chatbots as cheap therapists is incredibly appealing.
Without directly saying so, OpenAI has implied that they’re moving ChatGPT further into this space, with improvements to how the bot handles sensitive medical and mental health conversations.
This could be a huge boon for mental health. Many people appear more comfortable discussing their problems with an unthinking bot than with a human. The fact that a session with ChatGPT doesn’t cost $300 per hour is also a big plus!
ChatGPT’s capabilities will expand in other ways, too. A rumored “adult mode” will arrive in 2026, allowing ChatGPT to write risqué material. Prepare yourself for a wave of frenzied op-eds about how people are turning to these newly salacious for relationships instead of fellow humans.
4. AI-generated videos take over—and not just on Sora
OpenAI’s AI-powered Sora video generator is incredibly powerful, and their Sora-based social network is incredibly fun to use.
In 2026 and beyond, expect to see the reach and importance of AI-generated vertical video accelerate.
Vertical video is the perfect format for AI. The clips tend to be short, which caters to AI’s ability to generate about 10 seconds of reliable video before things go off the rails.
They also tend to be grabby and compelling. Again, AI excels at making videos of things like people falling into wedding cakes or having heated arguments with their roommates.
In 2026, expect the reach of AI-powered social networks like Sora to grow dramatically.
The biggest growth, though, will come from these videos migrating off the Sora platform and onto other social media. Already, my Facebook Reels feed is dominated by clearly AI-generated videos of things like a cat saving her kittens from a flood or grandmothers fighting grizzly bears.
In 2026, AI videos won’t stay put. They’ll travel into every vertical video space on the web—from TikTok to Nextdoor—further blurring the lines between what’s real and what’s imagined.
5. Electricity becomes the limiting factor
I have friends who build data centers for a living. They tell me the only thing stopping them from building more data centers is finding enough electrical power to keep up with AI’s demands.
Some companies are reportedly even going nuclear, building or recommissioning fully functional reactors to power their electricity-hungry AI chips.
The need for more electrical power for AI will start to limit the tech’s growth in 2026. It will also rub up against society’s other needs.
In 2026, I expect a populist backlash against the fact that data centers’ voracious energy demands are raising electricity rates for everyday people.
Ultimately, the deficiencies of today’s grid–strained as it is by the rise of AI–will drive innovative, new models that are good for everybody. Cheap solar power at midday may be redirected toward data centers, for example, or stored in giant batteries to keep servers running overnight.
This demand will create a huge market for green technologies, ultimately benefiting the planet and everyone on it.
6. AI invades the real world
No, the robot uprising isn’t here just yet. But AI is increasingly invading the real world.
Self-driving cars were once a novelty. In 2026, usage will explode, with Zoox, Waymo and their competitors–including entrants from China–serving more cities.
The rise of self-driving vehicles and other physical manifestations of AI technology will surprise people. You’ll blink, and one day it will feel like nearly every car on the road is self-driving—as it currently does in my home city of San Francisco.
I expect to see other experiments with physical AI in 2026, from robot baristas to caregiving machines, and plenty of military AI tech, too.
Again, though, the self-driving car blitz coming in 2026 will be the most profound and surprising (to everyday people, anyway) implementation of the technology. It will arrive far sooner than you think.
A pat on the back
So that’s what I anticipate for the year ahead. As someone who’s been in the AI space for a long time, I’ve missed some things.
But I also wrote, two years before ChatGPT, that “OpenAI and its founders could easily make billions (and likely challenge the advertising and content recommendation engines of rivals like Google) by throwing caution to the wind and throwing open the doors to GPT-3 to all comers.”
I still pat myself on the back for that one.
How will my 2026 predictions shape up? Ask me in a year!